Singularity/Library/PackageCache/com.unity.render-pipelines..../ShaderLibrary/SSAO.hlsl

463 lines
15 KiB
HLSL
Raw Normal View History

2024-05-06 14:45:45 -04:00
#ifndef UNIVERSAL_SSAO_INCLUDED
#define UNIVERSAL_SSAO_INCLUDED
// Includes
#include "Packages/com.unity.render-pipelines.core/ShaderLibrary/Common.hlsl"
#include "Packages/com.unity.render-pipelines.universal/ShaderLibrary/ShaderVariablesFunctions.hlsl"
#include "Packages/com.unity.render-pipelines.universal/ShaderLibrary/DeclareDepthTexture.hlsl"
#include "Packages/com.unity.render-pipelines.universal/ShaderLibrary/DeclareNormalsTexture.hlsl"
// Textures & Samplers
TEXTURE2D_X(_BaseMap);
TEXTURE2D_X(_ScreenSpaceOcclusionTexture);
SAMPLER(sampler_BaseMap);
SAMPLER(sampler_ScreenSpaceOcclusionTexture);
// Params
half4 _SSAOParams;
half4 _CameraViewTopLeftCorner[2];
half4x4 _CameraViewProjections[2]; // This is different from UNITY_MATRIX_VP (platform-agnostic projection matrix is used). Handle both non-XR and XR modes.
float4 _SourceSize;
float4 _ProjectionParams2;
float4 _CameraViewXExtent[2];
float4 _CameraViewYExtent[2];
float4 _CameraViewZExtent[2];
// Hardcoded random UV values that improves performance.
// The values were taken from this function:
// r = frac(43758.5453 * sin( dot(float2(12.9898, 78.233), uv)) ));
// Indices 0 to 19 are for u = 0.0
// Indices 20 to 39 are for u = 1.0
static half SSAORandomUV[40] =
{
0.00000000, // 00
0.33984375, // 01
0.75390625, // 02
0.56640625, // 03
0.98437500, // 04
0.07421875, // 05
0.23828125, // 06
0.64062500, // 07
0.35937500, // 08
0.50781250, // 09
0.38281250, // 10
0.98437500, // 11
0.17578125, // 12
0.53906250, // 13
0.28515625, // 14
0.23137260, // 15
0.45882360, // 16
0.54117650, // 17
0.12941180, // 18
0.64313730, // 19
0.92968750, // 20
0.76171875, // 21
0.13333330, // 22
0.01562500, // 23
0.00000000, // 24
0.10546875, // 25
0.64062500, // 26
0.74609375, // 27
0.67968750, // 28
0.35156250, // 29
0.49218750, // 30
0.12500000, // 31
0.26562500, // 32
0.62500000, // 33
0.44531250, // 34
0.17647060, // 35
0.44705890, // 36
0.93333340, // 37
0.87058830, // 38
0.56862750, // 39
};
// SSAO Settings
#define INTENSITY _SSAOParams.x
#define RADIUS _SSAOParams.y
#define DOWNSAMPLE _SSAOParams.z
// GLES2: In many cases, dynamic looping is not supported.
#if defined(SHADER_API_GLES) && !defined(SHADER_API_GLES3)
#define SAMPLE_COUNT 3
#else
#define SAMPLE_COUNT int(_SSAOParams.w)
#endif
// Function defines
#define SCREEN_PARAMS GetScaledScreenParams()
#define SAMPLE_BASEMAP(uv) SAMPLE_TEXTURE2D_X(_BaseMap, sampler_BaseMap, UnityStereoTransformScreenSpaceTex(uv));
// Constants
// kContrast determines the contrast of occlusion. This allows users to control over/under
// occlusion. At the moment, this is not exposed to the editor because it's rarely useful.
// The range is between 0 and 1.
static const half kContrast = half(0.5);
// The constant below controls the geometry-awareness of the bilateral
// filter. The higher value, the more sensitive it is.
static const half kGeometryCoeff = half(0.8);
// The constants below are used in the AO estimator. Beta is mainly used for suppressing
// self-shadowing noise, and Epsilon is used to prevent calculation underflow. See the paper
// (Morgan 2011 https://casual-effects.com/research/McGuire2011AlchemyAO/index.html)
// for further details of these constants.
