UnrealEngineUWP/Engine/Shaders/DistanceFieldSurfaceCacheLighting.usf

// Copyright 1998-2015 Epic Games, Inc. All Rights Reserved.

/*=============================================================================
	DistanceFieldSurfaceCacheLighting.usf
=============================================================================*/

#include "Common.usf"
#include "DeferredShadingCommon.usf"
#include "DistanceFieldLightingShared.usf"
#include "DistanceFieldAOShared.usf"

struct FObjectCullVertexOutput
{
	nointerpolation float4 PositionAndRadius : TEXCOORD0;
	nointerpolation uint ObjectIndex : TEXCOORD1;
};
 
float ConservativeRadiusScale;

/** Used when culling objects into screenspace tile lists */
void ObjectCullVS(
	float4 InPosition : ATTRIBUTE0,
	uint ObjectIndex : SV_InstanceID,
	out FObjectCullVertexOutput Output,
	out float4 OutPosition : SV_POSITION
	)
{
	//@todo - implement ConservativelyBoundSphere
	float4 ObjectPositionAndRadius = LoadObjectPositionAndRadius(ObjectIndex);
	//@todo - expand to handle conservative rasterization
	float EffectiveRadius = (ObjectPositionAndRadius.w + AOMaxDistance) * ConservativeRadiusScale;
	float3 WorldPosition = InPosition.xyz * EffectiveRadius + ObjectPositionAndRadius.xyz;
	OutPosition = mul(float4(WorldPosition, 1), View.WorldToClip);
	Output.PositionAndRadius = ObjectPositionAndRadius;
	Output.ObjectIndex = ObjectIndex;
} 

RWBuffer<uint> RWTileHeadDataUnpacked;
RWBuffer<uint> RWTileArrayData;

Buffer<float4> TileConeAxisAndCos;
Buffer<float4> TileConeDepthRanges;

float2 NumGroups;

/** Intersects a single object with the tile and adds to the intersection list if needed. */
void ObjectCullPS(
	FObjectCullVertexOutput Input, 
	in float4 SVPos : SV_POSITION,
	out float4 OutColor : SV_Target0)
{
	OutColor = 0;
	
	uint2 TilePosition = (uint2)SVPos.xy;
	uint TileIndex = TilePosition.y * NumGroups.x + TilePosition.x;
	float4 ConeAxisAndCos = TileConeAxisAndCos.Load(TileIndex);
	float4 ConeAxisDepthRanges = TileConeDepthRanges.Load(TileIndex);
	float3 TileConeVertex = 0;
	float3 TileConeAxis = ConeAxisAndCos.xyz;
	float TileConeAngleCos = ConeAxisAndCos.w;
	float TileConeAngleSin = sqrt(1 - TileConeAngleCos * TileConeAngleCos);

	// A value of 1 is conservative, but has a huge impact on performance
	float RadiusScale = .5f;
	float4 SphereCenterAndRadius = Input.PositionAndRadius;
	float3 ViewSpaceSphereCenter = mul(float4(SphereCenterAndRadius.xyz + View.PreViewTranslation.xyz, 1), View.TranslatedWorldToView).xyz;
	
	int SmallestGroupIndex = -1;

	UNROLL
	for (int GroupIndex = NUM_CULLED_OBJECT_LISTS - 1; GroupIndex >= 0; GroupIndex--)
	{
		uint StartIndex;
		uint EndIndex;
		GetPhaseParameters(GroupIndex, StartIndex, EndIndex);
		float GroupMaxSampleRadius = GetStepOffset(EndIndex) * 2 * RadiusScale;
	
		BRANCH
		if (SphereIntersectConeWithDepthRanges(float4(ViewSpaceSphereCenter, SphereCenterAndRadius.w + GroupMaxSampleRadius), TileConeVertex, TileConeAxis, TileConeAngleCos, TileConeAngleSin, ConeAxisDepthRanges))
		{
			SmallestGroupIndex = GroupIndex;
		}
	}

	if (SmallestGroupIndex >= 0)
	{
		uint ArrayIndex;
		InterlockedAdd(RWTileHeadDataUnpacked[TileIndex * 4 + 1 + (uint)SmallestGroupIndex], 1U, ArrayIndex);

		if (ArrayIndex < MAX_OBJECTS_PER_TILE)
		{
			// Note: indexing so that threads are writing to RWTileArrayData coherently, has a huge impact on speed, even though the array data for one record is no longer dense
			uint DataIndex = (ArrayIndex * (uint)(NumGroups.x * NumGroups.y + .5f) + TileIndex) * NUM_CULLED_OBJECT_LISTS + SmallestGroupIndex;

