#pragma kernel TileLightListGen_NoDepthRT		LIGHTLISTGEN=TileLightListGen_NoDepthRT
#pragma kernel TileLightListGen_DepthRT			LIGHTLISTGEN=TileLightListGen_DepthRT			ENABLE_DEPTH_TEXTURE_BACKPLANE
#pragma kernel TileLightListGen_DepthRT_MSAA	LIGHTLISTGEN=TileLightListGen_DepthRT_MSAA		ENABLE_DEPTH_TEXTURE_BACKPLANE		MSAA_ENABLED
#pragma kernel ClearAtomic


#include "..\common\ShaderBase.h"
#include "LightDefinitions.cs.hlsl"

//#define EXACT_EDGE_TESTS
#define PERFORM_SPHERICAL_INTERSECTION_TESTS
#define CONV_HULL_TEST_ENABLED

uniform int g_iNrVisibLights;
uniform float4x4 g_mInvScrProjection;
uniform float4x4 g_mScrProjection;

uniform float g_fClustScale;
uniform float g_fClustBase;
uniform float g_fNearPlane;
uniform float g_fFarPlane;
uniform int	  g_iLog2NumClusters;		// numClusters = (1<<g_iLog2NumClusters)

#include "ClusteredUtils.h"


#ifdef MSAA_ENABLED
Texture2DMS<float> g_depth_tex : register( t0 );
#else
Texture2D g_depth_tex : register( t0 );
#endif
StructuredBuffer<float3> g_vBoundsBuffer : register( t1 );
StructuredBuffer<SFiniteLightData> g_vLightData : register( t2 );
StructuredBuffer<SFiniteLightBound> g_data : register( t3 );


#define NR_THREADS			64

// output buffer
RWBuffer<uint> g_vLayeredLightList : register( u0 );
RWBuffer<uint> g_LayeredOffset : register( u1 );
RWBuffer<uint> g_LayeredSingleIdxBuffer : register( u2 );

#ifdef ENABLE_DEPTH_TEXTURE_BACKPLANE
RWBuffer<float> g_fModulUserscale : register( u3 );
#endif


#define MAX_NR_COARSE_ENTRIES		64

groupshared unsigned int coarseList[MAX_NR_COARSE_ENTRIES];
groupshared unsigned int clusterIdxs[MAX_NR_COARSE_ENTRIES/2];
groupshared float4 lightPlanes[4*6];

groupshared uint lightOffs;

#ifdef ENABLE_DEPTH_TEXTURE_BACKPLANE
groupshared int ldsZMax;
#endif

#ifdef EXACT_EDGE_TESTS
groupshared uint ldsIsLightInvisible;
groupshared uint lightOffs2;
#endif

#ifdef PERFORM_SPHERICAL_INTERSECTION_TESTS
groupshared uint lightOffsSph;
#endif


float GetLinearDepth(float zDptBufSpace)	// 0 is near 1 is far
{
	float3 vP = float3(0.0f,0.0f,zDptBufSpace);
	float4 v4Pres = mul(g_mInvScrProjection, float4(vP,1.0));
	return v4Pres.z / v4Pres.w;
}


float3 GetViewPosFromLinDepth(float2 v2ScrPos, float fLinDepth)
{
	float fSx = g_mScrProjection[0].x;
	float fCx = g_mScrProjection[0].z;
	float fSy = g_mScrProjection[1].y;
	float fCy = g_mScrProjection[1].z;

#ifdef LEFT_HAND_COORDINATES
	return fLinDepth*float3( ((v2ScrPos.x-fCx)/fSx), ((v2ScrPos.y-fCy)/fSy), 1.0 );
#else
	return fLinDepth*float3( -((v2ScrPos.x+fCx)/fSx), -((v2ScrPos.y+fCy)/fSy), 1.0 );
#endif
}

float GetOnePixDiagWorldDistAtDepthOne()
{
	float fSx = g_mScrProjection[0].x;
	float fSy = g_mScrProjection[1].y;

	return length( float2(1.0/fSx,1.0/fSy) );
}

void sortLightList(int localThreadID, int n);

#ifdef EXACT_EDGE_TESTS
int CullByExactEdgeTests(uint threadID, int iNrCoarseLights, uint2 viTilLL, uint2 viTilUR, float fTileFarPlane);
#endif
#ifdef PERFORM_SPHERICAL_INTERSECTION_TESTS
int SphericalIntersectionTests(uint threadID, int iNrCoarseLights, float2 screenCoordinate);
#endif