static const half kBeta = half(0.002);
static const half kEpsilon = half(0.0001);
#if defined(USING_STEREO_MATRICES)
#define unity_eyeIndex unity_StereoEyeIndex
#else
#define unity_eyeIndex 0
#endif
half4 PackAONormal(half ao, half3 n)
{
return half4(ao, n * half(0.5) + half(0.5));
}
half3 GetPackedNormal(half4 p)
{
return p.gba * half(2.0) - half(1.0);
}
half GetPackedAO(half4 p)
{
return p.r;
}
half EncodeAO(half x)
{
#if UNITY_COLORSPACE_GAMMA
return half(1.0 - max(LinearToSRGB(1.0 - saturate(x)), 0.0));
#else
return x;
#endif
}
half CompareNormal(half3 d1, half3 d2)
{
return smoothstep(kGeometryCoeff, half(1.0), dot(d1, d2));
}
// Trigonometric function utility
half2 CosSin(half theta)
{
half sn, cs;
sincos(theta, sn, cs);
return half2(cs, sn);
}
// Pseudo random number generator with 2D coordinates
half GetRandomUVForSSAO(float u, int sampleIndex)
{
return SSAORandomUV[u * 20 + sampleIndex];
}
float2 GetScreenSpacePosition(float2 uv)
{
return float2(uv * SCREEN_PARAMS.xy * DOWNSAMPLE);
}
// Sample point picker
half3 PickSamplePoint(float2 uv, int sampleIndex)
{
const float2 positionSS = GetScreenSpacePosition(uv);
const half gn = half(InterleavedGradientNoise(positionSS, sampleIndex));
const half u = frac(GetRandomUVForSSAO(half(0.0), sampleIndex) + gn) * half(2.0) - half(1.0);
const half theta = (GetRandomUVForSSAO(half(1.0), sampleIndex) + gn) * half(TWO_PI);
return half3(CosSin(theta) * sqrt(half(1.0) - u * u), u);
}
float SampleAndGetLinearEyeDepth(float2 uv)
{
float rawDepth = SampleSceneDepth(uv.xy);
#if defined(_ORTHOGRAPHIC)
return LinearDepthToEyeDepth(rawDepth);
#else
return LinearEyeDepth(rawDepth, _ZBufferParams);
#endif
}
// This returns a vector in world unit (not a position), from camera to the given point described by uv screen coordinate and depth (in absolute world unit).
half3 ReconstructViewPos(float2 uv, float depth)
{
// Screen is y-inverted.
uv.y = 1.0 - uv.y;
// view pos in world space
#if defined(_ORTHOGRAPHIC)
float zScale = depth * _ProjectionParams.w; // divide by far plane
float3 viewPos = _CameraViewTopLeftCorner[unity_eyeIndex].xyz
+ _CameraViewXExtent[unity_eyeIndex].xyz * uv.x
+ _CameraViewYExtent[unity_eyeIndex].xyz * uv.y
+ _CameraViewZExtent[unity_eyeIndex].xyz * zScale;
#else
float zScale = depth * _ProjectionParams2.x; // divide by near plane
float3 viewPos = _CameraViewTopLeftCorner[unity_eyeIndex].xyz
+ _CameraViewXExtent[unity_eyeIndex].xyz * uv.x
+ _CameraViewYExtent[unity_eyeIndex].xyz * uv.y;
viewPos *= zScale;
#endif
return half3(viewPos);
}
// Try reconstructing normal accurately from depth buffer.
// Low: DDX/DDY on the current pixel
// Medium: 3 taps on each direction | x | * | y |
// High: 5 taps on each direction: | z | x | * | y | w |
// https://atyuwen.github.io/posts/normal-reconstruction/
// https://wickedengine.net/2019/09/22/improved-normal-reconstruction-from-depth/
half3 ReconstructNormal(float2 uv, float depth, float3 vpos)
{
#if defined(_RECONSTRUCT_NORMAL_LOW)
return half3(normalize(cross(ddy(vpos), ddx(vpos))));
#else
float2 delta = float2(_SourceSize.zw * 2.0);
// Sample the neighbour fragments
float2 lUV = float2(-delta.x, 0.0);
float2 rUV = float2( delta.x, 0.0);
float2 uUV = float2(0.0, delta.y);
float2 dUV = float2(0.0, -delta.y);
float3 l1 = float3(uv + lUV, 0.0); l1.z = SampleAndGetLinearEyeDepth(l1.xy); // Left1
float3 r1 = float3(uv + rUV, 0.0); r1.z = SampleAndGetLinearEyeDepth(r1.xy); // Right1
float3 u1 = float3(uv + uUV, 0.0); u1.z = SampleAndGetLinearEyeDepth(u1.xy); // Up1
float3 d1 = float3(uv + dUV, 0.0); d1.z = SampleAndGetLinearEyeDepth(d1.xy); // Down1
// Determine the closest horizontal and vertical pixels...