			RWTileArrayData[DataIndex] = Input.ObjectIndex;
		}
	}
}

/** Computes the distance field normal, using a search through all the nearby objects to find the closest one, whose normal is used. */
void ComputeDistanceFieldNormalPS(
	in float4 UVAndScreenPos : TEXCOORD0, 
	in float4 SVPos : SV_POSITION,
	out float4 OutColor : SV_Target0)
{
	// Sample from the center of the top left full resolution texel
	float2 ScreenUV = float2((floor(SVPos.xy) * DOWNSAMPLE_FACTOR + View.ViewRectMin.xy + .5f) * View.ViewSizeAndSceneTexelSize.zw);
	float SceneDepth = CalcSceneDepth(ScreenUV);
	FGBufferData GBufferData = GetGBufferData(ScreenUV);

	OutColor = EncodeDownsampledGBuffer(GBufferData, SceneDepth);
}

RWTexture2D<float4> RWDistanceFieldNormal;

[numthreads(THREADGROUP_SIZEX, THREADGROUP_SIZEY, 1)]
void ComputeDistanceFieldNormalCS(
	uint3 GroupId : SV_GroupID,
	uint3 DispatchThreadId : SV_DispatchThreadID,
    uint3 GroupThreadId : SV_GroupThreadID) 
{
	float2 ScreenUV = float2((DispatchThreadId.xy * DOWNSAMPLE_FACTOR + View.ViewRectMin.xy + .5f) * View.ViewSizeAndSceneTexelSize.zw);
	float SceneDepth = CalcSceneDepth(ScreenUV);
	FGBufferData GBufferData = GetGBufferData(ScreenUV);

	float4 OutValue = EncodeDownsampledGBuffer(GBufferData, SceneDepth);
	RWDistanceFieldNormal[DispatchThreadId.xy] = OutValue;
}

void WriteDownsampledDepthPS(
	in float4 UVAndScreenPos : TEXCOORD0, 
	out float4 OutColor : SV_Target0,
	out float OutDepth : SV_DEPTH)
{
	float2 DistanceFieldUVs = UVAndScreenPos.xy;

	OutColor = 0;
	OutDepth = ConvertToDeviceZ(GetDownsampledDepth(DistanceFieldUVs));
}

// For some reason gives innaccurate results at lower resolutions
#define USE_SCREENVECTOR_WORLD_POSITION FINAL_INTERPOLATION_PASS

struct FIrradianceCacheSplatVertexOutput
{
	nointerpolation float4 PositionRadius : TEXCOORD0;
	nointerpolation float4 NormalAndFade : TEXCOORD1;
#if FINAL_INTERPOLATION_PASS
	nointerpolation float3 BentNormal : TEXCOORD2;
	#if SUPPORT_IRRADIANCE
		nointerpolation float3 Irradiance : TEXCOORD3;
	#endif
#endif
#if USE_SCREENVECTOR_WORLD_POSITION
	float3 ScreenVector : TEXCOORD5;
#endif
};

float InterpolationRadiusScale;
float2 NormalizedOffsetToPixelCenter;
float HackExpand;

// +1 = behind volume (safe from near plane clipping without depth testing), 
// -1 = in front of volume (required for conservative results with depth testing)
//@todo - using -1 currently causes artifacts with architectural scenes
float InterpolationBoundingDirection;

/** Expands a screen-facing polygon to cover a surface cache record for splatting. */
void IrradianceCacheSplatVS(
	float2 InPosition : ATTRIBUTE0,
	float2 InUV       : ATTRIBUTE1,
	uint JobIndex : SV_InstanceID,
	out FIrradianceCacheSplatVertexOutput Output,
	out float4 OutPosition : SV_POSITION
	)
{
	float4 PositionAndPackedRadius = IrradianceCachePositionRadius[JobIndex];

	float3 RecordPosition = PositionAndPackedRadius.xyz;
	float RecordRadius = abs(PositionAndPackedRadius.w);

	float OccluderRadius = IrradianceCacheOccluderRadius[JobIndex];
	RecordRadius = min(RecordRadius, OccluderRadius);

	float ViewToCenterLength = length(RecordPosition - View.ViewOrigin.xyz);
	float3 NormalizedViewToCenter = (RecordPosition - View.ViewOrigin.xyz) / ViewToCenterLength;

