#pragma kernel StpSetup

#pragma multi_compile _ ENABLE_DEBUG_MODE
#pragma multi_compile _ ENABLE_STENCIL_RESPONSIVE
#pragma multi_compile _ ENABLE_LARGE_KERNEL

#pragma multi_compile _ UNITY_DEVICE_SUPPORTS_NATIVE_16BIT

// TODO: Re-enable support for wave reductions (usage of UNITY_DEVICE_SUPPORTS_WAVE_ANY keyword)
//
// We've run into many platform specific problems when trying to use wave operations for STP's reductions so they're being
// disabled for now. Enabling support for wave operations also causes us to use DXC on the 32-bit path on some Qualcomm Android
// devices and this triggers visual artifacts that we have no other way to work around at the moment.

#pragma multi_compile _ DISABLE_TEXTURE2D_X_ARRAY

#pragma only_renderers d3d11 playstation xboxone xboxseries vulkan metal switch

#include "Packages/com.unity.render-pipelines.core/ShaderLibrary/Common.hlsl"
#include "Packages/com.unity.render-pipelines.core/ShaderLibrary/Color.hlsl"
#include "Packages/com.unity.render-pipelines.core/ShaderLibrary/UnityInstancing.hlsl"

#define STP_PAT 1

#include "Packages/com.unity.render-pipelines.core/Runtime/STP/StpCommon.hlsl"

//
// Input
//

TEXTURE2D_X(_StpInputColor);
TEXTURE2D_X(_StpInputDepth);
TEXTURE2D_X(_StpInputMotion);

#if defined(ENABLE_STENCIL_RESPONSIVE)
    TYPED_TEXTURE2D_X(uint2, _StpInputStencil);
#endif

//
// Intermediate Output
//

RW_TEXTURE2D_X(float4, _StpIntermediateColor);
RW_TEXTURE2D_X(float, _StpIntermediateConvergence);

//
// History Input/Output
//

TYPED_TEXTURE2D_X(uint, _StpPriorDepthMotion);
RW_TEXTURE2D_X(uint, _StpDepthMotion);

TEXTURE2D_X(_StpPriorLuma);
RW_TEXTURE2D_X(float2, _StpLuma);

TEXTURE2D_X(_StpPriorConvergence);

TEXTURE2D_X(_StpPriorFeedback);

#define STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET (SLICE_ARRAY_INDEX * STPSETUPPERVIEWCONSTANTS_COUNT)

#if defined(SHADER_API_PSSL) || defined(SHADER_API_SWITCH) || (defined(SHADER_API_METAL) && !defined(SHADER_API_MOBILE))
    // Force usage of the 32-bit reduction path even in 16-bit environments
    #define STP_FORCE_32BIT_REDUCTION
#endif

#if defined(SHADER_API_PSSL) || defined(SHADER_API_GAMECORE) || defined(SHADER_API_METAL) || (defined(SHADER_API_VULKAN) && defined(SHADER_API_MOBILE))
    // Force usage of group shared memory instead using wave operations
    #define STP_FORCE_GROUPSHARED
#endif

// Enable the use of wave operations when they're supported by the current hardware and usage of groupshared hasn't been forced.
#if defined(UNITY_HW_SUPPORTS_WAVE) && !defined(STP_FORCE_GROUPSHARED)
    #define STP_ENABLE_WAVEOPS
#endif

// STP requires a 4x4 reduction which must be implemented by either wave operations, or group shared memory.
#if !defined(STP_ENABLE_WAVEOPS)
#if defined(STP_16BIT) && !defined(STP_FORCE_32BIT_REDUCTION)
groupshared uint4 gs_StpScratch[STP_GROUP_SIZE];
#else
groupshared float4 gs_StpScratch[STP_GROUP_SIZE * 2];
#endif
#endif

