【DXR/UE4】DXR教程/UE4RTAO

DXR教程一

这个是Nvidia做的DXR的教程之一，总共14篇，主要是介绍DXR Api的使用，这套API还是非常繁琐的，如果一直关注细节的话很容易理不清思路。
事实上只需要关注TraceRay这个函数就可以很好的了解DXR的整体。这个流程中新的两个部分：加速结构（BLAS和TLAS）和ShaderTable都可以说是为TraceRay服务的，它们用于说明当ray与AccelerationStructure中的物体相交后执行哪部分程序。

Template<payload_t>
void TraceRay(RaytracingAccelerationStructure AccelerationStructure,uint RayFlags,uint InstanceInclusionMask,uint RayContributionToHitGroupIndex,uint MultiplierForGeometryContributionToHitGroupIndex,uint MissShaderIndex,RayDesc Ray,inout payload_t Payload);

第一个参数是使用的Top-Level Acceleration Structure。
第二个参数是Ray Flags，用于定义透明、裁剪一些提前操作。
第三个参数则用于拒绝某些Instance，

if(!((InstanceInclusionMask & InstanceMask) & 0xff)) { ignore intersection }

pInstanceDescs[i].InstanceMask = 0xFF;

当Instance的Mask设为0时，该instance会被忽略。
第四个参数是Ray Index，它的主要作用是索引次级光线。
第五个参数是两个Geometry的hit shader的距离。

教程13中的样例场景有3个实例，其中第一个实例有两个Geometry，一个是三角形，另一个是Plane，另外两个实例为三角形。同时击中后会产生一个再次发射一条阴影光线。

当RayGen Shader发射一条光线击中物体后，要确定接下来执行那个Hit Shader ，计算公式：

entryIndex =  
InstanceContributionToHitGroupIndex + 
GeometryIndex * MultiplierForGeometryContributionToShaderIndex + 
RayContributionToHitGroupIndex

假定击中的是Plane，那么就要执行hitshader，我们要去找hitShaderGroup的位置，HitGroupTable如下，它的起始的Offset是3个ShaderTableEntry的大小，包含8个HitShader。

	size_t hitOffset = 3 * mShaderTableEntrySize;raytraceDesc.HitGroupTable.StartAddress = mpShaderTable->GetGPUVirtualAddress() + hitOffset;raytraceDesc.HitGroupTable.StrideInBytes = mShaderTableEntrySize;raytraceDesc.HitGroupTable.SizeInBytes = mShaderTableEntrySize * 8;    // 8 hit-entries

找到HitShaderGroup的位置后，我们查找HitShader的索引。在上面EntryIndex的计算公式中，有两个是Shader中的参数：MultiplierForGeometryContributionToShaderIndex，
RayContributionToHitGroupIndex这两个在shader中指定，对于前者，是Instance中Geometry的数量，对于后者是Ray的index，因为是第一条光线所以Index=0。

InstanceContributionToHitGroupIndex在Top-Level AC中指定，用于说明当前的Instance在HitGroup中的位置，

	pInstanceDescs[0].InstanceID = 0;pInstanceDescs[0].InstanceContributionToHitGroupIndex = 0;``````for (int i = 1; i < 3; i++){pInstanceDescs[i].InstanceID = i;pInstanceDescs[i].InstanceContributionToHitGroupIndex = (i * 2) + 2;``````}

3个Instace的位置分别为0，4，6.

GeometryIndex用于说明这个Geometry在Instace中的位置，这个在Bottom-Level AC中指定，对于Instace0，Plane在其中的位置为1。

经过计算得到HitShader的Index为0+1*2+0=2，即执行hitGroup中的第三个Shader。

它的代码如下：

[shader("closesthit")]
void planeChs(inout RayPayload payload, in BuiltInTriangleIntersectionAttributes attribs)
{   ``````ShadowPayload Shadowpayload;TraceRay(gRtScene, 0 /*rayFlags*/, 0xFF, 1 /* ray index*/, 0, 1, ray, Shadowpayload);float factor = Shadowpayload.hit ? 0.1 : 1.0;payload.color = float4(0.9f, 0.9f, 0.9f, 1.0f) * factor;
}

也就是再发射次级光线进行阴影判断，所以Ray Index从0变成1，RayContributionToHitGroupIndex的值为1。
倒数第三个参数说明，如果未击中执行则执行MissGroup中的第2个Shader，index=1。