	// Only allow full interpolation expanding for small samples, it won't be noticeable for large samples but adds a lot of splatting cost
	float EffectiveInterpolationRadiusScale = lerp(InterpolationRadiusScale, 1, max(saturate(RecordRadius / 80), .2f));
	RecordRadius = max(RecordRadius, ViewToCenterLength * .01f) * EffectiveInterpolationRadiusScale;

	float OffsetFromCenter = InterpolationBoundingDirection * RecordRadius;
	// Distance from the camera position that NormalizedViewToCenter will intersect the near plane
	float NearPlaneDistance = View.NearPlane / dot(View.ViewForward, NormalizedViewToCenter);

	// Don't move closer than the near plane to avoid getting clipped
	// Clamping at the vertex level only works because it's a screen-aligned triangle
	// Small bias to push just off the near plane
	if (ViewToCenterLength + OffsetFromCenter < NearPlaneDistance + .001f)
	{
		OffsetFromCenter = -ViewToCenterLength + NearPlaneDistance + .001f;
	}

	// Construct a virtual sample position that won't be near plane clipped and will have a positive z after projection
	float3 VirtualPosition = RecordPosition + OffsetFromCenter * NormalizedViewToCenter;
	// Clipping the edge of the circle a bit can save perf, but introduces artifacts in the interpolation
	float RadiusSlack = 1.0f;
	// Compute new radius since we moved the center away from the camera, approximate as similar triangles
	// This is really incorrect because it's not taking into account perspective correction on a sphere - the widest part in screenspace is not at the world space center
	float VirtualRadius = RadiusSlack * RecordRadius * (ViewToCenterLength + OffsetFromCenter) / ViewToCenterLength;

#if !FINAL_INTERPOLATION_PASS
	// Combat the mysterious bug where samples from last frame do not cover this frame's pixel even with no camera movement, seems to be a precision issue
	VirtualRadius += HackExpand * ViewToCenterLength * 4;
#endif

	float4 RecordNormalAndFade = IrradianceCacheNormal[JobIndex];

	float2 CornerUVs = InUV;
	float3 CornerPosition = VirtualPosition + (View.ViewRight * CornerUVs.x + View.ViewUp * CornerUVs.y) * VirtualRadius;

	OutPosition = mul(float4(CornerPosition.xyz, 1), View.WorldToClip);

#if USE_SCREENVECTOR_WORLD_POSITION
	Output.ScreenVector = mul(float4(OutPosition.xy / OutPosition.w, 1, 0), View.ScreenToWorld).xyz;
#endif

	// Move the vertex over to compensate for the difference between the low res pixel center and the high res pixel center where shading is done
	OutPosition.xy += NormalizedOffsetToPixelCenter * OutPosition.w;

	Output.PositionRadius = float4(RecordPosition, RecordRadius); 

#if FADE_RECORDS_OVER_TIME
	float Fade = 1 - (abs(RecordNormalAndFade.w) - 1);
	/*
	FLATTEN
	if (PositionAndPackedRadius.w < 0)
	{
		// Fading out
		Fade = 1 - Fade;
	}
	else
	{
		Fade = max(Fade, .001f);
	}
	*/
#else
	float Fade = 1;
#endif

	Output.NormalAndFade = float4(RecordNormalAndFade.xyz, Fade);

#if FINAL_INTERPOLATION_PASS
	Output.BentNormal = IrradianceCacheBentNormal[JobIndex].xyz;
	#if SUPPORT_IRRADIANCE
		float ViewDistance = length(View.ViewOrigin.xyz - RecordPosition);
		float FadeAlpha = saturate(.001f * (.95f * AOMaxViewDistance - ViewDistance));
		Output.Irradiance = IrradianceCacheIrradiance[JobIndex].xyz * FadeAlpha;
	#endif
#endif
} 

float InterpolationAngleNormalization; 
float InvMinCosPointBehindPlane;

