UnrealEngineUWP/Engine/Shaders/DeferredLightingCommon.usf

// Copyright 1998-2016 Epic Games, Inc. All Rights Reserved.

/*=============================================================================
	DeferredLightingCommon.usf: Common definitions for deferred lighting.
=============================================================================*/

#ifndef __DEFERRED_LIGHTING_COMMON__
#define __DEFERRED_LIGHTING_COMMON__

#include "DeferredShadingCommon.usf"
#include "DynamicLightingCommon.usf"
#include "BRDF.usf"
#include "MonteCarlo.usf"
#include "IESLightProfilesCommon.usf"
#include "ShadingModels.usf"

/** 
 * Data about a single light.
 * Putting the light data in this struct allows the same lighting code to be used between standard deferred, 
 * Where many light properties are known at compile time, and tiled deferred, where all light properties have to be fetched from a buffer.
 */
struct FDeferredLightData
{
	float4 LightPositionAndInvRadius;
	float4 LightColorAndFalloffExponent;
	float3 LightDirection;
	float4 SpotAnglesAndSourceRadius;
	float MinRoughness;
	float ContactShadowLength;
	float2 DistanceFadeMAD;
	float4 ShadowMapChannelMask;
	/** Whether to use inverse squared falloff. */
	bool bInverseSquared;
	/** Whether this is a light with radial attenuation, aka point or spot light. */
	bool bRadialLight;
	/** Whether this light needs spotlight attenuation. */
	bool bSpotLight;
	/** Whether the light should apply shadowing. */
	uint ShadowedBits;
};

/** Data about a single light to be shaded with the simple shading model, designed for speed and limited feature set. */
struct FSimpleDeferredLightData
{
	float4 LightPositionAndInvRadius;
	float4 LightColorAndFalloffExponent;
	/** Whether to use inverse squared falloff. */
	bool bInverseSquared;
};

#ifdef THREADGROUP_SIZEX
#define REFERENCE_QUALITY	0
#else
#define REFERENCE_QUALITY	0
#endif

#undef LIGHT_SOURCE_SHAPE
#define LIGHT_SOURCE_SHAPE 1

#define LIGHT_SHAPE_SPHERE	1
#define LIGHT_SHAPE_RECT	2

static const uint LightSourceShape = LIGHT_SHAPE_SPHERE;

bool RayHitRect( float3 R, float3 RectCenter, float3 RectX, float3 RectY, float3 RectZ, float RectExtentX, float RectExtentY )
{
	// Intersect ray with plane
	float3 PointOnPlane = R * max( 0, dot( RectZ, RectCenter ) / dot( RectZ, R ) );

	bool InExtentX = abs( dot( RectX, PointOnPlane - RectCenter ) ) <= RectExtentX;
	bool InExtentY = abs( dot( RectY, PointOnPlane - RectCenter ) ) <= RectExtentY;
	return InExtentX && InExtentY;
}

float3 SphereLightingMIS( FGBufferData GBuffer, FDeferredLightData LightData, float3 LobeRoughness, float3 ToLight, float3 V, float3 N, uint2 Random )
{
	float3 Lighting = 0;

	LobeRoughness = max( 0.08, LobeRoughness );

	const float SourceRadius = max( 1, LightData.SpotAnglesAndSourceRadius.z );

	const float DistanceSqr = dot( ToLight, ToLight );
	const float3 ConeAxis = ToLight * rsqrt( DistanceSqr );
	const float ConeCos = sqrt( 1 - Square( SourceRadius ) / DistanceSqr );

	const float Area = PI * Square(SourceRadius);
	const float SampleColor = 1.0 / Area;
	
	const uint NumSets = 3;
	const uint NumSamples[ NumSets ] =
	{
		0,
		4,
		4,
	};