假设击中了Instance3，那么结果为：
entryIndex =
InstanceContributionToHitGroupIndex（6） +
GeometryIndex（0） * MultiplierForGeometryContributionToShaderIndex（0） +
RayContributionToHitGroupIndex（1）

即执行第8个shader，Index=7。

DXR剩下的部分（PipelineState，Resources等）类比光栅流程就可以了。

DXR教程二

第二份教程是在Falcor框架的基础上写的，同样是14节，与第一份教程相比，Falcor框架简化了DXR的操作，第一份教程的上千行代码在Falor框架下十几行就可以实现。

	pipeline->setPass(0, LightProbeGBufferPass::create());pipeline->setPass(1, GGXGlobalIlluminationPass::create("HDRColorOutput"));  pipeline->setPass(2, SimpleAccumulationPass::create("HDRColorOutput"));     pipeline->setPass(3, SimpleToneMappingPass::create("HDRColorOutput", ResourceManager::kOutputChannel)); 

总共四个Pass，第一个Pass是光栅化流程，输出一个GBuffer，包阔PosWorl，Normal，diffuse等信息。第2个Pass是RayTracing的核心，第三个Pass进行temporal accumulation，第四个Pass将结果输出。

第三个Pass的思路比较朴素，即如果前一帧和当前帧场景未发生改变，可以将前一帧（prevColor）的结果累加到当前帧（curColor），静止的帧数越多（gAccumCount），场景的噪点就会越小。

(gAccumCount * prevColor + curColor) / (gAccumCount + 1);

这个方法结合上相机位置的Jitter，不就在降噪的同时顺便进行了类似TAA的操作，Jitter是在第一个Pass中进行的。

第二个Pass的核心代码如下

	if (isGeometryValid){// Add any emissive color from primary raysshadeColor = gEmitMult * pixelEmissive.rgb;// (Optionally) do explicit direct lighting to a random light in the sceneif (gDoDirectGI)shadeColor += ggxDirect(randSeed, worldPos.xyz, worldNorm.xyz, V,difMatlColor.rgb, specMatlColor.rgb, roughness);// (Optionally) do indirect lighting for global illuminationif (gDoIndirectGI && (gMaxDepth > 0))shadeColor += ggxIndirect(randSeed, worldPos.xyz, worldNorm.xyz, noMapN,V, difMatlColor.rgb, specMatlColor.rgb, roughness, 0);}

如果当前击中点有效的话，进行直接光和间接光的计算。

在计算直接光的时候，为了减小计算量，教程在多个光源中随机选择一个光源计算，

	int lightToSample = min(int(nextRand(rndSeed) * gLightsCount), gLightsCount - 1);``````// Shoot our shadow ray to our randomly selected lightfloat shadowMult = float(gLightsCount) * shadowRayVisibility(hit, toLight, gMinT, distToLight);// Return the Lambertian shading color using the physically based Lambertian term (albedo / pi)return shadowMult * LdotN * lightIntensity * difColor / M_PI;