// In some cases, we have to expose the 32-bit reduction code in the 16-bit path
#if defined(STP_32BIT) || defined(STP_FORCE_32BIT_REDUCTION)
void StpPat4x4MaxF8(StpMU1 i, inout StpF4 a, inout StpF4 b)
{
#if defined(STP_ENABLE_WAVEOPS)
    a.x = max(a.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 1)));
    a.y = max(a.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 1)));
    a.z = max(a.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 1)));
    a.w = max(a.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 1)));
    b.x = max(b.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.x), WaveGetLaneIndex() ^ 1)));
    b.y = max(b.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.y), WaveGetLaneIndex() ^ 1)));
    b.z = max(b.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.z), WaveGetLaneIndex() ^ 1)));
    b.w = max(b.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.w), WaveGetLaneIndex() ^ 1)));

    a.x = max(a.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 2)));
    a.y = max(a.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 2)));
    a.z = max(a.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 2)));
    a.w = max(a.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 2)));
    b.x = max(b.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.x), WaveGetLaneIndex() ^ 2)));
    b.y = max(b.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.y), WaveGetLaneIndex() ^ 2)));
    b.z = max(b.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.z), WaveGetLaneIndex() ^ 2)));
    b.w = max(b.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.w), WaveGetLaneIndex() ^ 2)));

    a.x = max(a.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 4)));
    a.y = max(a.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 4)));
    a.z = max(a.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 4)));
    a.w = max(a.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 4)));
    b.x = max(b.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.x), WaveGetLaneIndex() ^ 4)));
    b.y = max(b.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.y), WaveGetLaneIndex() ^ 4)));
    b.z = max(b.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.z), WaveGetLaneIndex() ^ 4)));
    b.w = max(b.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.w), WaveGetLaneIndex() ^ 4)));

    a.x = max(a.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 8)));
    a.y = max(a.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 8)));
    a.z = max(a.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 8)));
    a.w = max(a.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 8)));
    b.x = max(b.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.x), WaveGetLaneIndex() ^ 8)));
    b.y = max(b.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.y), WaveGetLaneIndex() ^ 8)));
    b.z = max(b.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.z), WaveGetLaneIndex() ^ 8)));
    b.w = max(b.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.w), WaveGetLaneIndex() ^ 8)));
#else
    gs_StpScratch[i] = a;
    gs_StpScratch[i + STP_GROUP_SIZE] = b;

    GroupMemoryBarrierWithGroupSync();

    // 2x2 Reduction
    {
        StpMU1 offset = (i & ~StpMU1(3));

        StpMU1 a0 = offset + ((i + StpMU1(1)) & StpMU1(3));
        StpMU1 a1 = offset + ((i + StpMU1(2)) & StpMU1(3));
        StpMU1 a2 = offset + ((i + StpMU1(3)) & StpMU1(3));

        float4 x0 = gs_StpScratch[a0];
        float4 x1 = gs_StpScratch[a1];
        float4 x2 = gs_StpScratch[a2];

        float4 y0 = gs_StpScratch[a0 + STP_GROUP_SIZE];
        float4 y1 = gs_StpScratch[a1 + STP_GROUP_SIZE];
        float4 y2 = gs_StpScratch[a2 + STP_GROUP_SIZE];

        GroupMemoryBarrierWithGroupSync();

        a = max(max(max(a, x0), x1), x2);
        b = max(max(max(b, y0), y1), y2);
    }

    gs_StpScratch[i] = a;
    gs_StpScratch[i + STP_GROUP_SIZE] = b;

    GroupMemoryBarrierWithGroupSync();

    // 4x4 Reduction
    {
        StpMU1 offset = (i & ~StpMU1(15));

        StpMU1 a0 = offset + ((i + StpMU1(4)) & StpMU1(15));
        StpMU1 a1 = offset + ((i + StpMU1(8)) & StpMU1(15));
        StpMU1 a2 = offset + ((i + StpMU1(12)) & StpMU1(15));

        float4 x0 = gs_StpScratch[a0];
        float4 x1 = gs_StpScratch[a1];
        float4 x2 = gs_StpScratch[a2];

        float4 y0 = gs_StpScratch[a0 + STP_GROUP_SIZE];
        float4 y1 = gs_StpScratch[a1 + STP_GROUP_SIZE];
        float4 y2 = gs_StpScratch[a2 + STP_GROUP_SIZE];

        GroupMemoryBarrierWithGroupSync();

        a = max(max(max(a, x0), x1), x2);
        b = max(max(max(b, y0), y1), y2);
    }
#endif
}
void StpPat4x4SumF4(StpMU1 i, inout StpF4 a)
{
#if defined(STP_ENABLE_WAVEOPS)
    a.x += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 1));
    a.y += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 1));
    a.z += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 1));
    a.w += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 1));

    a.x += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 2));
    a.y += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 2));
    a.z += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 2));
    a.w += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 2));

    a.x += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 4));
    a.y += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 4));
    a.z += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 4));
    a.w += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 4));

    a.x += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 8));
    a.y += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 8));
    a.z += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 8));
    a.w += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 8));
#else
    gs_StpScratch[i] = a;

