// Put the following line to 0 or comment it to disable vignette weighting
#define USE_VIGNETTE_WEIGHTING 1

#pragma warning(disable : 3568)
#pragma exclude_renderers gles gles3 d3d11_9x

#include "../StdLib.hlsl"
#include "../Colors.hlsl"
#include "ExposureHistogram.hlsl"

RWStructuredBuffer<uint> _HistogramBuffer;
Texture2D<float4> _Source;
SamplerState sampler_LinearClamp;

CBUFFER_START(Params)
    float4 _ScaleOffsetRes; // x: scale, y: offset, z: width, w: height
CBUFFER_END

groupshared uint gs_histogram[HISTOGRAM_BINS];

#pragma kernel KEyeHistogram

#ifdef DISABLE_COMPUTE_SHADERS

TRIVIAL_COMPUTE_KERNEL(KEyeHistogram)
TRIVIAL_COMPUTE_KERNEL(KEyeHistogramClear)

#else

[numthreads(HISTOGRAM_THREAD_X, HISTOGRAM_THREAD_Y, 1)]
void KEyeHistogram(uint2 dispatchThreadId : SV_DispatchThreadID, uint2 groupThreadId : SV_GroupThreadID)
{
    // Pretty straightforward implementation of histogram gathering using atomic ops.
    // I tried a few methods (no atomic ops / heavy LDS leveraging) but this one turned out to be
    // the fastest on desktop (Nvidia - Kepler/Maxwell) and PS4. Still need to try it on GCN/desktop
    // but considering it runs very fast on PS4 we can expect it to run well (?).

    const uint localThreadId = groupThreadId.y * HISTOGRAM_THREAD_X + groupThreadId.x;

    // Clears the shared memory
    if (localThreadId < HISTOGRAM_BINS)
        gs_histogram[localThreadId] = 0u;

    float2 ipos = float2(dispatchThreadId) * 2.0;

    GroupMemoryBarrierWithGroupSync();

    // Gather local group histogram
    if (ipos.x < _ScaleOffsetRes.z && ipos.y < _ScaleOffsetRes.w)
    {
        uint weight = 1u;
        float2 sspos = ipos / _ScaleOffsetRes.zw;

        // Vignette weighting to put more focus on what's in the center of the screen
        #if USE_VIGNETTE_WEIGHTING
        {
            float2 d = abs(sspos - (0.5).xx);
            float vfactor = saturate(1.0 - dot(d, d));
            vfactor *= vfactor;
            weight = (uint)(64.0 * vfactor);
        }
        #endif

        float3 color = _Source.SampleLevel(sampler_LinearClamp, sspos, 0.0).xyz; // Bilinear downsample 2x
        float luminance = Luminance(color);
        float logLuminance = GetHistogramBinFromLuminance(luminance, _ScaleOffsetRes.xy);
        uint idx = (uint)(logLuminance * (HISTOGRAM_BINS - 1u));
        InterlockedAdd(gs_histogram[idx], weight);
    }

    GroupMemoryBarrierWithGroupSync();

    // Merge everything
    if (localThreadId < HISTOGRAM_BINS)
        InterlockedAdd(_HistogramBuffer[localThreadId], gs_histogram[localThreadId]);
}

#pragma kernel KEyeHistogramClear
[numthreads(HISTOGRAM_THREAD_X, 1, 1)]
void KEyeHistogramClear(uint dispatchThreadId : SV_DispatchThreadID)
{
    if (dispatchThreadId < HISTOGRAM_BINS)
        _HistogramBuffer[dispatchThreadId] = 0u;
}

#endif // DISABLE_COMPUTE_SHADERS