// returns 1 for intersection and 0 for none

float4 GetPlaneEq(const float3 vBoxX, const float3 vBoxY, const float3 vBoxZ, const float3 vCen, const float2 vScaleXZ, const int sideIndex);
float4 FetchPlane(int l, int p);


bool CheckIntersection(int l, int k, uint2 viTilLL, uint2 viTilUR, float fModulUserScale)
{
	unsigned int val = (clusterIdxs[l>>1]>>(16*(l&1)))&0xffff;
	bool bIsHit = ((val>>0)&0xff)<=((uint) k) && ((uint) k)<=((val>>8)&0xff);
	if(bIsHit)
	{
#ifdef CONV_HULL_TEST_ENABLED
		float depthAtNearZ = ClusterIdxToZ(k, fModulUserScale);
		float depthAtFarZ = ClusterIdxToZ(k+1, fModulUserScale);

		for(int p=0; p<6; p++)
		{
			float4 plane = lightPlanes[6*(l&3)+p];
		
			bool bAllInvisib = true;

			for(int i=0; i<8; i++)
			{
				float x = (i&1)==0 ? viTilLL.x : viTilUR.x;
				float y = (i&2)==0 ? viTilLL.y : viTilUR.y;
				float z = (i&4)==0 ? depthAtNearZ : depthAtFarZ;
				float3 vP = GetViewPosFromLinDepth( float2(x, y), z);

				bAllInvisib = bAllInvisib && dot(plane, float4(vP,1.0))>0;
			}

			if(bAllInvisib) bIsHit = false;
		}
#endif
	}

	return bIsHit;
}

bool CheckIntersectionBasic(int l, int k)
{
	unsigned int val = (clusterIdxs[l>>1]>>(16*(l&1)))&0xffff;
	return ((val>>0)&0xff)<=((uint) k) && ((uint) k)<=((val>>8)&0xff);
}


[numthreads(NR_THREADS, 1, 1)]
void LIGHTLISTGEN(uint threadID : SV_GroupIndex, uint3 u3GroupID : SV_GroupID)
{
	uint2 tileIDX = u3GroupID.xy;
	uint t=threadID;

	uint iWidth;
	uint iHeight;
#ifdef MSAA_ENABLED
	uint iNumSamplesMSAA;
	g_depth_tex.GetDimensions(iWidth, iHeight, iNumSamplesMSAA);
#else
	g_depth_tex.GetDimensions(iWidth, iHeight);
#endif
	uint nrTilesX = (iWidth+15)/16;
	uint nrTilesY = (iHeight+15)/16;

	uint2 viTilLL = 16*tileIDX;
	uint2 viTilUR = min( viTilLL+uint2(16,16), uint2(iWidth-1, iHeight-1) );

	if(t==0)
	{
		lightOffs = 0;

#ifdef ENABLE_DEPTH_TEXTURE_BACKPLANE
		ldsZMax = 0;
#endif
	}

#if !defined(XBONE) && !defined(PLAYSTATION4)
	GroupMemoryBarrierWithGroupSync();
#endif

	float dpt_ma=1.0;

#ifdef ENABLE_DEPTH_TEXTURE_BACKPLANE
	// establish min and max depth first
	dpt_ma=0.0;

	for(int idx=t; idx<256; idx+=NR_THREADS)
	{
		uint2 uPixCrd = min( uint2(viTilLL.x+(idx&0xf), viTilLL.y+(idx>>4)), uint2(iWidth-1, iHeight-1) );
#ifdef MSAA_ENABLED
		for(int i=0; i<iNumSamplesMSAA; i++)
		{
		const float fDpth = FetchDepthMSAA(g_depth_tex, uPixCrd, i);
#else
		const float fDpth = FetchDepth(g_depth_tex, uPixCrd);
#endif
		if(fDpth<VIEWPORT_SCALE_Z)		// if not skydome
		{
			dpt_ma = max(fDpth, dpt_ma);
		}
#ifdef MSAA_ENABLED
		}
#endif
	}

	InterlockedMax(ldsZMax, asuint(dpt_ma) );


