// Each #kernel tells which function to compile; you can have many kernels
#pragma kernel DeferredDirectionalShadow                    DEFERRED_DIRECTIONAL=DeferredDirectionalShadow
#pragma kernel DeferredDirectionalShadow_Contact            DEFERRED_DIRECTIONAL=DeferredDirectionalShadow_Contact          ENABLE_CONTACT_SHADOWS
#pragma kernel DeferredContactShadow

#ifdef SHADER_API_PSSL
#   pragma argument( scheduler=minpressure ) // instruct the shader compiler to prefer minimizing vgpr usage
#endif

#include "CoreRP/ShaderLibrary/Common.hlsl"
#include "../ShaderVariables.hlsl"
#include "../Material/NormalBuffer.hlsl"
#include "Lighting.hlsl"

#pragma only_renderers d3d11 ps4 xboxone vulkan metal switch

//#pragma enable_d3d11_debug_symbols

RWTexture2D<float4> _DeferredShadowTextureUAV;

CBUFFER_START(DeferredShadowParameters)
uint    _DirectionalShadowIndex;
float4  _DirectionalLightDirection;
float4  _PunctualLightPosition;
float4  _ContactShadowParamsParameters;
float4  _ContactShadowParamsParameters2;
int     _SampleCount;
CBUFFER_END

#define _ContactShadowLength                _ContactShadowParamsParameters.x
#define _ContactShadowDistanceScaleFactor   _ContactShadowParamsParameters.y
#define _ContactShadowFadeEnd               _ContactShadowParamsParameters.z
#define _ContactShadowFadeOneOverRange      _ContactShadowParamsParameters.w
#define _ContactShadowOpacity               _ContactShadowParamsParameters2.x

#define DEFERRED_SHADOW_TILE_SIZE 16

// Return 1.0 if occluded 0.0 if not
float4 ScreenSpaceShadowRayCast(float3 positionWS, float3 rayDirection, float rayLength)
{
    uint3 hashInput = uint3(abs(GetAbsolutePositionWS(positionWS)) * 1000);
    // Dither pattern is shifted by 0.5 because we want to jitter the ray starting position backward and forward (so we need values between -0.5 and 0.5)
    float ditherBias = 0.5;
    float dither = GenerateHashedRandomFloat(hashInput) - ditherBias;

    float3 rayStartWS = positionWS;
    float3 rayEndWS = rayStartWS + rayDirection * rayLength;

    float4 rayStartCS = TransformWorldToHClip(rayStartWS);
    float4 rayEndCS = TransformWorldToHClip(rayEndWS);

    // Here we compute a ray perpendicular to view space. This is the ray we use to compute the threshold for rejecting samples.
    // This is done this way so that the threshold is less dependent of ray slope.
    float4 rayOrthoViewSpace = rayStartCS + mul(GetViewToHClipMatrix(), float4(0, 0, rayLength, 0));
    rayOrthoViewSpace = rayOrthoViewSpace / rayOrthoViewSpace.w;

    rayStartCS.xyz = rayStartCS.xyz / rayStartCS.w;
    rayEndCS.xyz = rayEndCS.xyz / rayEndCS.w;

    // Pixel to light ray in clip space.
    float3 rayCS = rayEndCS.xyz - rayStartCS.xyz;

    // Depth at the start of the ray
    float startDepth = rayStartCS.z;
    // Depth range of the ray
    float rayDepth = rayCS.z;

    // Starting UV of the sampling loop
    float2 startUV = rayStartCS.xy * 0.5f + 0.5f;
    startUV.y = 1.0 - startUV.y;

    // Pixel to light ray in
    float2 rayUV = rayCS.xy * 0.5f;
    rayUV.y = -rayUV.y;

    float step = 1.0 / _SampleCount;
    float compareThreshold = abs(rayOrthoViewSpace.z - rayStartCS.z) * step;

    float occluded = 0.0f;

    for (int i = 0; i < _SampleCount; i++)
    {
        // Step for this sample
        float sampleStep = ((i + 1) * step + step * dither);

        // UVs for the current sample
        float2 sampleUV = (startUV + rayUV * sampleStep) * _ScreenToTargetScale.xy;
        // Ray depth for this sample
        float raySampleDepth = startDepth + rayDepth * sampleStep;