	UNROLL
	for( uint Set = 0; Set < NumSets; Set++ )
	{
		LOOP
		for( uint i = 0; i < NumSamples[ Set ]; i++ )
		{
			float2 E = Hammersley( i, NumSamples[ Set ], Random );
			//float2 E = CorrelatedMultiJitter2D( i, NumSamples[ Set ], Random.x * Random.y * (Set + 17) );
			
			float3 L, H;
			if( Set == 0 )
			{
				L = TangentToWorld( CosineSampleHemisphere( E ).xyz, N );
				H = normalize(V + L);
			}
			else if( Set == 1 )
			{
				H = TangentToWorld( ImportanceSampleGGX( E, LobeRoughness[1] ).xyz, N );
				L = 2 * dot( V, H ) * H - V;
			}
			else
			{
				L = TangentToWorld( UniformSampleCone( E, ConeCos ).xyz, ConeAxis );
				H = normalize(V + L);
			}

			float NoL = saturate( dot(N, L) );
			float NoH = saturate( dot(N, H) );
			float VoH = saturate( dot(V, H) );

			if( NoL > 0 && VoH > 0 )
			{
				BRANCH
				if( Set != 2 && dot( L, ConeAxis ) < ConeCos )
				{
					// Ray misses sphere
					continue;
				}

				float PDF[] =
				{
					NoL / PI,
					D_GGX( LobeRoughness[1], NoH ) * NoH / (4 * VoH),
					1.0 / ( 2 * PI * (1 - ConeCos) ),
				};

				// MIS balance heuristic
				float InvWeight = 0;
				UNROLL for( uint j = 0; j < NumSets; j++ )
				{
					InvWeight += PDF[j] * NumSamples[j];
				}
				float Weight = rcp( InvWeight );

				float3 Shading = SurfaceShading( GBuffer, LobeRoughness, 1, L, V, N, 1, Random ) * NoL;
				Shading += SubsurfaceShading( GBuffer, L, V, N, 1, Random );
			
				Lighting += SampleColor * Shading * Weight;
			}
		}
	}

	return Lighting;
}

// Find representative incoming light direction and energy modification
float3 AreaLightSpecular( FDeferredLightData LightData, inout float3 LobeRoughness, inout float3 ToLight, inout float3 L, float3 V, half3 N )
{
	float3 LobeEnergy = 1;

	LobeRoughness = max( LobeRoughness, LightData.MinRoughness );
	float3 m = LobeRoughness * LobeRoughness;
	
	const float SourceRadius = LightData.SpotAnglesAndSourceRadius.z;
	const float SourceLength = LightData.SpotAnglesAndSourceRadius.w;

	// TODO early out for point lights
	
	float3 R = reflect( -V, N );
	float InvDistToLight = rsqrt( dot( ToLight, ToLight ) );

	// Point lobe in off-specular peak direction
	float a = Square( LobeRoughness[1] );
	R = lerp( N, R, (1 - a) * ( sqrt(1 - a) + a ) );
	R = normalize( R );

	BRANCH
	if( SourceLength > 0 )
	{
		// Energy conservation
		// asin(x) is angle to sphere, atan(x) is angle to disk, saturate(x) is free and in the middle
		float LineAngle = saturate( SourceLength * InvDistToLight );
		LobeEnergy *= m / saturate( m + 0.5 * LineAngle );

		// Closest point on line segment to ray
		float3 L01 = LightData.LightDirection * SourceLength;
		float3 L0 = ToLight - 0.5 * L01;
		float3 L1 = ToLight + 0.5 * L01;

#if 1
		// Shortest distance
		float a = Square( SourceLength );
		float b = dot( R, L01 );
		float t = saturate( dot( L0, b*R - L01 ) / (a - b*b) );
#else
		// Smallest angle
		float A = Square( SourceLength );
		float B = 2 * dot( L0, L01 );
		float C = dot( L0, L0 );
		float D = dot( R, L0 );
		float E = dot( R, L01 );
		float t = saturate( (B*D - 2*C*E) / (B*E - 2*A*D) );
#endif

		ToLight = L0 + t * L01;
	}

	BRANCH
	if( SourceRadius > 0 )
	{
		// Energy conservation
		// asin(x) is angle to sphere, atan(x) is angle to disk, saturate(x) is free and in the middle
		float SphereAngle = saturate( SourceRadius * InvDistToLight );
		LobeEnergy *= Square( m / saturate( m + 0.5 * SphereAngle ) );
		
