// Copyright 1998-2017 Epic Games, Inc. All Rights Reserved.

/*=============================================================================
	PostProcessSelectionOutline.usf: Post processing outline effect.
=============================================================================*/

#include "Common.usf"
#include "PostProcessCommon.usf"

#if MSAA_SAMPLE_COUNT > 1
Texture2DMS<float, MSAA_SAMPLE_COUNT> PostprocessInput1MS;
Texture2DMS<uint2, MSAA_SAMPLE_COUNT> EditorPrimitivesStencil;
#else
// note: opengl compiler doesn't like Texture2D<float> so we don't do it
Texture2D PostprocessInput1MS;
Texture2D<uint2> EditorPrimitivesStencil;
#endif

// RGB: selection color, A:SelectionHighlightIntensity
float4 OutlineColor;
// RGB:subdued selection color (for inactive selection), A:unused
float4 SubduedOutlineColor;

float BSPSelectionIntensity;

// r. value of r.Editor.OpaqueGizmo .g:r.Editor.MovingPattern ba: unused
float4 EditorRenderParams;

struct FPixelInfo
{
	// if there is an selected object
	bool bObjectMask;
	// 
	int ObjectId;
	// if the current pixel is not occluded by some other object in front of it
	float fVisible;
	// hard version of fVisible
	bool bVisible;
};

// @param DeviceZPlane plane in screenspace .x: ddx(DeviceZ), y: ddy(DeviceZ) z:DeviceZ 
float ReconstructDeviceZ(float3 DeviceZPlane, float2 PixelOffset)
{
	return dot(DeviceZPlane, float3(PixelOffset.xy, 1));
}


// @param DeviceZPlane plane in screenspace .x: ddx(DeviceZ), y: ddy(DeviceZ) z:DeviceZ
FPixelInfo GetPixelInfo(int2 PixelPos, int SampleID, int OffsetX, int OffsetY, float3 DeviceZPlane, float2 DeviceZMinMax)
{
	FPixelInfo Ret;

	PixelPos += int2(OffsetX, OffsetY);

	// slightly more flicking artifacts on the silhouette
//	float2 ReconstructionOffset = int2(OffsetX, OffsetY) + 0.5f - View.TemporalAAParams.zw;
	// more stable on the silhouette
	float2 ReconstructionOffset = 0.5f - View.TemporalAAParams.zw;

#if MSAA_SAMPLE_COUNT > 1
	float DeviceZ = PostprocessInput1MS.Load(PixelPos, SampleID).r;
	int Stencil = EditorPrimitivesStencil.Load(PixelPos, SampleID) STENCIL_COMPONENT_SWIZZLE;

#if !COMPILER_GLSL && !PS4_PROFILE && !COMPILER_METAL && !COMPILER_HLSLCC
	// not yet supported on OpenGL, slightly better quality
	ReconstructionOffset += PostprocessInput1MS.GetSamplePosition(SampleID);
#endif

#else
	float DeviceZ = PostprocessInput1MS.Load(int3(PixelPos, 0)).r;
	int Stencil = EditorPrimitivesStencil.Load(int3(PixelPos, 0)) STENCIL_COMPONENT_SWIZZLE;
#endif

	float SceneDeviceZ = ReconstructDeviceZ(DeviceZPlane, ReconstructionOffset);

	// clamp SceneDeviceZ in (DeviceZMinMax.x .. DeviceZMinMax.z)
	// this avoids flicking artifacts on the silhouette by limiting the depth reconstruction error
	SceneDeviceZ = max(SceneDeviceZ, DeviceZMinMax.x);
	SceneDeviceZ = min(SceneDeviceZ, DeviceZMinMax.y);

	// test against far plane
	Ret.bObjectMask = DeviceZ != 0.0f;

	// outline even between multiple selected objects (best usability)
	Ret.ObjectId = Stencil;

#if (COMPILER_GLSL == 1 && FEATURE_LEVEL < FEATURE_LEVEL_SM4) || (COMPILER_METAL && FEATURE_LEVEL < FEATURE_LEVEL_SM4)
	// Stencil read in opengl is not supported on older versions
	Ret.ObjectId = Ret.bObjectMask ? 2 : 0;
#endif

	// Soft Bias with DeviceZ for best quality (larger bias than usual because SceneDeviceZ is only approximated)
	const float DeviceDepthFade = 0.00005f;

	// 2 to always bias over the current plane
	Ret.fVisible = saturate(2.0f - (SceneDeviceZ - DeviceZ) / DeviceDepthFade);

	Ret.bVisible = Ret.fVisible >= 0.5f;

	return Ret;
}

bool BoolXor(bool bA, bool bB)
{
	int a = (int)bA;
	int b = (int)bB;

	return (a ^ b) != 0;
}

// computes min and max at once
void MinMax(inout int2 Var, int Value)
{
	Var.x = min(Var.x, Value);
	Var.y = max(Var.y, Value);
}

float4 PerSample(int2 PixelPos, int SampleID, FPixelInfo CenterPixelInfo, float3 DeviceZPlane, float2 DeviceZMinMax)
{
	float4 Ret = 0;