        // Depth buffer depth for this sample
        float sampleDepth = SAMPLE_TEXTURE2D_LOD(_CameraDepthTexture, s_point_clamp_sampler, sampleUV, 0.0).x;

        bool Hit = false;
        float depthDiff = sampleDepth - raySampleDepth;
        Hit = depthDiff < compareThreshold && depthDiff > 0.0;// 1e-4;

        if (Hit)
            occluded = 1.0f;
    }

    // Off screen masking
    // We remove the occlusion if the ray is occluded and only if direction steps out of the screen
    float2 vignette = max(6.0 * abs(rayStartCS.xy + rayCS.xy * occluded * 0.5) - 5.0, 0.0);
    occluded *= saturate( 1.0 - dot(vignette, vignette) );

    return occluded;
}

float ComputeContactShadow(PositionInputs posInput, float3 direction)
{
    float contactShadow = 1.0;
    
    if (_ContactShadowLength > 0.0f)
    {
        //Here LightDirection is not the light direction but the light position
        float4 result = ScreenSpaceShadowRayCast(posInput.positionWS, direction, _ContactShadowLength * max(0.5, posInput.linearDepth * _ContactShadowDistanceScaleFactor));
        contactShadow = 1.0 - result.x * saturate((_ContactShadowFadeEnd - posInput.linearDepth) * _ContactShadowFadeOneOverRange);
    }

    return lerp(1.0, contactShadow, _ContactShadowOpacity);
}

[numthreads(DEFERRED_SHADOW_TILE_SIZE, DEFERRED_SHADOW_TILE_SIZE, 1)]
void DEFERRED_DIRECTIONAL(uint2 groupThreadId : SV_GroupThreadID, uint2 groupId : SV_GroupID)
{
    uint2 pixelCoord = groupId * DEFERRED_SHADOW_TILE_SIZE + groupThreadId;
    uint2 tileCoord = groupId;
    float contactShadow = 1.0;

    float depth = LOAD_TEXTURE2D(_CameraDepthTexture, pixelCoord.xy).x;

    if (depth == UNITY_RAW_FAR_CLIP_VALUE)
        return;

    PositionInputs posInput = GetPositionInput(pixelCoord.xy, _ScreenSize.zw, depth, UNITY_MATRIX_I_VP, UNITY_MATRIX_V, tileCoord);
    
    //Direction got from either the directional light direction or the difference of punctual light position and the pixel position
    float3 direction = normalize(_DirectionalLightDirection.xyz * _DirectionalLightDirection.w + (_PunctualLightPosition.xyz - posInput.positionWS) * _PunctualLightPosition.w);

    NormalData normalData;
    DecodeFromNormalBuffer(posInput.positionSS, normalData);
    float3 normalWS = normalData.normalWS;

    ShadowContext shadowContext = InitShadowContext();
    float shadow = GetDirectionalShadowAttenuation(shadowContext, posInput.positionWS, normalWS, _DirectionalShadowIndex, _DirectionalLightDirection.xyz);

#ifdef ENABLE_CONTACT_SHADOWS
    contactShadow = ComputeContactShadow(posInput, direction);
#endif

    _DeferredShadowTextureUAV[pixelCoord] = float4(shadow, contactShadow, 1.0, 1.0); // Note: RT is RG16 format
}

[numthreads(DEFERRED_SHADOW_TILE_SIZE, DEFERRED_SHADOW_TILE_SIZE, 1)]
void DeferredContactShadow(uint2 groupThreadId : SV_GroupThreadID, uint2 groupId : SV_GroupID)
{
    uint2 pixelCoord = groupId * DEFERRED_SHADOW_TILE_SIZE + groupThreadId;
    uint2 tileCoord = groupId;
    float contactShadow = 1.0;

    float depth = LOAD_TEXTURE2D(_CameraDepthTexture, pixelCoord.xy).x;

    if (depth == UNITY_RAW_FAR_CLIP_VALUE)
        return;

    PositionInputs posInput = GetPositionInput(pixelCoord.xy, _ScreenSize.zw, depth, UNITY_MATRIX_I_VP, UNITY_MATRIX_V, tileCoord);
    
    float3 direction = normalize(_PunctualLightPosition.xyz - posInput.positionWS);

    contactShadow = ComputeContactShadow(posInput, direction);

    _DeferredShadowTextureUAV[pixelCoord] = float4(1.0, contactShadow, 1.0, 1.0); // Note: RT is RG16 format
}