		// Closest point on sphere to ray
		float3 ClosestPointOnRay = dot( ToLight, R ) * R;
		float3 CenterToRay = ClosestPointOnRay - ToLight;
		float3 ClosestPointOnSphere = ToLight + CenterToRay * saturate( SourceRadius * rsqrt( dot( CenterToRay, CenterToRay ) ) );
		ToLight = ClosestPointOnSphere;
	}

	L = normalize( ToLight );

	return LobeEnergy;
}

/** Returns 0 for positions closer than the fade near distance from the camera, and 1 for positions further than the fade far distance. */
float DistanceFromCameraFade(float SceneDepth, FDeferredLightData LightData, float3 WorldPosition, float3 CameraPosition)
{
	// depth (non radial) based fading over distance
	float Fade = saturate(SceneDepth * LightData.DistanceFadeMAD.x + LightData.DistanceFadeMAD.y);
	return Fade * Fade;
}

void GetShadowTerms(FGBufferData GBuffer, FDeferredLightData LightData, float3 WorldPosition, float4 LightAttenuation, out float OpaqueShadowTerm, out float SSSShadowTerm)
{
	// Remapping the light attenuation buffer (see ShadowRendering.cpp)

	// LightAttenuation: Light function + per-object shadows in z, per-object SSS shadowing in w, 
	// Whole scene directional light shadows in x, whole scene directional light SSS shadows in y
	// Get static shadowing from the appropriate GBuffer channel
	float UsesStaticShadowMap = dot(LightData.ShadowMapChannelMask, float4(1, 1, 1, 1));
	float StaticShadowing = lerp(1, dot(GBuffer.PrecomputedShadowFactors, LightData.ShadowMapChannelMask), UsesStaticShadowMap);

	if (LightData.bRadialLight)
	{
		// Remapping the light attenuation buffer (see ShadowRendering.cpp)

		OpaqueShadowTerm = LightAttenuation.z * StaticShadowing;
		// SSS uses a separate shadowing term that allows light to penetrate the surface
		//@todo - how to do static shadowing of SSS correctly?
		SSSShadowTerm = LightAttenuation.w * StaticShadowing;
	}
	else
	{
		// Remapping the light attenuation buffer (see ShadowRendering.cpp)
		// Also fix up the fade between dynamic and static shadows
		// to work with plane splits rather than spheres.

		float DynamicShadowFraction = DistanceFromCameraFade(GBuffer.Depth, LightData, WorldPosition, View.WorldCameraOrigin);
		// For a directional light, fade between static shadowing and the whole scene dynamic shadowing based on distance + per object shadows
		OpaqueShadowTerm = lerp(LightAttenuation.x, StaticShadowing, DynamicShadowFraction);
		// Fade between SSS dynamic shadowing and static shadowing based on distance
		SSSShadowTerm = min(lerp(LightAttenuation.y, StaticShadowing, DynamicShadowFraction), LightAttenuation.w);
		
		// combine with light function
		OpaqueShadowTerm *= LightAttenuation.z;
		SSSShadowTerm *= LightAttenuation.z;
	}
}

float ShadowRayCast(
	float3 RayOriginTranslatedWorld, float3 RayDirection, float RayLength,
	int NumSteps, float StepOffset
)
{
	float4 RayStartClip	= mul( float4( RayOriginTranslatedWorld, 1 ), View.TranslatedWorldToClip );
	float4 RayDirClip	= mul( float4( RayDirection * RayLength, 0 ), View.TranslatedWorldToClip );
	float4 RayEndClip	= RayStartClip + RayDirClip;

	float3 RayStartScreen = RayStartClip.xyz / RayStartClip.w;
	float3 RayEndScreen = RayEndClip.xyz / RayEndClip.w;
	
	float3 RayStepScreen = RayEndScreen - RayStartScreen;

	float3 RayStartUVz = float3( RayStartScreen.xy * View.ScreenPositionScaleBias.xy + View.ScreenPositionScaleBias.wz, RayStartScreen.z );
	float3 RayStepUVz = float3( RayStepScreen.xy * View.ScreenPositionScaleBias.xy, RayStepScreen.z );

	float4 RayDepthClip	= RayStartClip + mul( float4( 0, 0, RayLength, 0 ), View.ViewToClip );
	float3 RayDepthScreen = RayDepthClip.xyz / RayDepthClip.w;

	const float Step = 1.0 / NumSteps;