在计算间接光时，根据概率选择采样Diffuse还是Spec。

UE4 RTAO

选择了各种RT效果中最简单的RTAO来实现，主要是为了熟悉下UE Ray Tracing的流程，之前在DXR的API文档上已经浪费了够多时间了。

SSAO的问题就是只有屏幕信息，无法判断柱子后有没有物体。

RTAO就没有这个问题

#include "../Common.ush"
#include "../DeferredShadingCommon.ush"
#include "../Montecarlo.ush"
#include "../PathTracing/Utilities/PathTracingRandomSequence.ush"
#include "../SceneTextureParameters.ush"
#include"RayTracingCommon.ush"
#include "RayTracingDeferredShadingCommon.ush"
RaytracingAccelerationStructure TLAS;
RWTexture2D<float> RWAmbientOcclusionMaskUAV;
RWTexture2D<float> RWAmbientOcclusionHitDistanceUAV;RAY_TRACING_ENTRY_RAYGEN(MyRTAORGS)
{//recompileShaders changedconst uint gNumRays = 2;float IntensityLocal = 0.2;float Visibility = 0.0;uint2 PixelCoord = DispatchRaysIndex().xy + View.ViewRectMin;RandomSequence RandSequence;uint LinearIndex = CalcLinearIndex(PixelCoord);RandomSequence_Initialize(RandSequence, LinearIndex, View.StateFrameIndex);FGBufferData GBufferData = GetGBufferDataFromSceneTexturesLoad(PixelCoord);float DeviceZ = SceneDepthTexture.Load(int3(PixelCoord, 0)).r;float3 WorldPosition;float3 CameraDirection;ReconstructWorldPositionAndCameraDirectionFromDeviceZ(PixelCoord, DeviceZ, WorldPosition, CameraDirection);float3 WorldNormal = GBufferData.WorldNormal;float ClosestRayHitDistance = 10000.0;for (uint index = 0; index < gNumRays;index++){uint DummyVariable;float2 RandSample = RandomSequence_GenerateSample2D(RandSequence, DummyVariable);float3 Direction_Tangent;Direction_Tangent.xy = UniformSampleDiskConcentric(RandSample);Direction_Tangent.z = sqrt(1.0 - dot(Direction_Tangent.xy, Direction_Tangent.xy));float3 Direction_World = TangentToWorld(Direction_Tangent.xyz, WorldNormal);RayDesc Ray;Ray.Origin = WorldPosition;Ray.Direction = Direction_World;Ray.TMin = 0.01;Ray.TMax = 196.0;uint RayFlags = 0;const uint InstanceInclusionMask = RAY_TRACING_MASK_SHADOW | RAY_TRACING_MASK_THIN_SHADOW;float RayWeight = 1.0;FMinimalPayload MinimalPayload = TraceVisibilityRay(TLAS,RayFlags,InstanceInclusionMask,PixelCoord,Ray);Visibility += RayWeight * (MinimalPayload.IsMiss() ? 1.0 : (1.0 - IntensityLocal) * (MinimalPayload.HitT / Ray.TMax));if (MinimalPayload.IsHit()){ClosestRayHitDistance = min(ClosestRayHitDistance, MinimalPayload.HitT);}}RWAmbientOcclusionMaskUAV[PixelCoord] = Visibility / float(gNumRays);RWAmbientOcclusionHitDistanceUAV[PixelCoord] = ClosestRayHitDistance;}