    GroupMemoryBarrierWithGroupSync();

    // 2x2 Reduction
    {
        StpMU1 offset = (i & ~StpMU1(3));

        StpMU1 a0 = offset + ((i + StpMU1(1)) & StpMU1(3));
        StpMU1 a1 = offset + ((i + StpMU1(2)) & StpMU1(3));
        StpMU1 a2 = offset + ((i + StpMU1(3)) & StpMU1(3));

        float4 x0 = gs_StpScratch[a0];
        float4 x1 = gs_StpScratch[a1];
        float4 x2 = gs_StpScratch[a2];

        GroupMemoryBarrierWithGroupSync();

        a = a + x0 + x1 + x2;
    }

    gs_StpScratch[i] = a;

    GroupMemoryBarrierWithGroupSync();

    // 4x4 Reduction
    {
        StpMU1 offset = (i & ~StpMU1(15));

        StpMU1 a0 = offset + ((i + StpMU1(4)) & StpMU1(15));
        StpMU1 a1 = offset + ((i + StpMU1(8)) & StpMU1(15));
        StpMU1 a2 = offset + ((i + StpMU1(12)) & StpMU1(15));

        float4 x0 = gs_StpScratch[a0];
        float4 x1 = gs_StpScratch[a1];
        float4 x2 = gs_StpScratch[a2];

        GroupMemoryBarrierWithGroupSync();

        a = a + x0 + x1 + x2;
    }
#endif
}
#endif

#if defined(STP_16BIT)
void StpPat4x4MaxH8(StpW1 i, inout StpH4 a, inout StpH4 b)
{
#if defined(STP_FORCE_32BIT_REDUCTION)
    StpPat4x4MaxF8(i, a, b);
#else
#if defined(STP_ENABLE_WAVEOPS)
    a.xy = max(a.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 1)));
    a.zw = max(a.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 1)));
    b.xy = max(b.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.xy), WaveGetLaneIndex() ^ 1)));
    b.zw = max(b.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.zw), WaveGetLaneIndex() ^ 1)));

    a.xy = max(a.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 2)));
    a.zw = max(a.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 2)));
    b.xy = max(b.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.xy), WaveGetLaneIndex() ^ 2)));
    b.zw = max(b.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.zw), WaveGetLaneIndex() ^ 2)));

    a.xy = max(a.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 4)));
    a.zw = max(a.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 4)));
    b.xy = max(b.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.xy), WaveGetLaneIndex() ^ 4)));
    b.zw = max(b.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.zw), WaveGetLaneIndex() ^ 4)));

    a.xy = max(a.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 8)));
    a.zw = max(a.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 8)));
    b.xy = max(b.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.xy), WaveGetLaneIndex() ^ 8)));
    b.zw = max(b.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.zw), WaveGetLaneIndex() ^ 8)));
#else
    gs_StpScratch[i] = StpU4(StpU1_H2(a.xy), StpU1_H2(a.zw), StpU1_H2(b.xy), StpU1_H2(b.zw));

    GroupMemoryBarrierWithGroupSync();

    // 2x2 Reduction
    {
        StpW1 offset = (i & ~StpW1(3));

        StpW1 a0 = offset + ((i + StpW1(1)) & StpW1(3));
        StpW1 a1 = offset + ((i + StpW1(2)) & StpW1(3));
        StpW1 a2 = offset + ((i + StpW1(3)) & StpW1(3));

        uint4 x0 = gs_StpScratch[a0];
        uint4 x1 = gs_StpScratch[a1];
        uint4 x2 = gs_StpScratch[a2];

        GroupMemoryBarrierWithGroupSync();

        a.xy = max(max(max(a.xy, StpH2_U1(x0.x)), StpH2_U1(x1.x)), StpH2_U1(x2.x));
        a.zw = max(max(max(a.zw, StpH2_U1(x0.y)), StpH2_U1(x1.y)), StpH2_U1(x2.y));
        b.xy = max(max(max(b.xy, StpH2_U1(x0.z)), StpH2_U1(x1.z)), StpH2_U1(x2.z));
        b.zw = max(max(max(b.zw, StpH2_U1(x0.w)), StpH2_U1(x1.w)), StpH2_U1(x2.w));
    }

    gs_StpScratch[i] = StpU4(StpU1_H2(a.xy), StpU1_H2(a.zw), StpU1_H2(b.xy), StpU1_H2(b.zw));

