// Each #kernel tells which function to compile; you can have many kernels
#pragma kernel CSMain
#pragma kernel CSMain _SKIP_3_MIP
// #define FFX_SPD_NO_WAVE_OPERATIONS 1
#pragma multi_compile _  FFX_SPD_NO_WAVE_OPERATIONS
#pragma enable_d3d11_debug_symbols
#pragma target 6.0
#pragma require wavebasic
#pragma require quadshuffle

#define FFX_GPU 1
#define FFX_HLSL 1
#define FFX_HALF 1

#define FFXM_GPU 1
#define FFXM_HALF 1
#define FFXM_HLSL 1

#include "ffx/ffx_core.h"
#include "ffx/ffx_common_types.h"
#include "Packages/com.unity.render-pipelines.core/ShaderLibrary/Common.hlsl"

Texture2D<FfxFloat32>  _InputDepth;
RWTexture2D<FfxFloat32>  _OutDepth;
float4 _MipOffsetAndSizeArray[16];

FfxUInt32   mips;
FfxUInt32x2   inputTextureSize;
FfxUInt32   numWorkGroups;
FFX_GROUPSHARED FfxUInt32 spdCounter;
RWTexture2D<FfxUInt32>    rw_internal_global_atomic;

FFX_GROUPSHARED FfxFloat32 spdIntermediate[16][16];
void SpdStoreIntermediate(FfxUInt32 x, FfxUInt32 y, FfxFloat32x4 value) {
    spdIntermediate[x][y] = value.x;
}

FfxFloat32x4 SpdLoadIntermediate(FfxUInt32 x, FfxUInt32 y) {
    FfxFloat32 f = spdIntermediate[x][y];
    return FfxFloat32x4(f.x, f.x, f.x, f.x);
}

FFX_GROUPSHARED FfxFloat32 spdIntermediateH[16][16];
void SpdStoreIntermediateH(FfxUInt32 x, FfxUInt32 y, FfxFloat32x4 value) {
    spdIntermediateH[x][y] = value.x;
}

FfxFloat32x4 SpdLoadIntermediateH(FfxUInt32 x, FfxUInt32 y) {
    FfxFloat32 f = spdIntermediateH[x][y];
    return FfxFloat32x4(f.x, f.x, f.x, f.x);
}

FfxFloat32x4 SpdReduce4(FfxFloat32x4 v0, FfxFloat32x4 v1, FfxFloat32x4 v2, FfxFloat32x4 v3) {
    #if UNITY_REVERSED_Z
    return min(min(v0, v1), min(v2, v3));
    #else
    return max(max(v0, v1), max(v2, v3));
    #endif
}

FfxFloat16x4 SpdReduce4H(FfxFloat16x4 v0, FfxFloat16x4 v1, FfxFloat16x4 v2, FfxFloat16x4 v3) {
    #if UNITY_REVERSED_Z
    return min(min(v0, v1), min(v2, v3));
    #else
    return max(max(v0, v1), max(v2, v3));
    #endif
}

FfxFloat32x4 SpdLoad(FfxInt32x2 coordinate, FfxUInt32 slice)
{
    uint4 cur = _MipOffsetAndSizeArray[6];
    return _OutDepth[coordinate + cur.xy].xxxx;   // 5 -> 6 as we store a copy of the depth buffer at index 0
}

void SpdStore(FfxInt32x2 pix, FfxFloat32x4 outValue, FfxUInt32 coordinate, FfxUInt32 slice)
{
#if _SKIP_3_MIP
    if(coordinate < 3)
    {
        return;
    }
#endif
    uint4 cur = _MipOffsetAndSizeArray[coordinate + 1];
    _OutDepth[pix + cur.xy] = outValue.x;    // + 1 as we store a copy of the depth buffer at index 0
}

FfxFloat32x4 SpdLoadH(FfxInt32x2 coordinate, FfxUInt32 slice)
{
    uint4 cur = _MipOffsetAndSizeArray[6];
    return _OutDepth[coordinate + cur.xy].xxxx;   // 5 -> 6 as we store a copy of the depth buffer at index 0
}

void SpdStoreH(FfxInt32x2 pix, FfxFloat32x4 outValue, FfxUInt32 coordinate, FfxUInt32 slice)
{
#if _SKIP_3_MIP
    if(coordinate < 3)
    {
        return;
    }
#endif
    uint4 cur = _MipOffsetAndSizeArray[coordinate + 1];
    _OutDepth[pix + cur.xy] = outValue.x;    // + 1 as we store a copy of the depth buffer at index 0
}

FfxFloat32x4 SpdLoadSourceImage(FfxInt32x2 coordinate, FfxUInt32 slice)
{
    return _InputDepth[coordinate].xxxx;
}

FfxFloat32x4 SpdLoadSourceImageH(FfxInt32x2 coordinate, FfxUInt32 slice)
{
    uint4 cur = _MipOffsetAndSizeArray[0];
    float2 uv = coordinate / float2(cur.zw);
    FfxInt32x2 uv2 = FfxInt32x2(uv * inputTextureSize);
    return _InputDepth[uv2].xxxx;
}

void IncreaseAtomicCounter(FFX_PARAMETER_IN FfxUInt32 slice, FFX_PARAMETER_INOUT FfxUInt32 counter)
{
    InterlockedAdd(rw_internal_global_atomic[FfxInt32x2(0, 0)] , 1, counter);
}

void ResetAtomicCounter(FFX_PARAMETER_IN FfxUInt32 slice)
{
    rw_internal_global_atomic[FfxInt32x2(0, 0)]  = 0;
}

FfxUInt32 SpdGetAtomicCounter()
{
    return spdCounter;
}

void SpdResetAtomicCounter(FfxUInt32 slice)
{
    ResetAtomicCounter(slice);
}

void SpdIncreaseAtomicCounter(FfxUInt32 slice)
{
    IncreaseAtomicCounter(slice, spdCounter);
}
#include "ffx/ffx_spd.h"

[numthreads(256,1,1)]
void CSMain (uint LocalThreadIndex : SV_GroupIndex, uint3 WorkGroupId : SV_GroupID)
{
    SpdDownsampleH(WorkGroupId.xy, LocalThreadIndex, mips, numWorkGroups, WorkGroupId.z, 0);
}