/** Computes surface cache weighting between the current pixel and the record being splatted. */
void IrradianceCacheSplatPS(
	FIrradianceCacheSplatVertexOutput Input, 
	in float4 SVPos : SV_POSITION
	,out float4 OutBentNormal : SV_Target0
#if FINAL_INTERPOLATION_PASS && SUPPORT_IRRADIANCE
	, out float4 OutIrradiance : SV_Target1
#endif
	)
{
	OutBentNormal = 0;
#if FINAL_INTERPOLATION_PASS && SUPPORT_IRRADIANCE
	OutIrradiance = 0;
#endif

	float2 BaseLevelScreenUV = float2(((floor(SVPos.xy)) * DownsampleFactorToBaseLevel + .5f) * BaseLevelTexelSize);

	float3 WorldNormal;
	float SceneDepth;
	bool bHasDistanceFieldRepresentation;
	bool bHasHeightfieldRepresentation;
	GetDownsampledGBuffer(BaseLevelScreenUV, WorldNormal, SceneDepth, bHasDistanceFieldRepresentation, bHasHeightfieldRepresentation);

#if USE_SCREENVECTOR_WORLD_POSITION

	float3 OpaqueWorldPosition = View.ViewOrigin.xyz + Input.ScreenVector * SceneDepth;

#else

	float2 ScreenUV = float2(((floor(SVPos.xy)) * CurrentLevelDownsampleFactor + View.ViewRectMin.xy + float2(.5f, .5f)) * View.ViewSizeAndSceneTexelSize.zw);
	SceneDepth = CalcSceneDepth(ScreenUV); 
	float2 ScreenPosition = (ScreenUV.xy - View.ScreenPositionScaleBias.wz) / View.ScreenPositionScaleBias.xy;
	float4 HomogeneousWorldPosition = mul(float4(ScreenPosition.xy * SceneDepth, SceneDepth, 1), View.ScreenToWorld);
	float3 OpaqueWorldPosition = HomogeneousWorldPosition.xyz / HomogeneousWorldPosition.w;

#endif
	
	float Distance = length(OpaqueWorldPosition - Input.PositionRadius.xyz);
	float DistanceError = saturate(Distance / Input.PositionRadius.w);
	float Weight = 0;

	BRANCH
	if (DistanceError < 1 && SceneDepth < AOMaxViewDistance)
	{ 
		float3 RecordNormal = Input.NormalAndFade.xyz;
		float NormalError = InterpolationAngleNormalization * sqrt(saturate(1 - dot(WorldNormal, RecordNormal)));

		// Don't use a lighting record if it's in front of the query point.
		// Query points behind the lighting record may have nearby occluders that the lighting record does not see.
		// Offset the comparison point along the negative normal to prevent self-occlusion
		float3 RecordToVertexVector = OpaqueWorldPosition - (Input.PositionRadius.xyz - 1 * RecordNormal.xyz);
		float DistanceToVertex = length(RecordToVertexVector);
		float PlaneDistance = dot(RecordNormal.xyz, RecordToVertexVector) / DistanceToVertex;
		
		// Setup an error metric that goes from 0 if the points are coplanar, to 1 if the point being shaded is at the angle corresponding to MinCosPointBehindPlane behind the plane
		float PointBehindPlaneError = min(max(PlaneDistance * InvMinCosPointBehindPlane, 0.0f), DistanceToVertex / 3.0f);

		float PrecisionScale = .1f;
		Weight = saturate(PrecisionScale * Input.NormalAndFade.w * (1 - max(DistanceError, max(NormalError, PointBehindPlaneError))));
		//Weight = saturate(PrecisionScale * Input.NormalAndFade.w * (1 - DistanceError));
		//Weight = Input.NormalAndFade.w;

		float VisualizePlacement = Distance < .1f * Input.PositionRadius.w;

#if FINAL_INTERPOLATION_PASS
		// Pixels without a distance field representation interpolate AO instead of a bent normal
		float3 InterpolatedValue = bHasDistanceFieldRepresentation ? Input.BentNormal : length(Input.BentNormal).xxx;
		OutBentNormal.rgb = InterpolatedValue * Weight;
		#if SUPPORT_IRRADIANCE
			OutIrradiance.rgb = Input.Irradiance * Weight;
		#endif
#endif
		OutBentNormal.a = Weight;
	}
	
