Skip to content


Steps to implement custom meshpass

  1. Add new render pass method FDeferredShadingSceneRenderer::RenderPrePass

  2. Debug Stats & flags:

    if (!ShouldRenderTranslucency(TranslucencyPass))
      return; // Early exit if nothing needs to be done.
    SCOPED_DRAW_EVENT(RHICmdList, Translucency);
    SCOPED_GPU_STAT(RHICmdList, Translucency);
  3. Call & Implement CreateCustomPassUniformBuffer(RHICmdList, View, PassUniformBuffer);

  4. Create Pass Render State with UniformBufferParms: FMeshPassProcessorRenderState DrawRenderState(View, BasePassUniformBuffer);

  5. Call & Implement SetupCustomPassState(DrawRenderState);

  6. Scene->UniformBuffers.UpdateViewUniformBuffer(View);

  7. FDeferredShadingSceneRenderer::RenderBasePassView

    1. SetupCustomPassView
      1. SetViewport
    2. DispatchDraw
  8. Implement FMeshPassProcessor. Ex: class FDepthPassMeshProcessor : public FMeshPassProcessor

  9. Register the pass: FRegisterPassProcessorCreateFunction RegisterCustomPass()
  10. Implement AddMeshBatch

    • Pass Filter
    • Select Shader to use. Ex: GetDepthPassShaders
    • Implement CalculateCustomPassMeshStaticSortKey() to calculate SortKey for mesh state
    • Gather bindings: BuildMeshDrawCommands()
  11. FRelevancePacket::MarkRelevant() && FRelevancePacket::ComputeDynamicMeshRelevance()

  12. Drawcall Merging: class FMeshDrawCommand::MatchesForDynamicInstancing

Notes: Update Shaders from Primitive.PrimitiveId => GetPrimitiveData(Parameters.PrimitiveId).

  1. Need to use GetPrimitiveData(Parameters.PrimitiveId)

  2. FDeferredShadingSceneRenderer::ClearGBufferAtMaxZ(FRHICommandList& RHICmdList)


Motivation: Need to execute lot of draws

  • Modular construction

  • Dynamic lighting/shadowing need extra mesh passes

  • Some techniques (DXR/GPU culling) needs to eval entire scene


  • Draw call merging based on RHI
  • No more per draw per frame shader params

Mesh Drawing/Render Refactor:

  • Very quick description would be: no more draw policies and static draw lists. Everything is now done through mesh processors, mesh draw commands (draw state) and EMeshPass lists.

  • Mesh draw commands are build and cached when you attach your prim to scene.

Mesh processors convert mesh batches to mesh draw commands.

EMeshPass is basically a list of visible mesh draw commands for a given pass, which is sorted, merged and drawn.

Now, as for render graph it’s currently not used too much in 4.22 :)

​ MeshBatch is now converted into mesh draw command, which contains all the draw state (binding etc).

​ FMeshBatch is like an interface for rendering.


  • FPrimitiveSceneProxy: Render Thread View of UE4 Gamecode Component
  • FMeshBatch: decouples FPrimitiveSceneProxy implementations from mesh passes
  • Contains everything pass needs to know
  • Agnostic of passes
  • DrawingPolicy: Converts FMeshBatch to RHIDrawCommands
  • RHICommandList: Platform independent render commands

Every frame, generate FMeshBatch from Scene proxies

New Way#

  • FMeshDrawCommand: Stateless description of draw command
  • FMeshPassProcessor: Builds one ore more FMeshDrawCommands from FMeshBatch
  • Select shader
  • Collect pass/vertex factory/material bindings
  • Describes a mesh pass

Shader Binding#


  • Shader Parms are submitted in big GPU buffers
  • Anything referenced by FMeshDrawCommands must not change frequently bc it will be invalidated
  • Instead, keep bindings stable and reference one big Uniform Buffer
  • Code path to update uniform buffer: RHIUpdateUniformBuffer
  • Uniform Buffers: FPersistentUniformBuffers
  • PrimitiveUniformBuffer: Data unique to primitive like LocalToWorldTransform
  • MaterialUniformBuffer: Material params
  • MaterialParameterCollections: MPC
  • PrecomputedLightingUniformBuffer: Which lightmaps to reference, etc
  • PassUniformBuffers: Any data that pass needs that’s per primitive that’s not shared across passes
  • Remember, shader bindings are cached so those need to be invalidated (eg: changing skylight)
  • VertexFactory: How UE4 generates vertex data from CPU to GPU
  • Only VFs that don’t require View are cached (eg: FLocalVertexFactory)

