// Precomputes data for IntegrateLD(). See that function for a detailed description.

#include "../../Core/ShaderLibrary/Common.hlsl"
#include "../../Core/ShaderLibrary/ImageBasedLighting.hlsl"

#ifdef SHADER_API_MOBILE
#define MAX_IBL_SAMPLE_CNT 34
#else
#define MAX_IBL_SAMPLE_CNT 89
#endif

RWTexture2D<float4> output; // [MAX_SAMPLE_CNT x UNITY_SPECCUBE_LOD_STEPS]

#pragma kernel ComputeGgxIblSampleData

[numthreads(MAX_IBL_SAMPLE_CNT, UNITY_SPECCUBE_LOD_STEPS, 1)]
void ComputeGgxIblSampleData(uint3 groupThreadId : SV_GroupThreadID)
{
    uint sampleIndex = groupThreadId.x;
    uint mipLevel    = groupThreadId.y + 1;
    uint sampleCount = GetIBLRuntimeFilterSampleCount(mipLevel);

    if (sampleIndex >= sampleCount)
    {
        output[groupThreadId.xy] = float4(0, 0, 0, 0);
        return;
    }

    float perceptualRoughness = MipmapLevelToPerceptualRoughness(mipLevel);
    float roughness = PerceptualRoughnessToRoughness(perceptualRoughness);

    float bias = 0.5 * roughness;

    float2 u = Fibonacci2d(sampleIndex, sampleCount);
    u.x = lerp(u.x, 0, bias);

    float3 localL;
    float  NdotL, NdotH, VdotH;

    SampleGGXDir(u, float3(0, 0, 1), k_identity3x3, roughness, localL, NdotL, NdotH, VdotH, true);

    if (NdotL <= 0)
    {
        // We are not supposed to generate wasteful samples.
        output[groupThreadId.xy] = float4(0, 0, 0, 0);
        return;
    }

    float pdf    = D_GGX(NdotH, roughness) * 0.25;
    float omegaS = rcp(sampleCount * pdf);

    output[groupThreadId.xy] = float4(localL, omegaS);
}