<@include gpu/Config.slh@>
<$VERSION_HEADER$>
//  Generated on <$_SCRIBE_DATE$>
//
//  ambient_occlusion.frag
//  fragment shader
//
//  Created by Niraj Venkat on 7/15/15.
//  Copyright 2015 High Fidelity, Inc.
//
//  Distributed under the Apache License, Version 2.0.
//  See the accompanying file LICENSE or http://www.apache.org/licenses/LICENSE-2.0.html
//

<@include DeferredBufferWrite.slh@>

<@include gpu/Transform.slh@>

<$declareStandardTransform()$>

// Based on NVidia HBAO implementation in D3D11
// http://www.nvidia.co.uk/object/siggraph-2008-HBAO.html

in vec2 varTexcoord;

uniform sampler2D depthTexture;
uniform sampler2D normalTexture;

uniform float g_scale;
uniform float g_bias;
uniform float g_sample_rad;
uniform float g_intensity;
uniform float bufferWidth;
uniform float bufferHeight;

const float PI = 3.14159265;

const vec2 FocalLen = vec2(1.0, 1.0);

const vec2 LinMAD = vec2(0.1-10.0, 0.1+10.0) / (2.0*0.1*10.0);

const vec2 AORes = vec2(1024.0, 768.0);
const vec2 InvAORes = vec2(1.0/1024.0, 1.0/768.0);
const vec2 NoiseScale = vec2(1024.0, 768.0) / 4.0;

const float AOStrength = 1.9;
const float R = 0.3;
const float R2 = 0.3*0.3;
const float NegInvR2 = - 1.0 / (0.3*0.3);
// can't use tan to initialize a const value
const float TanBias = 0.57735027; // tan(30.0 * PI / 180.0); 
const float MaxRadiusPixels = 50.0;

const int NumDirections = 6;
const int NumSamples = 4;

out vec4 outFragColor;

float ViewSpaceZFromDepth(float d){
	// [0,1] -> [-1,1] clip space
	d = d * 2.0 - 1.0;

	// Get view space Z
	return -1.0 / (LinMAD.x * d + LinMAD.y);
}

vec3 UVToViewSpace(vec2 uv, float z){
	//uv = UVToViewA * uv + UVToViewB;
	return vec3(uv * z, z);
}

vec3 GetViewPos(vec2 uv){
	float z = ViewSpaceZFromDepth(texture(depthTexture, uv).r);
	return UVToViewSpace(uv, z);
}

vec3 GetViewPosPoint(ivec2 uv){
	vec2 coord = vec2(gl_FragCoord.xy) + uv;
	//float z = texelFetch(texture0, coord, 0).r;
    float z = texture(depthTexture, uv).r;
	return UVToViewSpace(uv, z);
}

float TanToSin(float x){
	return x * inversesqrt(x*x + 1.0);
}

float InvLength(vec2 V){
	return inversesqrt(dot(V,V));
}

float Tangent(vec3 V){
	return V.z * InvLength(V.xy);
}

float BiasedTangent(vec3 V){
	return V.z * InvLength(V.xy) + TanBias;
}

float Tangent(vec3 P, vec3 S){
    return -(P.z - S.z) * InvLength(S.xy - P.xy);
}

float Length2(vec3 V){
	return dot(V,V);
}

vec3 MinDiff(vec3 P, vec3 Pr, vec3 Pl){
    vec3 V1 = Pr - P;
    vec3 V2 = P - Pl;
    return (Length2(V1) < Length2(V2)) ? V1 : V2;
}

vec2 SnapUVOffset(vec2 uv){
    return round(uv * AORes) * InvAORes;
}

float Falloff(float d2){
	return d2 * NegInvR2 + 1.0f;
}

float HorizonOcclusion(	vec2 deltaUV, vec3 P, vec3 dPdu, vec3 dPdv, float randstep, float numSamples){
	float ao = 0;

	// Offset the first coord with some noise
	vec2 uv = varTexcoord + SnapUVOffset(randstep*deltaUV);
	deltaUV = SnapUVOffset( deltaUV );

	// Calculate the tangent vector
	vec3 T = deltaUV.x * dPdu + deltaUV.y * dPdv;