    GroupMemoryBarrierWithGroupSync();

    // 4x4 Reduction
    {
        StpW1 offset = (i & ~StpW1(15));

        StpW1 a0 = offset + ((i + StpW1(4)) & StpW1(15));
        StpW1 a1 = offset + ((i + StpW1(8)) & StpW1(15));
        StpW1 a2 = offset + ((i + StpW1(12)) & StpW1(15));

        uint4 x0 = gs_StpScratch[a0];
        uint4 x1 = gs_StpScratch[a1];
        uint4 x2 = gs_StpScratch[a2];

        GroupMemoryBarrierWithGroupSync();

        a.xy = max(max(max(a.xy, StpH2_U1(x0.x)), StpH2_U1(x1.x)), StpH2_U1(x2.x));
        a.zw = max(max(max(a.zw, StpH2_U1(x0.y)), StpH2_U1(x1.y)), StpH2_U1(x2.y));
        b.xy = max(max(max(b.xy, StpH2_U1(x0.z)), StpH2_U1(x1.z)), StpH2_U1(x2.z));
        b.zw = max(max(max(b.zw, StpH2_U1(x0.w)), StpH2_U1(x1.w)), StpH2_U1(x2.w));
    }
#endif
#endif
}
void StpPat4x4SumH4(StpW1 i, inout StpH4 a)
{
#if defined(STP_FORCE_32BIT_REDUCTION)
    StpPat4x4SumF4(i, a);
#else
#if defined(STP_ENABLE_WAVEOPS)
    a.xy += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 1));
    a.zw += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 1));

    a.xy += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 2));
    a.zw += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 2));

    a.xy += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 4));
    a.zw += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 4));

    a.xy += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 8));
    a.zw += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 8));
#else
    gs_StpScratch[i].xy = StpU2(StpU1_H2(a.xy), StpU1_H2(a.zw));

    GroupMemoryBarrierWithGroupSync();

    // 2x2 Reduction
    {
        StpW1 offset = (i & ~StpW1(3));

        StpW1 a0 = offset + ((i + StpW1(1)) & StpW1(3));
        StpW1 a1 = offset + ((i + StpW1(2)) & StpW1(3));
        StpW1 a2 = offset + ((i + StpW1(3)) & StpW1(3));

        uint2 x0 = gs_StpScratch[a0].xy;
        uint2 x1 = gs_StpScratch[a1].xy;
        uint2 x2 = gs_StpScratch[a2].xy;

        GroupMemoryBarrierWithGroupSync();

        a.xy = a.xy + StpH2_U1(x0.x) + StpH2_U1(x1.x) + StpH2_U1(x2.x);
        a.zw = a.zw + StpH2_U1(x0.y) + StpH2_U1(x1.y) + StpH2_U1(x2.y);
    }

    gs_StpScratch[i].xy = StpU2(StpU1_H2(a.xy), StpU1_H2(a.zw));

    GroupMemoryBarrierWithGroupSync();

    // 4x4 Reduction
    {
        StpW1 offset = (i & ~StpW1(15));

        StpW1 a0 = offset + ((i + StpW1(4)) & StpW1(15));
        StpW1 a1 = offset + ((i + StpW1(8)) & StpW1(15));
        StpW1 a2 = offset + ((i + StpW1(12)) & StpW1(15));

        uint2 x0 = gs_StpScratch[a0].xy;
        uint2 x1 = gs_StpScratch[a1].xy;
        uint2 x2 = gs_StpScratch[a2].xy;

        GroupMemoryBarrierWithGroupSync();

        a.xy = a.xy + StpH2_U1(x0.x) + StpH2_U1(x1.x) + StpH2_U1(x2.x);
        a.zw = a.zw + StpH2_U1(x0.y) + StpH2_U1(x1.y) + StpH2_U1(x2.y);
    }
#endif
#endif
}
StpH1 StpPatPriConH(StpF2 p) { return (StpH1)SAMPLE_TEXTURE2D_X_LOD(_StpPriorConvergence, s_linear_clamp_sampler, p, 0); }

