#pragma kernel PerlinWorleyNoiseEvaluator
#pragma kernel WorleyNoiseEvaluator
#pragma kernel PerlinNoiseEvaluator

#pragma only_renderers d3d11 playstation xboxone xboxseries vulkan metal switch

#include "Packages/com.unity.render-pipelines.core/ShaderLibrary/Common.hlsl"
#include "Packages/com.unity.render-pipelines.high-definition/Runtime/ShaderLibrary/ShaderVariables.hlsl"
#include "Packages/com.unity.render-pipelines.high-definition/Runtime/Lighting/VolumetricClouds/WorleyUtilities.hlsl"

// Input and output data
int _NumLayers;
int _Layer;
RWTexture2D<float4> _WorleyEvaluationOutput;

// Due to a limitation, we cannot do the whole thing in a single NxNxN dispatch so we have to do N dispatches of NxN.
[numthreads(8, 8, 1)]
void PerlinWorleyNoiseEvaluator(uint3 dispatchThreadID : SV_DispatchThreadID)
{
    // Convert the dispatch coordinates to to the generation space [0,1]x[0,1]x[0,1]
    float3 position = float3(((float2)dispatchThreadID.xy + 0.5) / _NumLayers, ((float)_Layer + 0.5) / _NumLayers);

    // Generate the perlin FBM
    const float baseFrequence = 4.0;
    float perlinFBM = lerp(1.0, EvaluatePerlinFractalBrownianMotion(position, baseFrequence, 4.0), 0.5);
    perlinFBM = abs(perlinFBM * 2.0 - 1.0); // billowy perlin noise

    // Generate all the worley noises that are required
    const float3 positionFrequence = position * baseFrequence;
    float worley1 = 1.0 - WorleyNoise(positionFrequence, baseFrequence * 1.0f);
    float worley2 = 1.0 - WorleyNoise(positionFrequence * 2.0f, baseFrequence * 2.0f);
    float worley4 = 1.0 - WorleyNoise(positionFrequence * 4.0f, baseFrequence * 4.0f);

    // Compute the worley fractal brownian motion
    float worleyFBM = worley1 * 0.625 + worley2 * 0.25 + worley4 * 0.125;

    // Combine both signals
    float result = remap(perlinFBM, 0.0, 1.0, worleyFBM, 1.0); // Perlin-Worley

    // Output to the render texture.
    _WorleyEvaluationOutput[dispatchThreadID.xy] = float4(result, 0.0, 0.0, 1.0);
}

// Due to a limitation, we cannot do the whole thing in a single NxNxN dispatch so we have to do N dispatches of NxN.
[numthreads(8, 8, 1)]
void WorleyNoiseEvaluator(uint3 dispatchThreadID : SV_DispatchThreadID)
{
    // Convert the dispatch coordinates to to the generation space [0,1]x[0,1]x[0,1]
    float3 position = float3(((float2)dispatchThreadID.xy + 0.5) / _NumLayers, ((float)_Layer + 0.5) / _NumLayers);

    // Generate all the worley noises that are required
    const float baseFrequence = 2.0;
    const float3 positionFrequence = position * baseFrequence;
    float worley2 = 1.0 - WorleyNoise(positionFrequence * 2.0f, baseFrequence * 2.0f);
    float worley4 = 1.0 - WorleyNoise(positionFrequence * 4.0f, baseFrequence * 4.0f);
    float worley8 = 1.0 - WorleyNoise(positionFrequence * 8.0f, baseFrequence * 8.0f);

    // Compute the worley fractal brownian motion
    float worleyFBM = worley2 * 0.625 + worley4 * 0.25 + worley8 * 0.125;

    // Output to the render texture.
    _WorleyEvaluationOutput[dispatchThreadID.xy] = float4(worleyFBM, 0.0, 0.0, 1.0);
}

// Due to a limitation, we cannot do the whole thing in a single NxNxN dispatch so we have to do N dispatches of NxN.
[numthreads(8, 8, 1)]
void PerlinNoiseEvaluator(uint3 dispatchThreadID : SV_DispatchThreadID)
{
    // Convert the dispatch coordinates to to the generation space [0,1]x[0,1]x[0,1]
    float3 position = float3(((float2)dispatchThreadID.xy + 0.5) / _NumLayers, ((float)_Layer + 0.5) / _NumLayers);
    // Compute the worley fractal brownian motion
    const float baseFrequence = 4.0;
    float perlinFBM = lerp(1.0, EvaluatePerlinFractalBrownianMotion(position, baseFrequence, 8.0), 0.5);

    // Output to the render texture.
    _WorleyEvaluationOutput[dispatchThreadID.xy] = float4(perlinFBM, 0.0, 0.0, 1.0);
}