High level Frame with Caching#

  • FPrimitiveSceneInfo::AddToScene
  • If Static, cache FMeshBatch’s on FPrimitiveSceneInfo
    • Also generate FMeshDrawCommands and store on the scene
  • FScene::SetSkyLight
  • Invalidate cached FMeshDrawCommands
    • FPrimitiveSceneInfo::BeginDeferredUpdateStaticMeshes: Invalidate specific mesh
    • FScene->bScenesPrimitivesNeedStaticMeshElementUpdate: Invalidate entire Scene’s cached commands
    • Ex: FScene::SetSkyLight()
  • InitViews
  • Foreach Primitive
    • If Static Relevance
    • Compute LOD and add cached FMeshDrawCommands to visible list
    • If Dynamic Relevance
    • Gather FMeshBatch’s with GetDynamicMeshElements
  • RenderDepth
  • Foreach FMeshDrawCommand in visible list, Draw

Drawcall Merging#

  • FMeshDrawCommand captures all the state for the draw
  • Currently only D3D11 style merging implemented, so only drawcalls with identitcal shader bindings merge
  • Future will be D3D12 Execute Indirect that can change state in-between

  • Happens in FMeshDrawCommand::MatchesForDynamicInstancing

  • Shader Parameters must be crafted to enable this

Pass Types Description
Pass Parameters These are placed in the pass uniform buffer, where any draws in the pass can merge.
FLocalVertexFactory Parameters These are placed in a uniform buffer owned by UStaticMesh where any draws with the same UStaticMesh can merge.
Material Instance Parameters These are palced in a material uniform buffer where any draws using the same Material Instance can merge.
Lightmap Resource Parameters These are placed in a LightmapResourceCluster uniform buffer where any draws using the same LightmapTexture can merge.
Primitive Parameters These are placed in a scene-wide primitive data buffer called GPUScene and indexed in the shader using PrimitiveID.
  • Assigning state buckets for merging is slow so its cached at FPrimitive::AddToScene
  • cpp>TSet<FMeshDrawCommandStateBucket, MeshDrawCommandKeyFuncs> CachedMeshDrawCommandStateBuckets;

  • Actual merging operation is just transformation on visible FMeshDrawCommand list

  • For each drawcommand in the same statebucket, it gets replaced with an instanced mesh draw command

  • To be effective, need to get rid of per-draw bindings

  • Created Pass uniform buffer frequency

  • Moved 64 per-draw bindings into OpaqueBasePassUniformBuffer

  SHADER_PARAMETER_STRUCT(FForwardLightData, Forward)
  SHADER_PARAMETER_STRUCT(FReflectionUniformParameters, Reflection)
  SHADER_PARAMETER_STRUCT(FPlanarReflectionUniformParameters, PlanarReflection) // Single global planar reflection for the forward pass.
  SHADER_PARAMETER_STRUCT(FFogUniformParameters, Fog)
  SHADER_PARAMETER_TEXTURE(Texture2D, SSProfilesTexture)

