#include "ShaderLibrary/Common.hlsl"

Texture2D<float> _Source;
RWTexture2D<float> _Result;

SamplerState sampler_LinearClamp;

CBUFFER_START(cb)
    float4 _Size;
CBUFFER_END

#pragma kernel KMain
[numthreads(8, 8, 1)]
void KMain(uint2 groupId : SV_GroupID, uint2 groupThreadId : SV_GroupThreadID, uint2 dispatchThreadId : SV_DispatchThreadID)
{
    // Upper-left pixel coordinate of quad that this thread will read
    int2 threadUL = dispatchThreadId;

    // Downsample the block
    float2 offset = float2(threadUL);
    float p00 = _Source.SampleLevel(sampler_LinearClamp, (offset                   ) * _Size.zw, 0.0).x;
    float p10 = _Source.SampleLevel(sampler_LinearClamp, (offset + float2(1.0, 0.0)) * _Size.zw, 0.0).x;
    float p01 = _Source.SampleLevel(sampler_LinearClamp, (offset + float2(0.0, 1.0)) * _Size.zw, 0.0).x;
    float p11 = _Source.SampleLevel(sampler_LinearClamp, (offset + float2(1.0, 1.0)) * _Size.zw, 0.0).x;

    float depth = min(min(min(p00, p01), p10), p11);

    // Write to the final target
    _Result[dispatchThreadId] = depth;
}