	// [0]:center, [1..4]:borders
	FPixelInfo PixelInfo[5];
	
	PixelInfo[0] = CenterPixelInfo;

	// diagonal cross is thicker than vertical/horizontal cross
	PixelInfo[1] = GetPixelInfo(PixelPos, SampleID,  1,  1, DeviceZPlane, DeviceZMinMax);
	PixelInfo[2] = GetPixelInfo(PixelPos, SampleID, -1, -1, DeviceZPlane, DeviceZMinMax);
	PixelInfo[3] = GetPixelInfo(PixelPos, SampleID,  1, -1, DeviceZPlane, DeviceZMinMax);
	PixelInfo[4] = GetPixelInfo(PixelPos, SampleID, -1,  1, DeviceZPlane, DeviceZMinMax);

	// with (.x != .y) we can detect a border around and between each object
	int2 BorderMinMax = int2(255,0);
	{
		MinMax(BorderMinMax, PixelInfo[1].ObjectId);
		MinMax(BorderMinMax, PixelInfo[2].ObjectId);
		MinMax(BorderMinMax, PixelInfo[3].ObjectId);
		MinMax(BorderMinMax, PixelInfo[4].ObjectId);
	}

	int2 VisibleBorderMinMax = int2(255,0);
	{
		FLATTEN	if(PixelInfo[1].bVisible) MinMax(VisibleBorderMinMax, PixelInfo[1].ObjectId);
		FLATTEN	if(PixelInfo[2].bVisible) MinMax(VisibleBorderMinMax, PixelInfo[2].ObjectId);
		FLATTEN	if(PixelInfo[3].bVisible) MinMax(VisibleBorderMinMax, PixelInfo[3].ObjectId);
		FLATTEN	if(PixelInfo[4].bVisible) MinMax(VisibleBorderMinMax, PixelInfo[4].ObjectId);
	}

	// this border around the object
	bool bVisibleBorder = VisibleBorderMinMax.y != 0;
	bool bBorder = BorderMinMax.x != BorderMinMax.y;

	bool bInnerBorder = BorderMinMax.y == 4;
	
	// moving diagonal lines
//	float PatternMask = saturate(sin((PixelPos.x + PixelPos.y) * 0.8f + 8.0f * View.RealTime * EditorRenderParams.g));
	float PatternMask = ((PixelPos.x/2 + PixelPos.y/2) % 2) * 0.6f;
	// non moving high frequency pattern
//	float PatternMask = ((PixelPos.x + PixelPos.y) % 2) * 0.7f;

	// the contants express the two opacity values we see when the primitive is hidden
	float LowContrastPatternMask = lerp(0.2, 1.0f, PatternMask);
	
	float3  SelectionColor;
	FLATTEN if(VisibleBorderMinMax.x >= 128 )
	{
		SelectionColor = SubduedOutlineColor.rgb;
	}
	else
	{
		SelectionColor = OutlineColor.rgb;
	}
	FLATTEN if(bBorder)
	{
		FLATTEN if(bVisibleBorder)
		{
			// unoccluded border
			Ret = float4(SelectionColor, 1);
		}
		else
		{
			// occluded border
			Ret = lerp(float4(0, 0, 0, 0), float4(SelectionColor, 1), LowContrastPatternMask);
		}
	}

	FLATTEN if(bInnerBorder)
	{
		// even occluded object is filled
		// occluded object is rendered differently(flickers with TemporalAA)
		float VisibleMask = lerp(PatternMask, 1.0f, PixelInfo[0].fVisible);

		// occluded parts are more tranparent
//		Ret = lerp(Ret, float4(SelectionColor.rgb, 1), OutlineColor.a * VisibleMask);
		// darken occluded parts
		Ret = lerp(Ret, float4(SelectionColor * VisibleMask, 1), OutlineColor.a );
	}

	// highlight inner part of the object
	if(PixelInfo[0].bObjectMask)
	{
		float VisibleMask = lerp(PatternMask, 1.0f, PixelInfo[0].fVisible);
		float InnerHighlightAmount = PixelInfo[0].ObjectId == 1 ? BSPSelectionIntensity : OutlineColor.a;
		Ret = lerp(Ret, float4(SelectionColor, 1), VisibleMask * InnerHighlightAmount);
	}

	return Ret;
}

void MainPS(noperspective float4 UVAndScreenPos : TEXCOORD0, float4 SvPosition : SV_POSITION, out float4 OutColor : SV_Target0)
{
	float2 UV = UVAndScreenPos.xy;
 	int2 PixelPos = (int2)SvPosition.xy;
	
	OutColor = Texture2DSample(PostprocessInput0, PostprocessInput0Sampler, UVAndScreenPos.xy);