// These are separate to support inline operation (pass merged instead of loads).
StpF2 StpPatDatMotH(StpW2 o) { return LOAD_TEXTURE2D_X_LOD(_StpInputMotion, o, 0).xy; }
StpH3 StpPatDatColH(StpW2 o) { return (StpH3)LOAD_TEXTURE2D_X_LOD(_StpInputColor, o, 0).rgb; }
StpF1 StpPatDatZH(StpW2 o) { return LOAD_TEXTURE2D_X_LOD(_StpInputDepth, o, 0).x; }
// This provides a place to convert Z from depth to linear if not inlined and actually loaded.
StpF1 StpPatFixZH(StpF1 z) { return 1.0 / (STP_ZBUFFER_PARAMS_Z * z + STP_ZBUFFER_PARAMS_W); }
StpU1 StpPatDatRH(StpW2 o) {
    #if defined(ENABLE_STENCIL_RESPONSIVE)
        return GetStencilValue(LOAD_TEXTURE2D_X_LOD(_StpInputStencil, o, 0).xy);
    #endif // defined(ENABLE_STENCIL_RESPONSIVE)
    return StpU1_(0); }
StpH1 StpPatFixRH(StpU1 v) {
    // Activate the "responsive" feature when we don't have valid history textures.
    bool hasValidHistory = DecodeHasValidHistory(STP_COMMON_CONSTANT);
    bool excludeTaa = false;
    #if defined(ENABLE_STENCIL_RESPONSIVE)
        excludeTaa = (v & DecodeStencilMask(STP_COMMON_CONSTANT)) != 0;
    #endif // defined(ENABLE_STENCIL_RESPONSIVE)
    return (hasValidHistory && !excludeTaa) ? StpH1_(1.0) : StpH1_(0.0); }

StpH1 StpPatDitH(StpW2 o) { return StpDitH1(o); }
StpH4 StpPatPriFedH(StpF2 p) { return (StpH4)SAMPLE_TEXTURE2D_X_LOD(_StpPriorFeedback, s_linear_clamp_sampler, p, 0); }
StpH4 StpPatPriFedR4H(StpF2 p) { return (StpH4)GATHER_RED_TEXTURE2D_X(_StpPriorFeedback, s_point_clamp_sampler, p); }
StpH4 StpPatPriFedG4H(StpF2 p) { return (StpH4)GATHER_GREEN_TEXTURE2D_X(_StpPriorFeedback, s_point_clamp_sampler, p); }
StpH4 StpPatPriFedB4H(StpF2 p) { return (StpH4)GATHER_BLUE_TEXTURE2D_X(_StpPriorFeedback, s_point_clamp_sampler, p); }
StpH2 StpPatPriLumH(StpF2 p) { return (StpH2)SAMPLE_TEXTURE2D_X_LOD(_StpPriorLuma, s_linear_clamp_sampler, p, 0); }
StpU4 StpPatPriMot4H(StpF2 p) { return GATHER_RED_TEXTURE2D_X(_StpPriorDepthMotion, s_point_clamp_sampler, p); }

void StpPatStMotH(StpW2 p, StpU1 v) { _StpDepthMotion[COORD_TEXTURE2D_X(p)] = v; }
void StpPatStColH(StpW2 p, StpH4 v) { _StpIntermediateColor[COORD_TEXTURE2D_X(p)] = v; }
void StpPatStLumH(StpW2 p, StpH2 v) { _StpLuma[COORD_TEXTURE2D_X(p)] = v; }
void StpPatStCnvH(StpW2 p, StpH1 v) { _StpIntermediateConvergence[COORD_TEXTURE2D_X(p >> StpW1(2))] = v; }
#endif

#if defined(STP_32BIT)
StpMF1 StpPatPriConF(StpF2 p) { return (StpMF1)SAMPLE_TEXTURE2D_X_LOD(_StpPriorConvergence, s_linear_clamp_sampler, p, 0); }