	#define VISUALIZE_SPLAT_OVERDRAW 0
	#if VISUALIZE_SPLAT_OVERDRAW && FINAL_INTERPOLATION_PASS
		OutBentNormal.rgb = 0;
		//OutBentNormal.a = Input.Position.w <= SceneDepth ? .01f : 0.0f;
		OutBentNormal.a = .005f;// + (Weight > .001f ? .01f : 0);
	#endif

	#define VISUALIZE_RECORD_POINTS 0
	#if VISUALIZE_RECORD_POINTS && FINAL_INTERPOLATION_PASS
		OutBentNormal.rgb = 1;
		OutBentNormal.a = (DistanceError < .05f) * .1f;
	#endif
}

Buffer<float4> DebugBuffer;

// 1 / ((1 - FadeDistanceFraction) * AOMaxViewDistance)
float DistanceFadeScale;
float SelfOcclusionReplacement;

/** Normalizes the splatted surface cache values, packs depth in alpha. */
void AOCombinePS(
	in float4 UVAndScreenPos : TEXCOORD0
	,out float4 OutBentNormal : SV_Target0
#if SUPPORT_IRRADIANCE
	,out float4 OutIrradiance : SV_Target1
#endif
	)
{
#if SUPPORT_IRRADIANCE
	OutIrradiance = 0;
#endif

	float3 WorldNormal;
	float SceneDepth;
	bool bHasDistanceFieldRepresentation;
	bool bHasHeightfieldRepresentation;
	GetDownsampledGBuffer(UVAndScreenPos.xy, WorldNormal, SceneDepth, bHasDistanceFieldRepresentation, bHasHeightfieldRepresentation);

#define VISUALIZE_ACCUMULATED_WEIGHTS 0
#if VISUALIZE_ACCUMULATED_WEIGHTS

	float3 BentNormalAO = Texture2DSampleLevel(BentNormalAOTexture, BentNormalAOSampler, UVAndScreenPos.xy, 0).aaa;
	OutBentNormal = float4(BentNormalAO, SceneDepth); 

#else

	float4 BentNormalAccumulation = Texture2DSampleLevel(BentNormalAOTexture, BentNormalAOSampler, UVAndScreenPos.xy, 0);

	if (BentNormalAccumulation.w > 0)
	{
		OutBentNormal.rgb = BentNormalAccumulation.xyz / BentNormalAccumulation.w;
		OutBentNormal.a = SceneDepth;

		if (!bHasDistanceFieldRepresentation)
		{
			// Pixels without a distance field representation interpolate AO instead of a bent normal
			// Construct a bent normal from the interpolated AO
			OutBentNormal.rgb = OutBentNormal.r * WorldNormal;
		}

		#if SUPPORT_IRRADIANCE
			OutIrradiance.rgb = Texture2DSampleLevel(IrradianceTexture, IrradianceSampler, UVAndScreenPos.xy, 0).xyz / BentNormalAccumulation.w;

			if (!bHasDistanceFieldRepresentation && !bHasHeightfieldRepresentation)
			{
				OutIrradiance.rgb *= SelfOcclusionReplacement;
			}
		#endif
	}
	else
	{
		OutBentNormal.rgb = float3(0, 0, .1f);
		// Sign bit stores whether texel is valid
		OutBentNormal.a = -SceneDepth;
	}

	float BentNormalLength = length(OutBentNormal.rgb);
	// Fade to unoccluded in the distance
	float FadeAlpha = saturate((AOMaxViewDistance - SceneDepth) * DistanceFadeScale);
	float3 NormalizedBentNormal = OutBentNormal.rgb / max(BentNormalLength, .0001f);
	OutBentNormal.rgb = NormalizedBentNormal * lerp(1, BentNormalLength, FadeAlpha);

	// Clamp Nan's before they get to the filter
	OutBentNormal.rgb = clamp(OutBentNormal.rgb, -1, 1);

	FLATTEN
	if (SceneDepth > AOMaxViewDistance)
	{
		// Mark as valid for the gap filling pass, so we don't get overwritten
		OutBentNormal.a = abs(OutBentNormal.a);
	}