#if !defined(XBONE) && !defined(PLAYSTATION4)
	GroupMemoryBarrierWithGroupSync();
#endif
	dpt_ma = asfloat(ldsZMax);
#endif

	float3 vTileLL = float3(viTilLL.x/(float) iWidth, viTilLL.y/(float) iHeight, 0.0);
	float3 vTileUR = float3((viTilLL.x+16)/(float) iWidth, (viTilLL.y+16)/(float) iHeight, dpt_ma);
	vTileUR.xy = min(vTileUR.xy,float2(1.0,1.0)).xy;
	

	// build coarse list using AABB
	for(int l=(int) t; l<(int) g_iNrVisibLights; l += NR_THREADS)
	{
		const float3 vMi = g_vBoundsBuffer[l];
		const float3 vMa = g_vBoundsBuffer[l+g_iNrVisibLights];

#ifdef ENABLE_DEPTH_TEXTURE_BACKPLANE
		if( all(vMa.xy>vTileLL.xy) && all(vMi.xyz<vTileUR.xyz))
#else
		if( all(vMa.xy>vTileLL.xy) && all(vMi.xy<vTileUR.xy))
#endif
		{
			unsigned int uInc = 1;
			unsigned int uIndex;
			InterlockedAdd(lightOffs, uInc, uIndex);
			if(uIndex<MAX_NR_COARSE_ENTRIES) coarseList[uIndex] = l;		// add to light list
		}
	}

#if !defined(XBONE) && !defined(PLAYSTATION4)
	GroupMemoryBarrierWithGroupSync();
#endif

	int iNrCoarseLights = lightOffs<MAX_NR_COARSE_ENTRIES ? lightOffs : MAX_NR_COARSE_ENTRIES;
	
#ifdef PERFORM_SPHERICAL_INTERSECTION_TESTS
	iNrCoarseLights = SphericalIntersectionTests( t, iNrCoarseLights, float2(min(viTilLL.xy+uint2(16/2,16/2), uint2(iWidth-1, iHeight-1))) );
#endif

#ifdef ENABLE_DEPTH_TEXTURE_BACKPLANE

#ifdef LEFT_HAND_COORDINATES
	float fTileFarPlane = GetLinearDepth(dpt_ma);
#else
	float fTileFarPlane = -GetLinearDepth(dpt_ma);
#endif
	float fDenom = fTileFarPlane - g_fNearPlane;
	float fModulUserScale = fDenom>0.0 ? ((g_fFarPlane-g_fNearPlane)/fDenom) : 1.0;		// readjust to new range.
#else
	float fTileFarPlane = g_fFarPlane;
	float fModulUserScale = 1.0;
#endif


#ifdef EXACT_EDGE_TESTS
	iNrCoarseLights = CullByExactEdgeTests(t, iNrCoarseLights, viTilLL.xy, viTilUR.xy, fTileFarPlane);
#endif

// sort lights
#if !defined(XBONE) && !defined(PLAYSTATION4)
	sortLightList((int) t, iNrCoarseLights);
#endif

	//////////// cell specific code
	{
		for(int l=(int) t; l<((iNrCoarseLights+1)>>1); l += NR_THREADS)
		{
			const int l0 = coarseList[2*l+0], l1 = coarseList[min(2*l+1,iNrCoarseLights)];
			const unsigned int clustIdxMi0 = (const unsigned int) min(255,SnapToClusterIdx(GetLinearDepth(g_vBoundsBuffer[l0].z), fModulUserScale));
			const unsigned int clustIdxMa0 = (const unsigned int) min(255,SnapToClusterIdx(GetLinearDepth(g_vBoundsBuffer[l0+g_iNrVisibLights].z), fModulUserScale));
			const unsigned int clustIdxMi1 = (const unsigned int) min(255,SnapToClusterIdx(GetLinearDepth(g_vBoundsBuffer[l1].z), fModulUserScale));
			const unsigned int clustIdxMa1 = (const unsigned int) min(255,SnapToClusterIdx(GetLinearDepth(g_vBoundsBuffer[l1+g_iNrVisibLights].z), fModulUserScale));

			clusterIdxs[l] = (clustIdxMa1<<24) | (clustIdxMi1<<16) | (clustIdxMa0<<8) | (clustIdxMi0<<0);
		}
	}