// These are separate to support inline operation (pass merged instead of loads).
StpF2 StpPatDatMotF(StpMU2 o) { return LOAD_TEXTURE2D_X_LOD(_StpInputMotion, o, 0).xy; }
StpMF3 StpPatDatColF(StpMU2 o) { return (StpMF3)LOAD_TEXTURE2D_X_LOD(_StpInputColor, o, 0).rgb; }
StpF1 StpPatDatZF(StpMU2 o) { return LOAD_TEXTURE2D_X_LOD(_StpInputDepth, o, 0).x; }
// This provides a place to convert Z from depth to linear if not inlined and actually loaded.
StpF1 StpPatFixZF(StpF1 z) { return 1.0 / (STP_ZBUFFER_PARAMS_Z * z + STP_ZBUFFER_PARAMS_W); }
StpU1 StpPatDatRF(StpMU2 o) {
    #if defined(ENABLE_STENCIL_RESPONSIVE)
        return GetStencilValue(LOAD_TEXTURE2D_X_LOD(_StpInputStencil, o, 0).xy);
    #endif // defined(ENABLE_STENCIL_RESPONSIVE)
    return StpU1_(0); }
StpMF1 StpPatFixRF(StpU1 v) {
    // Activate the "responsive" feature when we don't have valid history textures.
    bool hasValidHistory = DecodeHasValidHistory(STP_COMMON_CONSTANT);
    bool excludeTaa = false;
    #if defined(ENABLE_STENCIL_RESPONSIVE)
        excludeTaa = (v & DecodeStencilMask(STP_COMMON_CONSTANT)) != 0;
    #endif // defined(ENABLE_STENCIL_RESPONSIVE)
    return (hasValidHistory && !excludeTaa) ? StpMF1_(1.0) : StpMF1_(0.0); }

StpMF1 StpPatDitF(StpMU2 o) { return (StpMF1)StpDitF1(o); }
StpMF4 StpPatPriFedF(StpF2 p) { return (StpMF4)SAMPLE_TEXTURE2D_X_LOD(_StpPriorFeedback, s_linear_clamp_sampler, p, 0); }
StpMF4 StpPatPriFedR4F(StpF2 p) { return (StpMF4)GATHER_RED_TEXTURE2D_X(_StpPriorFeedback, s_point_clamp_sampler, p); }
StpMF4 StpPatPriFedG4F(StpF2 p) { return (StpMF4)GATHER_GREEN_TEXTURE2D_X(_StpPriorFeedback, s_point_clamp_sampler, p); }
StpMF4 StpPatPriFedB4F(StpF2 p) { return (StpMF4)GATHER_BLUE_TEXTURE2D_X(_StpPriorFeedback, s_point_clamp_sampler, p); }
StpMF2 StpPatPriLumF(StpF2 p) { return (StpMF2)SAMPLE_TEXTURE2D_X_LOD(_StpPriorLuma, s_linear_clamp_sampler, p, 0); }
StpU4 StpPatPriMot4F(StpF2 p) { return GATHER_RED_TEXTURE2D_X(_StpPriorDepthMotion, s_point_clamp_sampler, p); }

void StpPatStMotF(StpMU2 p, StpU1 v) { _StpDepthMotion[COORD_TEXTURE2D_X(p)] = v; }
void StpPatStColF(StpMU2 p, StpMF4 v) { _StpIntermediateColor[COORD_TEXTURE2D_X(p)] = v; }
void StpPatStLumF(StpMU2 p, StpMF2 v) { _StpLuma[COORD_TEXTURE2D_X(p)] = v; }
void StpPatStCnvF(StpMU2 p, StpMF1 v) { _StpIntermediateConvergence[COORD_TEXTURE2D_X(p >> StpMU1(2))] = v; }
#endif

#define THREADING_BLOCK_SIZE STP_GROUP_SIZE
#include "Packages/com.unity.render-pipelines.core/ShaderLibrary/Threading.hlsl"

[numthreads(STP_GROUP_SIZE, 1, 1)]
void StpSetup(Threading::Group group)
{
    UNITY_XR_ASSIGN_VIEW_INDEX(group.groupID.z);

#if defined(STP_16BIT)
    StpW1 lane = StpW1_(group.groupIndex);
    StpW2 groupPos = ComputeGroupPos(StpW2(group.groupID.xy));
    StpW2 pos = groupPos + StpRemapLaneTo8x16H(lane);
#else
    StpMU1 lane = StpMU1_(group.groupIndex);
    StpMU2 groupPos = ComputeGroupPos(StpMU2(group.groupID.xy));
    StpMU2 pos = groupPos + StpRemapLaneTo8x16F(lane);
#endif

#if defined(STP_16BIT)
    StpPatH(
        lane,
        pos,
#else
    StpPatF(
        lane,
        pos,
#endif

        asuint(_StpSetupConstants0),
        asuint(_StpSetupConstants1),
        asuint(_StpSetupConstants2),
        asuint(_StpSetupConstants3),
        asuint(_StpSetupConstants4),
        asuint(_StpSetupConstants5),

        asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 0]),
        asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 1]),
        asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 2]),
        asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 3]),
        asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 4]),
        asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 5]),
        asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 6]),
        asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 7])
    );
}