	// *2 to get less morie pattern in extreme cases, larger values make object appear not grounded in reflections
	const float CompareTolerance = abs( RayDepthScreen.z - RayStartScreen.z ) * Step * 2;

	float SampleTime = StepOffset * Step + Step;
	
	float Coverage = 0;

	UNROLL
	for( int i = 0; i < NumSteps; i++ )
	{
		float3 SampleUVz = RayStartUVz + RayStepUVz * SampleTime;
		float SampleDepth = SceneDepthTexture.SampleLevel( SceneDepthTextureSampler, SampleUVz.xy, 0 ).r;

		float DepthDiff = SampleUVz.z - SampleDepth;
		bool Hit = abs( DepthDiff + CompareTolerance ) < CompareTolerance;

		Coverage += Hit ? 1.0 : 0.0;

		SampleTime += Step;
	}

	float Shadow = Coverage > 0 ? 1 : 0;

	// Off screen masking
	float2 Vignette = saturate( abs( RayStartScreen.xy ) * 6 - 5 );
	//Shadow *= saturate( 1.0 - dot( Vignette, Vignette ) );

	return 1 - Shadow;
}

/** Calculates lighting for a given position, normal, etc with a fully featured lighting model designed for quality. */
float4 GetDynamicLighting(float3 WorldPosition, float3 CameraVector, FGBufferData GBuffer, float AmbientOcclusion, uint ShadingModelID, FDeferredLightData LightData, float4 LightAttenuation, uint2 Random)
{
	FLightAccumulator LightAccumulator = (FLightAccumulator)0;

	float3 V = -CameraVector;
	float3 N = GBuffer.WorldNormal;
	float3 ToLight = LightData.LightDirection;
	float3 L = ToLight;	// no need to normalize
	float NoL = saturate( dot(N, L) );
	float DistanceAttenuation = 1;
	float LightRadiusMask = 1;
	float SpotFalloff = 1;

	if (LightData.bRadialLight)
	{
		ToLight = LightData.LightPositionAndInvRadius.xyz - WorldPosition;
		
		float DistanceSqr = dot( ToLight, ToLight );
		L = ToLight * rsqrt( DistanceSqr );

		if (LightData.bInverseSquared)
		{
			const float SourceRadius = LightData.SpotAnglesAndSourceRadius.z;
			const float SourceLength = LightData.SpotAnglesAndSourceRadius.w;

			BRANCH
			if( SourceLength > 0 )
			{
				// Line segment irradiance
				float3 L01 = LightData.LightDirection * SourceLength;
				float3 ToLight0 = ToLight - 0.5 * L01;
				float3 ToLight1 = ToLight + 0.5 * L01;

				float LengthSqr0 = dot( ToLight0, ToLight0 );
				float LengthSqr1 = dot( ToLight1, ToLight1 );
				float rLength0 = rsqrt( LengthSqr0 );
				float rLength1 = rsqrt( LengthSqr1 );
				float Length0 = LengthSqr0 * rLength0;
				float Length1 = LengthSqr1 * rLength1;

				DistanceAttenuation = rcp( ( Length0 * Length1 + dot( ToLight0, ToLight1 ) ) * 0.5 + 1 );
				NoL = saturate( 0.5 * ( dot(N, ToLight0) * rLength0 + dot(N, ToLight1) * rLength1 ) );
			}
			else
			{
				DistanceAttenuation = rcp( DistanceSqr + 1 );
				NoL = saturate( dot( N, L ) );

				if( SourceRadius > 0 )
				{
				#if 1	//HORIZON
					NoL = dot( N, L );

					float SinAlphaSqr = saturate( Square( SourceRadius ) / DistanceSqr );
					float SinAlpha = sqrt( SinAlphaSqr );

					if( NoL < SinAlpha )
					{
					#if 0
						// Accurate sphere irradiance
						float CosBeta = NoL;
						float SinBeta = sqrt( 1 - CosBeta * CosBeta );
						float TanBeta = SinBeta / CosBeta;
					
						float x = sqrt( 1 / SinAlphaSqr - 1 );
						float y = -x / TanBeta;
						float z = SinBeta * sqrt(1 - y*y);