#endif
}

RWTexture2D<float4> RWVisualizeMeshDistanceFields;

bool RayHitSphere( float3 RayOrigin, float3 UnitRayDirection, float3 SphereCenter, float SphereRadius )
{
	float3 ClosestPointOnRay = max( 0, dot( SphereCenter - RayOrigin, UnitRayDirection ) ) * UnitRayDirection;
	float3 CenterToRay = RayOrigin + ClosestPointOnRay - SphereCenter;
	return dot( CenterToRay, CenterToRay ) <= Square( SphereRadius );
}

#define MAX_INTERSECTING_OBJECTS 512
groupshared uint IntersectingObjectIndices[MAX_INTERSECTING_OBJECTS];

groupshared uint NumIntersectingObjects;

void RayTraceThroughTileCulledDistanceFields(float3 WorldRayStart, float3 WorldRayEnd, float MaxRayTime, out float MinRayTime, out float TotalStepsTaken)
{
	MinRayTime = MaxRayTime;
	TotalStepsTaken = 0;

	LOOP
	for (uint ListObjectIndex = 0; ListObjectIndex < min(NumIntersectingObjects, MAX_INTERSECTING_OBJECTS); ListObjectIndex++)
	{
		uint ObjectIndex = IntersectingObjectIndices[ListObjectIndex];
		float4 SphereCenterAndRadius = LoadObjectPositionAndRadius(ObjectIndex);

		float3 LocalPositionExtent = LoadObjectLocalPositionExtent(ObjectIndex);
		float4x4 WorldToVolume = LoadObjectWorldToVolume(ObjectIndex);
		float4 UVScaleAndVolumeScale = LoadObjectUVScale(ObjectIndex);
		float3 UVAdd = LoadObjectUVAdd(ObjectIndex);

		float3 VolumeRayStart = mul(float4(WorldRayStart, 1), WorldToVolume).xyz;
		float3 VolumeRayEnd = mul(float4(WorldRayEnd, 1), WorldToVolume).xyz;
		float3 VolumeRayDirection = VolumeRayEnd - VolumeRayStart;
		float VolumeRayLength = length(VolumeRayDirection);
		VolumeRayDirection /= VolumeRayLength;

		float2 IntersectionTimes = LineBoxIntersect(VolumeRayStart, VolumeRayEnd, -LocalPositionExtent, LocalPositionExtent);

		if (IntersectionTimes.x < IntersectionTimes.y && IntersectionTimes.x < 1)
		{
			float SampleRayTime = IntersectionTimes.x * VolumeRayLength;

			float MinDistance = 1000000;
			float3 IntersectionPosition = float3(0, 0, 0);

			uint StepIndex = 0;
			uint MaxSteps = 256;

			LOOP
			for (; StepIndex < MaxSteps; StepIndex++)
			{
				float3 SampleVolumePosition = VolumeRayStart + VolumeRayDirection * SampleRayTime;
				float3 ClampedSamplePosition = clamp(SampleVolumePosition, -LocalPositionExtent, LocalPositionExtent);
				float3 VolumeUV = DistanceFieldVolumePositionToUV(ClampedSamplePosition, UVScaleAndVolumeScale.xyz, UVAdd);
				float DistanceField = Texture3DSampleLevel(DistanceFieldTexture, DistanceFieldSampler, VolumeUV, 0).x;
				MinDistance = min(MinDistance, DistanceField);
				IntersectionPosition = SampleVolumePosition;

				float MinStepSize = 1.0f / (4 * MaxSteps);
				float StepDistance = max(DistanceField, MinStepSize);
				SampleRayTime += StepDistance;

				// Terminate the trace if we reached a negative area or went past the end of the ray
				if (DistanceField < 0 
					|| SampleRayTime > IntersectionTimes.y * VolumeRayLength)
				{
					break;
				}
			}

			//Result = max(Result, StepIndex / (float)MaxSteps);

			if (MinDistance * UVScaleAndVolumeScale.w < 0 || StepIndex == MaxSteps)
			{
				MinRayTime = min(MinRayTime, SampleRayTime * UVScaleAndVolumeScale.w);
			}