	// Scout outwards from the center pixel to determine if any neighboring pixels are selected
	int ScountStep = 2;
#if MSAA_SAMPLE_COUNT > 1
	int4 StencilScout = int4(
		EditorPrimitivesStencil.Load(PixelPos + int2( ScountStep,  0), 0) STENCIL_COMPONENT_SWIZZLE,
		EditorPrimitivesStencil.Load(PixelPos + int2(-ScountStep,  0), 0) STENCIL_COMPONENT_SWIZZLE,
		EditorPrimitivesStencil.Load(PixelPos + int2( 0,  ScountStep), 0) STENCIL_COMPONENT_SWIZZLE,
		EditorPrimitivesStencil.Load(PixelPos + int2( 0, -ScountStep), 0) STENCIL_COMPONENT_SWIZZLE
	);
#else
	int4 StencilScout = int4(
		EditorPrimitivesStencil.Load(int3(PixelPos + int2( ScountStep,  0), 0)) STENCIL_COMPONENT_SWIZZLE,
		EditorPrimitivesStencil.Load(int3(PixelPos + int2(-ScountStep,  0), 0)) STENCIL_COMPONENT_SWIZZLE,
		EditorPrimitivesStencil.Load(int3(PixelPos + int2( 0,  ScountStep), 0)) STENCIL_COMPONENT_SWIZZLE,
		EditorPrimitivesStencil.Load(int3(PixelPos + int2( 0, -ScountStep), 0)) STENCIL_COMPONENT_SWIZZLE
	);
#endif
	int ScoutSum = dot(saturate(StencilScout), 1);

	// If this sum is zero, none of our neighbors are selected pixels and we can skip all the heavy processing
	if (ScoutSum > 0) {
		float SceneDepth = CalcSceneDepth(UV);

		// This allows to reconstruct depth with a small pixel offset without many tetxure lookups (4xMSAA * 5 neighbors -> 20 samples)
		// It's an approximation assuming the surface is a plane
		float3 DeviceZPlane;
		DeviceZPlane.z = Texture2DSampleLevel(SceneDepthTexture, SceneDepthTextureSampler, UV, 0).r;

		float Left = Texture2DSampleLevel(SceneDepthTexture, SceneDepthTextureSampler, UV + float2(-1, 0) * View.BufferSizeAndInvSize.zw, 0).r;
		float Right= Texture2DSampleLevel(SceneDepthTexture, SceneDepthTextureSampler, UV + float2(1, 0) * View.BufferSizeAndInvSize.zw, 0).r;
		float Top  = Texture2DSampleLevel(SceneDepthTexture, SceneDepthTextureSampler, UV + float2(0, -1) * View.BufferSizeAndInvSize.zw, 0).r;
		float Bottom = Texture2DSampleLevel(SceneDepthTexture, SceneDepthTextureSampler, UV + float2(0, 1) * View.BufferSizeAndInvSize.zw, 0).r;

		float2 DeviceZMinMax;
		DeviceZMinMax.x = min(DeviceZPlane.z, min(min(Left, Right), min(Top, Bottom)));
		DeviceZMinMax.y = max(DeviceZPlane.z, max(max(Left, Right), max(Top, Bottom)));

		DeviceZPlane.x = (Right - Left) / 2;
		DeviceZPlane.y = (Bottom - Top) / 2;

		// even stronger appromiation (faster but too many artifacts)
	//	DeviceZPlane.x = ddx(DeviceZPlane.z);
	//	DeviceZPlane.y = ddy(DeviceZPlane.z);

		FPixelInfo CenterPixelInfo = GetPixelInfo(PixelPos, 0, 0,  0, DeviceZPlane, DeviceZMinMax);

		float4 Sum = 0;
#if COMPILER_GLSL && FEATURE_LEVEL >= FEATURE_LEVEL_SM4
	UNROLL
#endif
		for(int SampleID = 0; SampleID < MSAA_SAMPLE_COUNT; ++SampleID)
		{
			Sum += PerSample(PixelPos, SampleID, CenterPixelInfo, DeviceZPlane, DeviceZMinMax);
		}
		float4 Avg = Sum / MSAA_SAMPLE_COUNT;

		OutColor = Texture2DSample(PostprocessInput0, PostprocessInput0Sampler, UVAndScreenPos.xy);

		// Dest color has gamma applied so we need to remove it before applying linear color
		OutColor.rgb = pow( OutColor.rgb, 2.2f );

		OutColor.rgb = lerp(OutColor.rgb, Avg.rgb, Avg.a);

		// Re-apply gamma corrrection
		OutColor.rgb = pow( OutColor.rgb, 1/2.2f );
	}
/*
	if(MSAA_SAMPLE_COUNT == 2)
	{
		OutColor = 1;
	}
*/

//	OutColor.rgb = Avg.rgb;
//	OutColor.rgb = Avg.a;

	// SceneColor from the pass before
//	OutColor = Texture2DSample(PostprocessInput0, PostprocessInput0Sampler, UVAndScreenPos.xy);
}