#if !defined(XBONE) && !defined(PLAYSTATION4)
	GroupMemoryBarrierWithGroupSync();
#endif

	int nrClusters = (1<<g_iLog2NumClusters);



	//////////////////////////////////////////////////////////
	
	uint start = 0;
	int i=(int) t;
	int iSpaceAvail = 0;
	int iSum = 0;
	if(i<nrClusters)
	{
		for(int l=0; l<iNrCoarseLights; l++)
		{
			iSum += (CheckIntersectionBasic(l, i) ? 1 : 0);
		}

		iSpaceAvail = min(iSum,MAX_NR_COARSE_ENTRIES);							// combined storage for both direct lights and reflection
		InterlockedAdd(g_LayeredSingleIdxBuffer[0], iSpaceAvail, start);		// alloc list memory
	}

	int modelListCount[NR_LIGHT_MODELS]={0,0};		// direct light count and reflection lights
	uint offs = start;
	for(int ll=0; ll<iNrCoarseLights; ll+=4)
	{
		int p = i>>2;
		int m = i&3;
		if(i<24) lightPlanes[6*m+p] = FetchPlane(min(iNrCoarseLights-1,ll+m), p);
#if !defined(XBONE) && !defined(PLAYSTATION4)
		GroupMemoryBarrierWithGroupSync();
#endif

		for(int l=ll; l<min(iNrCoarseLights,(ll+4)); l++)
		{
			if(offs<(start+iSpaceAvail) && i<nrClusters && CheckIntersection(l, i, viTilLL.xy, viTilUR.xy, fModulUserScale) )
			{
				uint lightModel = g_vLightData[ coarseList[l] ].uLightModel;
				++modelListCount[ lightModel==REFLECTION_LIGHT ? 1 : 0];
				g_vLayeredLightList[offs++] = coarseList[l];			// reflection lights will be last since we sorted
			}
		}

#if !defined(XBONE) && !defined(PLAYSTATION4)
		GroupMemoryBarrierWithGroupSync();
#endif
	}

	uint localOffs=0;
	offs = i*nrTilesX*nrTilesY + tileIDX.y*nrTilesX + tileIDX.x;
	for(int m=0; m<NR_LIGHT_MODELS; m++)
	{
		int numLights = min(modelListCount[m],31);		// only allow 5 bits
		if(i<nrClusters)
		{
			g_LayeredOffset[offs] = (start+localOffs) | (((uint) numLights)<<27);
			offs += (nrClusters*nrTilesX*nrTilesY);
			localOffs += modelListCount[m];		// use unclamped count for localOffs
		}
	}

#ifdef ENABLE_DEPTH_TEXTURE_BACKPLANE
	g_fModulUserscale[tileIDX.y*nrTilesX + tileIDX.x] = fModulUserScale;
#endif
}


// NOTE! returns 1 when value_in==0
unsigned int LimitPow2AndClamp(unsigned int value_in, unsigned int maxValue)
{
	unsigned int value = 1;
	
	while(value<value_in && (value<<1)<=maxValue)
		value<<=1;

	return value;
}


void sortLightList(int localThreadID, int length)
{
	// closest pow2 integer greater than or equal to length
	const int N = (const int) LimitPow2AndClamp((unsigned int) length, MAX_NR_COARSE_ENTRIES);			// N is 1 when length is zero but will still not enter first for-loop

	// bitonic sort can only handle arrays with a power of two length. Fill remaining entries with greater than possible index.
	for(int t=length+localThreadID; t<N; t+=NR_THREADS) { coarseList[t]=MAX_NR_COARSE_ENTRIES; }		// impossible index
	GroupMemoryBarrierWithGroupSync();

	for(int k=2; k<=N; k=2*k)
	{
		for(int j=k>>1; j>0; j=j>>1)
		{
			for(int i=localThreadID; i<N; i+=NR_THREADS)
			{
				int ixj=i^j;
				if((ixj)>i)
				{
					const unsigned int Avalue = coarseList[i];
					const unsigned int Bvalue = coarseList[ixj];

					const bool mustSwap = ((i&k)!=0^(Avalue>Bvalue)) && Avalue!=Bvalue;
					if(mustSwap)
					{
						coarseList[i]=Bvalue;
						coarseList[ixj]=Avalue;
					}
				}
			}