  SHADER_PARAMETER_STRUCT(FSharedBasePassUniformParameters, Shared)
  SHADER_PARAMETER_STRUCT(FSceneTexturesUniformParameters, SceneTextures)
  // Material SSR
  SHADER_PARAMETER(FVector4, HZBUvFactorAndInvFactor)
  SHADER_PARAMETER(FVector4, PrevScreenPositionScaleBias)
  SHADER_PARAMETER(float, PrevSceneColorPreExposureInv)
  SHADER_PARAMETER_SAMPLER(SamplerState, PrevSceneColorSampler)
  // Translucency Lighting Volume
  SHADER_PARAMETER_TEXTURE(Texture3D, TranslucencyLightingVolumeAmbientInner)
  SHADER_PARAMETER_SAMPLER(SamplerState, TranslucencyLightingVolumeAmbientInnerSampler)
  SHADER_PARAMETER_TEXTURE(Texture3D, TranslucencyLightingVolumeAmbientOuter)
  SHADER_PARAMETER_SAMPLER(SamplerState, TranslucencyLightingVolumeAmbientOuterSampler)
  SHADER_PARAMETER_TEXTURE(Texture3D, TranslucencyLightingVolumeDirectionalInner)
  SHADER_PARAMETER_SAMPLER(SamplerState, TranslucencyLightingVolumeDirectionalInnerSampler)
  SHADER_PARAMETER_TEXTURE(Texture3D, TranslucencyLightingVolumeDirectionalOuter)
  SHADER_PARAMETER_SAMPLER(SamplerState, TranslucencyLightingVolumeDirectionalOuterSampler)
  • Remaining per draw bindings
  • We can upload these to a scene-wide unified buffer
  • Dynamically index them from the shader via InstanceId
  • Ex:

    • GlobalConstantBuffer values
    • PreviousLocalToWorld BasePass Velocity
    • Dither LOD
    • PrimitiveUniformBuffer
    • PrecomputedLightingUniformBuffer
    • DistanceCullFadeUniformBuffer
  • GPU Scene Primitive Data buffer: GPU TArray implementation with dynamic resizing

  • Renderthread tracks Primitive add / update / remove, uses a compute shader to update them on the next frame (only operate on deltas)
  • Need to use GetPrimitiveData(Parameters.PrimitiveId) in shaders now


  • Static Mesh Components that are in different lightmaps wont get merged
  • Tweak DefaultEngine.ini MaxLightmapRadius
  • Deferred primtive update is gone (r.DeferUniformBufferUpdatesUntilVisible)
  • before, if an out of view primitive moved, its primitive buffer wouldn’t be updated
  • Now it has to bc of caching
  • Forward renderer now supports only single global planar reflection
  • FMeshDrawCommand Performance Hazards
  • Going over inline allocator limit will cause performance hazard
  • FMeshDrawShaderBindings assumes 2 shader frequencies (Vertex + Pixel): cpp>TArray<FMeshDrawShaderBindingsLayout, TInlineAllocator<2>>ShaderLayouts
  • FMeshDrawCommand assumes 10 shader bindings among all frequencies: cpp>const int32 NumInlineShaderBindings = 10;
  • FMeshDrawCommand assumes 4 vertex streams from the vertex factory: cpp>typedef TArray<FVertexInputStream, TInlineAllocator<4>>FVertexInputStreamArray;

Upgrade Notes#

Static Draw Lists and Primitive Sets#

  • static draw lists (TStaticMeshDrawList) and primitive sets (for example, FTranslucentPrimSet, FCustomDepthPrimSet) were replaced by FParallelMeshDrawCommandPass. FParallelMeshDrawCommandPass encapsulates a single per pass visible mesh draw command list
  • per scene mesh lists were replaced by visible mesh lists
  • previously drawing a Static Mesh pass would traverse the entire list of per-pass Static Meshes in-scene (TStaticMeshDrawList) to select visible ones by checking FViewInfo::StaticMeshVisibilityMap for each Static Mesh
  • new design, drawing is just a simple traversal of a visible mesh draw command array (FMeshDrawCommandPassSetupTaskContext::MeshDrawCommands)
  • pipeline also has an immediate mode mesh rendering emulation through DrawDynamicMeshPass function.
  • should only be used for non-performance critical mesh passes, as it doesn’t support caching, automatic instancing, and makes multiple dynamic memory allocations. For example, it replaces DrawViewElements, which was responsible for rendering Editor-only helper meshes

Drawing Policies#

  • Drawing policies like FDepthDrawingPolicy or FBasePassDrawingPolicy were replaced by FDepthPassMeshProcessor and FBasePassMeshProcessor
  • Specific pass mesh processors are derived from a FMeshProcessor base class and are responsible for converting each FMeshBatch into a set of mesh draw commands for the pass.