	// Get the angle of the tangent vector from the viewspace axis
	float tanH = BiasedTangent(T);
	float sinH = TanToSin(tanH);

	float tanS;
	float d2;
	vec3 S;

	// Sample to find the maximum angle
	for(float s = 1; s <= numSamples; ++s){
		uv += deltaUV;
		S = GetViewPos(uv);
		tanS = Tangent(P, S);
		d2 = Length2(S - P);

		// Is the sample within the radius and the angle greater?
		if(d2 < R2 && tanS > tanH)
		{
			float sinS = TanToSin(tanS);
			// Apply falloff based on the distance
			ao += Falloff(d2) * (sinS - sinH);

			tanH = tanS;
			sinH = sinS;
		}
	}
	return ao;
}

vec2 RotateDirections(vec2 Dir, vec2 CosSin){
    return vec2(Dir.x*CosSin.x - Dir.y*CosSin.y, Dir.x*CosSin.y + Dir.y*CosSin.x);
}

void ComputeSteps(inout vec2 stepSizeUv, inout float numSteps, float rayRadiusPix, float rand){
    // Avoid oversampling if numSteps is greater than the kernel radius in pixels
    numSteps = min(NumSamples, rayRadiusPix);

    // Divide by Ns+1 so that the farthest samples are not fully attenuated
    float stepSizePix = rayRadiusPix / (numSteps + 1);

    // Clamp numSteps if it is greater than the max kernel footprint
    float maxNumSteps = MaxRadiusPixels / stepSizePix;
    if (maxNumSteps < numSteps)
    {
        // Use dithering to avoid AO discontinuities
        numSteps = floor(maxNumSteps + rand);
        numSteps = max(numSteps, 1);
        stepSizePix = MaxRadiusPixels / numSteps;
    }

    // Step size in uv space
    stepSizeUv = stepSizePix * InvAORes;
}

float getRandom(vec2 uv){
    return fract(sin(dot(uv.xy ,vec2(12.9898,78.233))) * 43758.5453);
}

void main(void){
	float numDirections = NumDirections;

	vec3 P, Pr, Pl, Pt, Pb;
	P = GetViewPos(varTexcoord);

	// Sample neighboring pixels
    Pr = GetViewPos(varTexcoord + vec2( InvAORes.x, 0));
    Pl = GetViewPos(varTexcoord + vec2(-InvAORes.x, 0));
    Pt = GetViewPos(varTexcoord + vec2( 0, InvAORes.y));
    Pb = GetViewPos(varTexcoord + vec2( 0,-InvAORes.y));

    // Calculate tangent basis vectors using the minimum difference
    vec3 dPdu = MinDiff(P, Pr, Pl);
    vec3 dPdv = MinDiff(P, Pt, Pb) * (AORes.y * InvAORes.x);

    // Get the random samples from the noise function
	vec3 random = vec3(getRandom(varTexcoord.xy), getRandom(varTexcoord.yx), getRandom(varTexcoord.xx));

	// Calculate the projected size of the hemisphere
    vec2 rayRadiusUV = 0.5 * R * FocalLen / -P.z;
    float rayRadiusPix = rayRadiusUV.x * AORes.x;

    float ao = 1.0;

    // Make sure the radius of the evaluated hemisphere is more than a pixel
    if(rayRadiusPix > 1.0){
    	ao = 0.0;
    	float numSteps;
    	vec2 stepSizeUV;

    	// Compute the number of steps
    	ComputeSteps(stepSizeUV, numSteps, rayRadiusPix, random.z);

		float alpha = 2.0 * PI / numDirections;

		// Calculate the horizon occlusion of each direction
		for(float d = 0; d < numDirections; ++d){
			float theta = alpha * d;

			// Apply noise to the direction
			vec2 dir = RotateDirections(vec2(cos(theta), sin(theta)), random.xy);
			vec2 deltaUV = dir * stepSizeUV;

			// Sample the pixels along the direction
			ao += HorizonOcclusion(	deltaUV,
									P,
									dPdu,
									dPdv,
									random.z,
									numSteps);
		}

		// Average the results and produce the final AO
		ao = 1.0 - ao / numDirections * AOStrength;
	}

    outFragColor = vec4(vec3(ao), 1.0);
}