			GroupMemoryBarrierWithGroupSync();
		}
	}
}



float4 GetPlaneEq(const float3 vBoxX, const float3 vBoxY, const float3 vBoxZ, const float3 vCen, const float2 vScaleXY, const int sideIndex)
{
	const int iAbsSide = (sideIndex == 0 || sideIndex == 1) ? 0 : ((sideIndex == 2 || sideIndex == 3) ? 1 : 2);
	const float fS = (sideIndex & 1) != 0 ? 1 : (-1);

	float3 vA = fS*(iAbsSide == 0 ? vBoxX : (iAbsSide == 1 ? (-vBoxY) : vBoxZ));
	float3 vB = fS*(iAbsSide == 0 ? (-vBoxY) : (iAbsSide == 1 ? (-vBoxX) : (-vBoxY)));
	float3 vC = iAbsSide == 0 ? vBoxZ : (iAbsSide == 1 ? vBoxZ : (-vBoxX));

	bool bIsTopQuad = iAbsSide == 2 && (sideIndex & 1) != 0;		// in this case all 4 verts get scaled.
	bool bIsSideQuad = (iAbsSide == 0 || iAbsSide == 1);		// if side quad only two verts get scaled (impacts q1 and q2)

	if (bIsTopQuad) { vB *= vScaleXY.y; vC *= vScaleXY.x; }

	float3 vA2 = vA;
	float3 vB2 = vB;

	if (bIsSideQuad) { vA2 *= (iAbsSide == 0 ? vScaleXY.x : vScaleXY.y); vB2 *= (iAbsSide == 0 ? vScaleXY.y : vScaleXY.x); }

	float3 p0 = vCen + (vA + vB - vC);		// vCen + vA is center of face when vScaleXY is 1.0
	float3 vN = cross( vB2, 0.5*(vA-vA2) - vC );

#ifdef LEFT_HAND_COORDINATES
	vN = -vN;
#endif

	return float4(vN, -dot(vN,p0));
}


float4 FetchPlane(int l, int p)
{
	SFiniteLightBound lgtDat = g_data[coarseList[l]];
	
	const float3 vBoxX = lgtDat.vBoxAxisX.xyz;
	const float3 vBoxY = lgtDat.vBoxAxisY.xyz;
	const float3 vBoxZ = -lgtDat.vBoxAxisZ.xyz;           // flip an axis to make it right handed since Determinant(worldToView)<0
	const float3 vCen = lgtDat.vCen.xyz;
	const float fRadius = lgtDat.fRadius;
	const float2 vScaleXY = lgtDat.vScaleXY;

	return GetPlaneEq(vBoxX, vBoxY, vBoxZ, vCen, vScaleXY, p);
}





#ifdef PERFORM_SPHERICAL_INTERSECTION_TESTS
int SphericalIntersectionTests(uint threadID, int iNrCoarseLights, float2 screenCoordinate)
{
#ifdef LEFT_HAND_COORDINATES
	float3 V = GetViewPosFromLinDepth( screenCoordinate, 1.0);
#else
	float3 V = GetViewPosFromLinDepth( screenCoordinate, -1.0);
#endif

	float onePixDiagDist = GetOnePixDiagWorldDistAtDepthOne();
	float worldDistAtDepthOne = 8*onePixDiagDist;		// scale by half a tile
	

	int iNrVisib = 0;
	for(int l=threadID; l<iNrCoarseLights; l+=NR_THREADS)
	{
		SFiniteLightBound lgtDat = g_data[coarseList[l]];
	
		const float3 vCen = lgtDat.vCen.xyz;
		float fRad = lgtDat.fRadius;

#if 1
		float3 maxZdir = float3(-vCen.z*vCen.x, -vCen.z*vCen.y, vCen.x*vCen.x + vCen.y*vCen.y);		// cross(vCen,cross(Zaxis,vCen))
		float len = length(maxZdir);
		float scalarProj = len>0.0001 ? (maxZdir.z/len) : len;	// since len>=(maxZdir.z/len) we can use len as an approximate value when len<=epsilon
		float fOffs = scalarProj*fRad;
#else
		float fOffs = fRad;		// more false positives due to larger radius but works too
#endif