Shader Bindings#

  • Previously all shader parameters were directly set on RHICmdList by drawing policies

  • Now, All parameters are gathered into FMeshDrawSingleShaderBindings (which later are set on RHICmdList by calling SetOnCommandList during drawing)

  • Old: FDrawingPolicyRenderState to pass common high-level mesh pass render state (e.g. pass uniform buffer)

  • New: Renames FDrawingPolicyRenderState to FMeshPassProcessorRenderState without major changes to its functionality

  • Old: Other parts of shader bindings were filled inside of shader’s SetParameters and SetMesh functions

  • New: Replaced by GetShaderBindings and GetElementShaderBindings, and pass per-draw parameters inside a customizable ShaderElementDataType

  • Since each FMeshPassProcessor must go through BuildMeshDrawCommands() to call the pass shader’s GetShaderBindings(), we need a mechanism to pass arbitrary data from the FMeshPassProcessor to the GetShaderBindings() call. This is accomplished with the ShaderElementData parameter to BuildMeshDrawCommands()

  • Many loose parameters were pulled out into per-pass or other uniform buffers (do not use loose parameters)

  • Old: uniform buffers like ViewUniformBuffer or DepthPassUniformBuffer were recreated every frame with new data

  • New: those are persistent and global (kept inside FScene::FPersistentUniformBuffers)

  • instead of recreating them to pass new data — their contents are updated using new RHI function — to RHIUpdateUniformBuffer
  • This indirection enables shaders to receive per-frame data even though their mesh draw commands are cached


  • Extended with two extra relevance flags:
  • bVelocityRelevance is required for separate velocity pass
  • bTranslucentSelfShadow is required for translucency self shadow
  • New: all dynamic draws now rely on view relevance, and disabling certain passes in view relevance will also disable its rendering


  • The new pipeline introduced GPUScene, which is a structured buffer containing primitive uniform buffer data for all the primitives in the scene.
  • Currently only the local vertex factory (Static Mesh Component) and the SM5 feature level can utilize this rendering path.
  • Shaders need to use GetPrimitiveData(PrimitiveId) instead of accessing the Primitive uniform buffer directly to compile with GPUScene enabled.
  • Primitive.Member => GetPrimitiveData(Parameters.PrimitiveId).Member



Add Immediate Mode Custom Mesh Pass#

void DrawDecalMeshCommands(FRenderingCompositePassContext& Context, EDecalRenderStage CurrentDecalStage, FDecalRenderingCommon::ERenderTargetMode RenderTargetMode)
  FRHICommandListImmediate& RHICmdList = Context.RHICmdList;
  const FViewInfo& View = Context.View;

  const bool bPerPixelDBufferMask = IsUsingPerPixelDBufferMask(View.GetShaderPlatform());

  FGraphicsPipelineStateInitializer GraphicsPSOInit;
  FDecalRenderTargetManager RenderTargetManager(Context.RHICmdList, Context.GetShaderPlatform(), CurrentDecalStage);
  RenderTargetManager.SetRenderTargetMode(RenderTargetMode, true, bPerPixelDBufferMask);

  DrawDynamicMeshPass(View, RHICmdList,
    [&View, CurrentDecalStage, RenderTargetMode](FDynamicPassMeshDrawListContext* DynamicMeshPassContext)
    FMeshDecalMeshProcessor PassMeshProcessor(

    for (int32 MeshBatchIndex = 0; MeshBatchIndex < View.MeshDecalBatches.Num(); ++MeshBatchIndex)
      const FMeshBatch* Mesh = View.MeshDecalBatches[MeshBatchIndex].Mesh;
      const FPrimitiveSceneProxy* PrimitiveSceneProxy = View.MeshDecalBatches[MeshBatchIndex].Proxy;
      const uint64 DefaultBatchElementMask = ~0ull;

      PassMeshProcessor.AddMeshBatch(*Mesh, DefaultBatchElementMask, PrimitiveSceneProxy);

void RenderMeshDecals(FRenderingCompositePassContext& Context, EDecalRenderStage CurrentDecalStage)
  FRHICommandListImmediate& RHICmdList = Context.RHICmdList;
  FSceneRenderTargets& SceneContext = FSceneRenderTargets::Get(RHICmdList);
  const FViewInfo& View = Context.View;
  FScene* Scene = (FScene*)View.Family->Scene;