						DistanceAttenuation = SinAlphaSqr * ( NoL * acos(y) - x * z ) + atan( z / x );
						DistanceAttenuation /= PI * Square( SourceRadius );
						NoL = 1;
					#else
						// Hermite spline approximation
						// Fairly accurate with SinAlpha < 0.8
						// y=0 and dy/dx=0 at -SinAlpha
						// y=SinAlpha and dy/dx=1 at SinAlpha
						NoL = max( NoL, -SinAlpha );
						NoL = Square( SinAlpha + NoL ) / ( 4 * SinAlpha );
					#endif
					}
				#endif
				}
			}

			// TODO optimize
			LightRadiusMask = Square( saturate( 1 - Square( DistanceSqr * Square(LightData.LightPositionAndInvRadius.w) ) ) );
		}
		else
		{
			DistanceAttenuation = 1;
			NoL = saturate( dot( N, L ) );
			
			LightRadiusMask = RadialAttenuation(ToLight * LightData.LightPositionAndInvRadius.w, LightData.LightColorAndFalloffExponent.w);

#if REFERENCE_QUALITY
			// anti Area
			LightRadiusMask *= DistanceSqr + 1;
#endif
		}

		if (LightData.bSpotLight)
		{
			SpotFalloff = SpotAttenuation(L, -LightData.LightDirection, LightData.SpotAnglesAndSourceRadius.xy);
		}
	}

	LightAccumulator.EstimatedCost += 0.3f;		// running the PixelShader at all has a cost

	BRANCH
	if (LightRadiusMask > 0 && SpotFalloff > 0)
	{
		float SurfaceShadow = 1;
		float SubsurfaceShadow = 1;

		BRANCH
		if (LightData.ShadowedBits)
		{
			GetShadowTerms(GBuffer, LightData, WorldPosition, LightAttenuation, SurfaceShadow, SubsurfaceShadow);

			// greatly reduces shadow mapping artifacts
			// Commented out because it reduces character shading quality.
			//SurfaceShadow *= saturate(dot(N, L) * 6 - 0.2);

#ifndef THREADGROUP_SIZEX
			BRANCH
			if( LightData.ShadowedBits > 1 && LightData.ContactShadowLength > 0 )
			{
				uint FrameRandom = View.StateFrameIndexMod8 * 1551;

				float StepOffset = float( Random.x & 0xffff ) / (1<<16);
				StepOffset -= 0.5;

				float Shadow = ShadowRayCast( WorldPosition + View.PreViewTranslation, L, LightData.ContactShadowLength * GBuffer.Depth, 8, StepOffset );
				
				SurfaceShadow *= Shadow;
				//SubsurfaceShadow *= Shadow;
			}
#endif
		}
		else
		{
			SurfaceShadow = AmbientOcclusion;
		}

#ifndef THREADGROUP_SIZEX
		BRANCH
		if( LightData.ShadowedBits < 2 && GBuffer.ShadingModelID == SHADINGMODELID_HAIR )
		{
			uint FrameRandom = View.StateFrameIndexMod8 * 1551;

			float StepOffset = float( Random.x & 0xffff ) / (1<<16);
			StepOffset -= 0.5;

			SubsurfaceShadow = ShadowRayCast( WorldPosition + View.PreViewTranslation, L, 0.1 * GBuffer.Depth, 8, StepOffset );
		}
#endif

		float SurfaceAttenuation	= (DistanceAttenuation * LightRadiusMask * SpotFalloff) * SurfaceShadow;
		float SubsurfaceAttenuation	= (DistanceAttenuation * LightRadiusMask * SpotFalloff) * SubsurfaceShadow;

		LightAccumulator.EstimatedCost += 0.3f;		// add the cost of getting the shadow terms

		{
			const float3 LightColor = LightData.LightColorAndFalloffExponent.rgb;

			const float ClearCoatRoughness	= GBuffer.CustomData.y;

			float3 LobeRoughness = float3(ClearCoatRoughness, GBuffer.Roughness, 1);
			