#include"MyRTAO.h"
#include "DeferredShadingRenderer.h"#if RHI_RAYTRACING#include "ClearQuad.h"
#include "SceneRendering.h"
#include "SceneRenderTargets.h"
#include "SceneUtils.h"
#include "RenderTargetPool.h"
#include "RHIResources.h"
#include "UniformBuffer.h"
#include "RHI/Public/PipelineStateCache.h"
#include "Raytracing/RaytracingOptions.h"
#include "Raytracing/RayTracingMaterialHitShaders.h"//Note
#include "SceneTextureParameters.h"#include "PostProcess/PostProcessing.h"
#include "PostProcess/SceneFilterRendering.h"class FMyRTAORGS : public FGlobalShader
{DECLARE_GLOBAL_SHADER(FMyRTAORGS)SHADER_USE_ROOT_PARAMETER_STRUCT(FMyRTAORGS, FGlobalShader)static bool ShouldCompilePermutation(const FGlobalShaderPermutationParameters& Parameters){return ShouldCompileRayTracingShadersForProject(Parameters.Platform);}BEGIN_SHADER_PARAMETER_STRUCT(FParameters, )SHADER_PARAMETER_SRV(RaytracingAccelerationStructure, TLAS)SHADER_PARAMETER_RDG_TEXTURE_UAV(RWTexture2D<float>, RWAmbientOcclusionMaskUAV)SHADER_PARAMETER_RDG_TEXTURE_UAV(RWTexture2D<float>, RWAmbientOcclusionHitDistanceUAV)SHADER_PARAMETER_STRUCT_REF(FViewUniformShaderParameters, ViewUniformBuffer)SHADER_PARAMETER_STRUCT_INCLUDE(FSceneTextureParameters, SceneTextures)END_SHADER_PARAMETER_STRUCT()
};
IMPLEMENT_GLOBAL_SHADER(FMyRTAORGS, "/Engine/Private/RayTracing/MyRTAORGS.usf", "MyRTAORGS", SF_RayGen);
void FDeferredShadingSceneRenderer::MyRenderRTAO(FRDGBuilder& GraphBuilder,FViewInfo& View,const FSceneTextureParameters& SceneTextures,FRDGTextureRef* OutAmbientOcclusionTexture)
#if RHI_RAYTRACING
{RDG_EVENT_SCOPE(GraphBuilder, "My_RTAO");IScreenSpaceDenoiser::FAmbientOcclusionInputs DenoiserInputs;{FRDGTextureDesc Desc = FRDGTextureDesc::Create2D(SceneTextures.SceneDepthTexture->Desc.Extent,PF_R16F,FClearValueBinding::None,TexCreate_ShaderResource | TexCreate_RenderTargetable | TexCreate_UAV);DenoiserInputs.Mask = GraphBuilder.CreateTexture(Desc, TEXT("M_RTAO_TEX"));DenoiserInputs.RayHitDistance = GraphBuilder.CreateTexture(Desc, TEXT("M_RTAO_Hit_TEX"));}FMyRTAORGS::FParameters *PassParameters = GraphBuilder.AllocParameters<FMyRTAORGS::FParameters>();PassParameters->TLAS = View.RayTracingScene.RayTracingSceneRHI->GetShaderResourceView();PassParameters->RWAmbientOcclusionMaskUAV = GraphBuilder.CreateUAV(DenoiserInputs.Mask);PassParameters->RWAmbientOcclusionHitDistanceUAV = GraphBuilder.CreateUAV(DenoiserInputs.RayHitDistance);PassParameters->ViewUniformBuffer = View.ViewUniformBuffer;PassParameters->SceneTextures = SceneTextures;TShaderMapRef<FMyRTAORGS> RayGenerationShader(GetGlobalShaderMap(FeatureLevel));FIntPoint RayTracingResolution = View.ViewRect.Size();ClearUnusedGraphResources(RayGenerationShader, PassParameters);GraphBuilder.AddPass(RDG_EVENT_NAME("M_RTAO"),PassParameters,ERDGPassFlags::Compute,[PassParameters, this, &View, RayGenerationShader, RayTracingResolution](FRHICommandList& RHICmdList){FRayTracingShaderBindingsWriter GlobalResources;SetShaderParameters(GlobalResources, RayGenerationShader, *PassParameters);// TODO: Provide material support for opacity maskFRayTracingPipelineState* Pipeline = View.RayTracingMaterialPipeline;//------------------------------------------------------// Declare default pipelineFRayTracingPipelineStateInitializer Initializer;Initializer.MaxPayloadSizeInBytes = 64; // sizeof(FPackedMaterialClosestHitPayload)FRHIRayTracingShader* RayGenShaderTable[] = { RayGenerationShader.GetRayTracingShader() };Initializer.SetRayGenShaderTable(RayGenShaderTable);FRHIRayTracingShader* HitGroupTable[] = { View.ShaderMap->GetShader<FOpaqueShadowHitGroup>().GetRayTracingShader() };Initializer.SetHitGroupTable(HitGroupTable);Initializer.bAllowHitGroupIndexing = false; // Use the same hit shader for all geometry in the scene by disabling SBT indexing.Pipeline = PipelineStateCache::GetAndOrCreateRayTracingPipelineState(RHICmdList, Initializer);//---------------------------------------------------------FRHIRayTracingScene* RayTracingSceneRHI = View.RayTracingScene.RayTracingSceneRHI;RHICmdList.RayTraceDispatch(Pipeline, RayGenerationShader.GetRayTracingShader(), RayTracingSceneRHI, GlobalResources, RayTracingResolution.X, RayTracingResolution.Y);});int32 DenoiserMode = 1;if (DenoiserMode != 0){IScreenSpaceDenoiser::FAmbientOcclusionRayTracingConfig RayTracingConfig;RayTracingConfig.RayCountPerPixel = 2;const IScreenSpaceDenoiser* DefaultDenoiser = IScreenSpaceDenoiser::GetDefaultDenoiser();const IScreenSpaceDenoiser* DenoiserToUse = DenoiserMode == 1 ? DefaultDenoiser : GScreenSpaceDenoiser;RDG_EVENT_SCOPE(GraphBuilder, "%s%s(AmbientOcclusion) %dx%d",DenoiserToUse != DefaultDenoiser ? TEXT("ThirdParty ") : TEXT(""),DenoiserToUse->GetDebugName(),View.ViewRect.Width(), View.ViewRect.Height());IScreenSpaceDenoiser::FAmbientOcclusionOutputs DenoiserOutputs = DenoiserToUse->DenoiseAmbientOcclusion(GraphBuilder,View,&View.PrevViewInfo,SceneTextures,DenoiserInputs,RayTracingConfig);*OutAmbientOcclusionTexture = DenoiserOutputs.AmbientOcclusionMask;
}else{*OutAmbientOcclusionTexture = DenoiserInputs.Mask;}
}
#else
{unimplemented();
}
#endif
#endif