			TotalStepsTaken += StepIndex;
		}
	}
}

[numthreads(THREADGROUP_SIZEX, THREADGROUP_SIZEY, 1)]
void VisualizeMeshDistanceFieldCS(
	uint3 GroupId : SV_GroupID,
	uint3 DispatchThreadId : SV_DispatchThreadID,
    uint3 GroupThreadId : SV_GroupThreadID) 
{
	uint ThreadIndex = GroupThreadId.y * THREADGROUP_SIZEX + GroupThreadId.x;

	float2 ScreenUV = float2((DispatchThreadId.xy * DOWNSAMPLE_FACTOR + View.ViewRectMin.xy + .5f) * View.ViewSizeAndSceneTexelSize.zw);
	float2 ScreenPosition = (ScreenUV.xy - View.ScreenPositionScaleBias.wz) / View.ScreenPositionScaleBias.xy;

	float SceneDepth = CalcSceneDepth(ScreenUV);
	float4 HomogeneousWorldPosition = mul(float4(ScreenPosition * SceneDepth, SceneDepth, 1), View.ScreenToWorld);
	float3 OpaqueWorldPosition = HomogeneousWorldPosition.xyz / HomogeneousWorldPosition.w;

	float TraceDistance = 20000;
	float3 WorldRayStart = View.ViewOrigin.xyz;
	float3 WorldRayEnd = WorldRayStart + normalize(OpaqueWorldPosition - View.ViewOrigin.xyz) * TraceDistance;
	float3 WorldRayDirection = WorldRayEnd - WorldRayStart;
	float3 UnitWorldRayDirection = normalize(WorldRayDirection);

	if (ThreadIndex == 0)
	{
		NumIntersectingObjects = 0;
	}

	GroupMemoryBarrierWithGroupSync();

	uint NumCulledObjects = GetCulledNumObjects();

	LOOP
	for (uint ObjectIndex = ThreadIndex; ObjectIndex < NumCulledObjects; ObjectIndex += THREADGROUP_TOTALSIZE)
	{
		float4 SphereCenterAndRadius = LoadObjectPositionAndRadius(ObjectIndex);

		//@todo - make independent of current pixel
		BRANCH
		if (RayHitSphere(WorldRayStart, UnitWorldRayDirection, SphereCenterAndRadius.xyz, SphereCenterAndRadius.w))
		{
			uint ListIndex;
			InterlockedAdd(NumIntersectingObjects, 1U, ListIndex);

			if (ListIndex < MAX_INTERSECTING_OBJECTS)
			{
				IntersectingObjectIndices[ListIndex] = ObjectIndex; 
			}
		}
	}

	GroupMemoryBarrierWithGroupSync();

	float MinRayTime;
	float TotalStepsTaken;

	// Trace once to find the distance to first intersection
	RayTraceThroughTileCulledDistanceFields(WorldRayStart, WorldRayEnd, TraceDistance, MinRayTime, TotalStepsTaken);

	float TempMinRayTime;
	// Recompute the ray end point
	WorldRayEnd = WorldRayStart + UnitWorldRayDirection * MinRayTime;
	// Trace a second time to only accumulate steps taken before the first intersection, improves visualization
	RayTraceThroughTileCulledDistanceFields(WorldRayStart, WorldRayEnd, MinRayTime, TempMinRayTime, TotalStepsTaken);

	float3 Result = saturate(TotalStepsTaken / 200.0f);

	if (MinRayTime < TraceDistance)
	{
		Result += .1f;
	}

	RWVisualizeMeshDistanceFields[DispatchThreadId.xy] = float4(Result, 0);
}

Texture2D VisualizeDistanceFieldTexture;
SamplerState VisualizeDistanceFieldSampler;

void VisualizeDistanceFieldUpsamplePS(in float4 UVAndScreenPos : TEXCOORD0, out float4 OutColor : SV_Target0)
{
	// Distance field AO was computed at 0,0 regardless of viewrect min
	float2 DistanceFieldUVs = UVAndScreenPos.xy - View.ViewRectMin.xy * View.ViewSizeAndSceneTexelSize.zw;

	float3 Value = Texture2DSampleLevel(VisualizeDistanceFieldTexture, VisualizeDistanceFieldSampler, DistanceFieldUVs, 0).xyz;

	OutColor = float4(Value, 1);
}