#ifdef LEFT_HAND_COORDINATES
		fRad = fRad + (vCen.z+fOffs)*worldDistAtDepthOne;
#else
		fRad = fRad + (vCen.z-fOffs)*worldDistAtDepthOne;
#endif
		
		float a = dot(V,V);
		float CdotV = dot(vCen,V);
		float c = dot(vCen,vCen) - fRad*fRad;

		float fDescDivFour = CdotV*CdotV - a*c;
		if(!(c<0 || (fDescDivFour>0 && CdotV>0)))		// if ray hit bounding sphere
			coarseList[l]=0xffffffff;
	}

#if !defined(XBONE) && !defined(PLAYSTATION4)
		GroupMemoryBarrierWithGroupSync();
#endif

	// to greedy to double buffer coarseList lds on this so serializing removal of gaps.
	if(threadID==0)
	{
		int offs = 0;
		for(int l=0; l<iNrCoarseLights; l++)
		{	if(coarseList[l]!=0xffffffff) coarseList[offs++] = coarseList[l]; }
		lightOffsSph = offs;
	}

#if !defined(XBONE) && !defined(PLAYSTATION4)
	GroupMemoryBarrierWithGroupSync();
#endif

	return lightOffsSph;
}
#endif







#ifdef EXACT_EDGE_TESTS
float3 GetHullVertex(const float3 vBoxX, const float3 vBoxY, const float3 vBoxZ, const float3 vCen, const float2 vScaleXY, const int p)
{
	const bool bIsTopVertex = (p&4)!=0;
	float3 vScales = float3( ((p&1)!=0 ? 1.0f : (-1.0f))*(bIsTopVertex ? vScaleXY.x : 1.0), ((p&2)!=0 ? 1.0f : (-1.0f))*(bIsTopVertex ? vScaleXY.y : 1.0), (p&4)!=0 ? 1.0f : (-1.0f) );
	return (vScales.x*vBoxX + vScales.y*vBoxY + vScales.z*vBoxZ) + vCen;
}

void GetHullEdge(out int idx0, out int idx_twin, out float3 vP0, out float3 vE0, const int e0, const float3 vBoxX, const float3 vBoxY, const float3 vBoxZ, const float3 vCen, const float2 vScaleXY)
{
	int iAxis = e0>>2;
	int iSwizzle = e0&0x3;
	bool bIsSwizzleOneOrTwo = ((iSwizzle-1)&0x2)==0;

	const int i0 = iAxis==0 ? (2*iSwizzle+0) : ( iAxis==1 ? (iSwizzle+(iSwizzle&2)) : iSwizzle);
	const int i1 = i0 + (1<<iAxis);
	const bool bSwap = iAxis==0 ? (!bIsSwizzleOneOrTwo) : (iAxis==1 ? false : bIsSwizzleOneOrTwo);
	
	idx0 = bSwap ? i1 : i0;
	idx_twin = bSwap ? i0 : i1;
	float3 p0 = GetHullVertex(vBoxX, vBoxY, vBoxZ, vCen, vScaleXY, idx0);
	float3 p1 = GetHullVertex(vBoxX, vBoxY, vBoxZ, vCen, vScaleXY, idx_twin);

	vP0 = p0;
	vE0 = p1-p0;
}

float3 GetTileVertex(uint2 viTilLL, uint2 viTilUR, int i, float fTileFarPlane)
{
	float x = (i&1)==0 ? viTilLL.x : viTilUR.x;
	float y = (i&2)==0 ? viTilLL.y : viTilUR.y;
	float z = (i&4)==0 ? g_fNearPlane : fTileFarPlane;
#ifndef LEFT_HAND_COORDINATES
	z = -z;
#endif
	return GetViewPosFromLinDepth( float2(x, y), z);
}

void GetFrustEdge(out float3 vP0, out float3 vE0, const int e0, uint2 viTilLL, uint2 viTilUR, float fTileFarPlane)
{
	int iSection = e0>>2;		// section 0 is side edges, section 1 is near edges and section 2 is far edges
	int iSwizzle = e0&0x3;

	int i=iSwizzle + (2*(iSection&0x2));	// offset by 4 at section 2
	vP0 = GetTileVertex(uint2(viTilLL.x, viTilUR.y), uint2(viTilUR.x, viTilLL.y), i, fTileFarPlane);
	vE0 = iSection==0 ? vP0 : (((iSwizzle&0x2)==0 ? 1.0f : (-1.0f))*((iSwizzle&0x1)==(iSwizzle>>1) ? Vec3(1,0,0) : Vec3(0,1,0)));
}

int CullByExactEdgeTests(uint threadID, int iNrCoarseLights, uint2 viTilLL, uint2 viTilUR, float fTileFarPlane)
{
	if(threadID==0) lightOffs2 = 0;

	const bool bOnlyNeedFrustumSideEdges = true;
	const int nrFrustEdges = bOnlyNeedFrustumSideEdges ? 4 : 8;	// max 8 since we never need to test 4 far edges of frustum since they are identical vectors to near edges and plane is placed at vP0 on light hull.

