// Definitions
//--------------------------------------------------------------------------------------------------

// #pragma enable_d3d11_debug_symbols
#pragma only_renderers d3d11 playstation xboxone xboxseries vulkan metal switch

#pragma kernel MainCS

//#pragma multi_compile _ SSR_APPROX

//--------------------------------------------------------------------------------------------------
// Included headers
//--------------------------------------------------------------------------------------------------

#include "Packages/com.unity.render-pipelines.core/ShaderLibrary/Common.hlsl"
#include "Packages/com.unity.render-pipelines.core/ShaderLibrary/Packing.hlsl"
#include "Packages/com.unity.render-pipelines.high-definition/Runtime/ShaderLibrary/ShaderVariables.hlsl"
#include "Packages/com.unity.render-pipelines.high-definition/Runtime/Lighting/ScreenSpaceLighting/ScreenSpaceLighting.hlsl"
#include "Packages/com.unity.render-pipelines.high-definition/Runtime/Lighting/ScreenSpaceLighting/ShaderVariablesScreenSpaceReflection.cs.hlsl"
#include "Packages/com.unity.render-pipelines.high-definition/Runtime/Material/Builtin/BuiltinData.hlsl"

#ifdef DEBUG_DISPLAY
#include "Packages/com.unity.render-pipelines.core/ShaderLibrary/Debug.hlsl"
#include "Packages/com.unity.render-pipelines.high-definition/Runtime/Debug/DebugDisplay.hlsl"
#endif

RW_TEXTURE2D_ARRAY(float, _LensFlareOcclusion);

int _MultipassID;

groupshared float localOcclusion[8][8][SHADEROPTIONS_XR_MAX_VIEWS];

[numthreads(8, 8, 2)]
void MainCS(uint3 dispatchThreadId : SV_DispatchThreadID, uint3 groupThreadID : SV_GroupThreadID)
{
    uint threadIdZ;
    // For SinglePass XR
    if (_MultipassID < 0)
        threadIdZ = dispatchThreadId.z;
    else // For Multipass XR
        threadIdZ = (uint)_MultipassID;

    // Row[0] is reserved to store the results
    if (_MultipassID < 0 || (_MultipassID >= 0 && dispatchThreadId.z == threadIdZ))
        localOcclusion[groupThreadID.x][groupThreadID.y][threadIdZ] = _LensFlareOcclusion[uint3(dispatchThreadId.x, dispatchThreadId.y + 1, threadIdZ)].x;

    GroupMemoryBarrierWithGroupSync();

    if (groupThreadID.y < 4 && (_MultipassID < 0 || (_MultipassID >= 0 && dispatchThreadId.z == threadIdZ)))
        localOcclusion[groupThreadID.x][groupThreadID.y][threadIdZ] = max(localOcclusion[groupThreadID.x][groupThreadID.y][threadIdZ], localOcclusion[groupThreadID.x][groupThreadID.y + 4][threadIdZ]);
    GroupMemoryBarrierWithGroupSync();

    if (groupThreadID.y < 2 && (_MultipassID < 0 || (_MultipassID >= 0 && dispatchThreadId.z == threadIdZ)))
        localOcclusion[groupThreadID.x][groupThreadID.y][threadIdZ] = max(localOcclusion[groupThreadID.x][groupThreadID.y][threadIdZ], localOcclusion[groupThreadID.x][groupThreadID.y + 2][threadIdZ]);
    GroupMemoryBarrierWithGroupSync();

    if (groupThreadID.y == 0)
    {
        if (_MultipassID < 0 || (_MultipassID >= 0 && dispatchThreadId.z == threadIdZ))
            _LensFlareOcclusion[uint3(dispatchThreadId.x, 0, threadIdZ)] = max(localOcclusion[groupThreadID.x][0][threadIdZ], localOcclusion[groupThreadID.x][1][threadIdZ]);
    }
}