  SCOPED_DRAW_EVENT(RHICmdList, MeshDecals);

  FSceneTexturesUniformParameters SceneTextureParameters;
  SetupSceneTextureUniformParameters(SceneContext, View.FeatureLevel, ESceneTextureSetupMode::All, SceneTextureParameters);

  if (View.MeshDecalBatches.Num() > 0)
    switch (CurrentDecalStage)
    case DRS_BeforeBasePass:
      DrawDecalMeshCommands(Context, CurrentDecalStage, FDecalRenderingCommon::RTM_DBuffer);

    case DRS_AfterBasePass:
      DrawDecalMeshCommands(Context, CurrentDecalStage, FDecalRenderingCommon::RTM_SceneColorAndGBufferDepthWriteWithNormal);

    case DRS_BeforeLighting:
      DrawDecalMeshCommands(Context, CurrentDecalStage, FDecalRenderingCommon::RTM_GBufferNormal);
      DrawDecalMeshCommands(Context, CurrentDecalStage, FDecalRenderingCommon::RTM_SceneColorAndGBufferWithNormal);

    case DRS_Mobile:
      DrawDecalMeshCommands(Context, CurrentDecalStage, FDecalRenderingCommon::RTM_SceneColor);

    case DRS_AmbientOcclusion:
      DrawDecalMeshCommands(Context, CurrentDecalStage, FDecalRenderingCommon::RTM_AmbientOcclusion);

    case DRS_Emissive:
      DrawDecalMeshCommands(Context, CurrentDecalStage, FDecalRenderingCommon::RTM_SceneColor);

Render Graph#

Main file: UnrealEngine\Engine\Source\Runtime\RenderCore\Public\RenderGraph.h

Main class: FRDGBuilder Currently used very lightly in 4.22 but looks like Compute Shaders can be dispatched with it FComputeShaderUtils::AddPass(…)



  • FMeshDrawCommand::DebugData is a debug data struct
  • Whether to assert in cases where the layout is released before uniform buffers created with that layout

  • Whether to assert when a uniform buffer is being deleted while still referenced by a mesh draw command

  • Enabling this requires -norhithread to work correctly since FRHIResource lifetime is managed by both the RT and RHIThread
Command Desc
r.MeshDrawCommands.DynamicInstancing Whether to dynamically combine multiple compatible visible Mesh Draw Commands into one instanced draw on vertex factories that support it.
r.MeshDrawCommands.LogDynamicInstancingStats Whether to log dynamic instancing stats on the next frame
r.MeshDrawCommands.LogMeshDrawCommandMemoryStats Whether to log mesh draw command memory stats on the next frame
r.GPUScene.UploadEveryFrame Forces GPU Scene to be fully updated every frame, which is useful for diagnosing issues with stale GPU Scene data.
r.GPUScene.ValidatePrimitiveBuffer This downloads GPU Scene to the CPU and validates its contents against primitive uniform buffers.
r.RHICmdUseThread To Use a separate thread for RHICmdList
r.RHIThread.Enable To Disable RHI Thread
r.RHICmdBypass Set to 1 to disable
r.RHICmdUseParallelAlgorithms True to use parallel algorithms. Ignored if r.RHICmdBypass is 1.
r.MeshDrawCommands.ParallelPassSetup Whether to setup mesh draw command pass in parallel.
r.RHICmdBasePassDeferredContexts Disable the parallel tasks for base pass draw dispatch, causing those to happen on the RenderingThread.
r.MeshDrawCommands.UseCachedCommands Whether to render from cached mesh draw commands (on vertex factories that support it), or to generate draw commands every frame.
r.RDG.ImmediateMode Toggle get render graph executing passes as they get created to easily debug crashes caused by pass wiring logic.
r.RDG.EmitWarnings Toggle render graph emitting warnings about inefficiencies.
r.GPUScene.ValidatePrimitiveBuffer Readback the GPU primitive data and assert if it doesn’t match the RT primitive data.
r.GPUScene.UploadEveryFrame Upload the entire scene’s primitive data every frame

Last update: November 14, 2019