	const int totNrEdgePairs = 12*nrFrustEdges;
	for(int l=0; l<iNrCoarseLights; l++)
	{
		if(threadID==0) ldsIsLightInvisible=0;

#if !defined(XBONE) && !defined(PLAYSTATION4)
		GroupMemoryBarrierWithGroupSync();
#endif
		const int idxCoarse = coarseList[l];
		[branch]if(g_vLightData[idxCoarse].uLightType!=SPHERE_LIGHT)		// don't bother doing edge tests for sphere lights since these have camera aligned bboxes.
		{
			SFiniteLightBound lgtDat = g_data[idxCoarse];
	
			const float3 vBoxX = lgtDat.vBoxAxisX.xyz;
			const float3 vBoxY = lgtDat.vBoxAxisY.xyz;
			const float3 vBoxZ = -lgtDat.vBoxAxisZ.xyz;           // flip an axis to make it right handed since Determinant(worldToView)<0
			const float3 vCen = lgtDat.vCen.xyz;
			const float2 vScaleXY = lgtDat.vScaleXY;

			for(int i=threadID; i<totNrEdgePairs; i+=NR_THREADS)
			{
				int e0 = (int) (((uint)i)/((uint) nrFrustEdges)); // should become a shift right
				int e1 = i - e0*nrFrustEdges;

				int idx_cur=0, idx_twin=0;
				float3 vP0, vE0;
				GetHullEdge(idx_cur, idx_twin, vP0, vE0, e0, vBoxX, vBoxY, vBoxZ, vCen, vScaleXY);
				
			
				float3 vP1, vE1;
				GetFrustEdge(vP1, vE1, e1, viTilLL, viTilUR, fTileFarPlane);
				
				// potential separation plane
				float3 vN = cross(vE0, vE1);
			
				int positive=0, negative=0;
				for(int k=1; k<8; k++)		// only need to test 7 verts (technically just 6).
				{
					int j = (idx_cur+k)&0x7;
					float3 vPh = GetHullVertex(vBoxX, vBoxY, vBoxZ, vCen, vScaleXY, j);
					float fSignDist = idx_twin==j ? 0.0 : dot(vN, vPh-vP0);
					if(fSignDist>0) ++positive; else if(fSignDist<0) ++negative;
				}
				int resh = (positive>0 && negative>0) ? 0 : (positive>0 ? 1 : (negative>0 ? (-1) : 0));

				positive=0; negative=0;
				for(int j=0; j<8; j++)
				{
					float3 vPf = GetTileVertex(viTilLL, viTilUR, j, fTileFarPlane);
					float fSignDist = dot(vN, vPf-vP0);
					if(fSignDist>0) ++positive; else if(fSignDist<0) ++negative;
				}
				int resf = (positive>0 && negative>0) ? 0 : (positive>0 ? 1 : (negative>0 ? (-1) : 0));

				bool bFoundSepPlane = (resh*resf)<0;

				if(bFoundSepPlane) InterlockedOr(ldsIsLightInvisible, 1);
			}
		}
#if !defined(XBONE) && !defined(PLAYSTATION4)
		GroupMemoryBarrierWithGroupSync();
#endif
		if(threadID==0 && ldsIsLightInvisible==0)
		{
			coarseList[lightOffs2++] = coarseList[l];
		}
	}
#if !defined(XBONE) && !defined(PLAYSTATION4)
		GroupMemoryBarrierWithGroupSync();
#endif
	return lightOffs2;
}
#endif



[numthreads(1, 1, 1)]
void ClearAtomic(uint threadID : SV_GroupIndex, uint3 u3GroupID : SV_GroupID)
{
	g_LayeredSingleIdxBuffer[0]=0;
}