#if REFERENCE_QUALITY
			float LightMask = LightRadiusMask * SpotFalloff * SurfaceShadow;
			LightAccumulator_Add( LightAccumulator, SphereLightingMIS( GBuffer, LightData, LobeRoughness, ToLight, V, N, Random ), 0, LightColor * LightMask );
#else
			float3 LobeEnergy = AreaLightSpecular(LightData, LobeRoughness, ToLight, L, V, N);

			// accumulate diffuse and specular
			{
#if 1	// for testing if there is a perf impact
				// correct screen space subsurface scattering
				float3 SurfaceLightingDiff = SurfaceShading(GBuffer, LobeRoughness, LobeEnergy, L, V, N, float2(1, 0), Random);
				float3 SurfaceLightingSpec = SurfaceShading(GBuffer, LobeRoughness, LobeEnergy, L, V, N, float2(0, 1), Random);
				LightAccumulator_Add(LightAccumulator, SurfaceLightingDiff, SurfaceLightingSpec, LightColor * (NoL * SurfaceAttenuation));
				//LightAccumulator_Add(LightAccumulator, SurfaceLightingDiff, SurfaceLightingSpec, LightColor * (SurfaceAttenuation));
#else
				// wrong screen space subsurface scattering but potentially faster
				float3 SurfaceLighting = SurfaceShading(GBuffer, LobeRoughness, LobeEnergy, L, V, N, float2(1, 1), Random);
				LightAccumulator_Add(LightAccumulator, SurfaceLighting, 0, LightColor * (NoL * SurfaceAttenuation));
#endif
			}

			// accumulate subsurface
			{
				float3 SubsurfaceLighting = SubsurfaceShading(GBuffer, L, V, N, SubsurfaceShadow, Random);

				LightAccumulator_Add(LightAccumulator, SubsurfaceLighting, 0, LightColor * SubsurfaceAttenuation);

				LightAccumulator.EstimatedCost += 0.4f;		// add the cost of the lighting computations (should sum up to 1 form one light)
			}
#endif
		}
	}

	return LightAccumulator_GetResult(LightAccumulator);
}

/** 
 * Calculates lighting for a given position, normal, etc with a simple lighting model designed for speed. 
 * All lights rendered through this method are unshadowed point lights with no shadowing or light function or IES.
 * A cheap specular is used instead of the more correct area specular, no fresnel.
 */
float3 GetSimpleDynamicLighting(float3 WorldPosition, float3 CameraVector, FScreenSpaceData ScreenSpaceData, FSimpleDeferredLightData LightData)
{
	float3 V = -CameraVector;
	float3 N = ScreenSpaceData.GBuffer.WorldNormal;
	float3 ToLight = LightData.LightPositionAndInvRadius.xyz - WorldPosition;
	float DistanceAttenuation = 1;
	
	float DistanceSqr = dot( ToLight, ToLight );
	float3 L = ToLight * rsqrt( DistanceSqr );
	float NoL = saturate( dot( N, L ) );

	if (LightData.bInverseSquared)
	{
		// Sphere falloff (technically just 1/d2 but this avoids inf)
		DistanceAttenuation = 1 / ( DistanceSqr + 1 );
	
		float LightRadiusMask = Square( saturate( 1 - Square( DistanceSqr * Square(LightData.LightPositionAndInvRadius.w) ) ) );
		DistanceAttenuation *= LightRadiusMask;
	}
	else
	{
		DistanceAttenuation = RadialAttenuation(ToLight * LightData.LightPositionAndInvRadius.w, LightData.LightColorAndFalloffExponent.w);
	}

	float3 OutLighting = 0;

	BRANCH
	if (DistanceAttenuation > 0)
	{
		const float3 LightColor = LightData.LightColorAndFalloffExponent.rgb;

		// Apply SSAO to the direct lighting since we're not going to have any other shadowing
		float Attenuation = DistanceAttenuation * ScreenSpaceData.AmbientOcclusion;

		OutLighting += LightColor * (NoL * Attenuation) * SimpleShading( ScreenSpaceData.GBuffer.DiffuseColor, ScreenSpaceData.GBuffer.SpecularColor, ScreenSpaceData.GBuffer.Roughness, L, V, N );
	}

	return OutLighting;
}

#endif