// horizontal: left = 0.0 right = 1.0
// vertical : down = 0.0 up = 1.0
#if defined(_RECONSTRUCT_NORMAL_MEDIUM)
uint closest_horizontal = l1.z > r1.z ? 0 : 1;
uint closest_vertical = d1.z > u1.z ? 0 : 1;
#else
float3 l2 = float3(uv + lUV * 2.0, 0.0); l2.z = SampleAndGetLinearEyeDepth(l2.xy); // Left2
float3 r2 = float3(uv + rUV * 2.0, 0.0); r2.z = SampleAndGetLinearEyeDepth(r2.xy); // Right2
float3 u2 = float3(uv + uUV * 2.0, 0.0); u2.z = SampleAndGetLinearEyeDepth(u2.xy); // Up2
float3 d2 = float3(uv + dUV * 2.0, 0.0); d2.z = SampleAndGetLinearEyeDepth(d2.xy); // Down2
const uint closest_horizontal = abs( (2.0 * l1.z - l2.z) - depth) < abs( (2.0 * r1.z - r2.z) - depth) ? 0 : 1;
const uint closest_vertical = abs( (2.0 * d1.z - d2.z) - depth) < abs( (2.0 * u1.z - u2.z) - depth) ? 0 : 1;
#endif
// Calculate the triangle, in a counter-clockwize order, to
// use based on the closest horizontal and vertical depths.
// h == 0.0 && v == 0.0: p1 = left, p2 = down
// h == 1.0 && v == 0.0: p1 = down, p2 = right
// h == 1.0 && v == 1.0: p1 = right, p2 = up
// h == 0.0 && v == 1.0: p1 = up, p2 = left
// Calculate the view space positions for the three points...
float3 P1;
float3 P2;
if (closest_vertical == 0)
{
P1 = closest_horizontal == 0 ? l1 : d1;
P2 = closest_horizontal == 0 ? d1 : r1;
}
else
{
P1 = closest_horizontal == 0 ? u1 : r1;
P2 = closest_horizontal == 0 ? l1 : u1;
}
// Use the cross product to calculate the normal...
return half3(normalize(cross(ReconstructViewPos(P2.xy, P2.z) - vpos, ReconstructViewPos(P1.xy, P1.z) - vpos)));
#endif
}
// For when we don't need to output the depth or view position
// Used in the blur passes
half3 SampleNormal(float2 uv)
{
#if defined(_SOURCE_DEPTH_NORMALS)
return half3(SampleSceneNormals(uv));
#else
float depth = SampleAndGetLinearEyeDepth(uv);
half3 vpos = ReconstructViewPos(uv, depth);
return ReconstructNormal(uv, depth, vpos);
#endif
}
void SampleDepthNormalView(float2 uv, out float depth, out half3 normal, out half3 vpos)
{
depth = SampleAndGetLinearEyeDepth(uv);
vpos = ReconstructViewPos(uv, depth);
#if defined(_SOURCE_DEPTH_NORMALS)
normal = half3(SampleSceneNormals(uv));
#else
normal = ReconstructNormal(uv, depth, vpos);
#endif
}
// Distance-based AO estimator based on Morgan 2011
// "Alchemy screen-space ambient obscurance algorithm"
// http://graphics.cs.williams.edu/papers/AlchemyHPG11/
half4 SSAO(Varyings input) : SV_Target
{
UNITY_SETUP_STEREO_EYE_INDEX_POST_VERTEX(input);
float2 uv = input.uv;
// Parameters used in coordinate conversion
half3x3 camTransform = (half3x3)_CameraViewProjections[unity_eyeIndex]; // camera viewProjection matrix
// Get the depth, normal and view position for this fragment
float depth_o;
half3 norm_o;
half3 vpos_o;
SampleDepthNormalView(uv, depth_o, norm_o, vpos_o);
// This was added to avoid a NVIDIA driver issue.
const half rcpSampleCount = half(rcp(SAMPLE_COUNT));
half ao = 0.0;
for (int s = 0; s < SAMPLE_COUNT; s++)
{
// Sample point
half3 v_s1 = PickSamplePoint(uv, s);
// Make it distributed between [0, _Radius]
v_s1 *= sqrt((half(s) + half(1.0)) * rcpSampleCount) * RADIUS;
v_s1 = faceforward(v_s1, -norm_o, v_s1);
half3 vpos_s1 = vpos_o + v_s1;
// Reproject the sample point
half3 spos_s1 = mul(camTransform, vpos_s1);
#if defined(_ORTHOGRAPHIC)
float2 uv_s1_01 = clamp((spos_s1.xy + float(1.0)) * float(0.5), float(0.0), float(1.0));
#else
float zdist = -dot(UNITY_MATRIX_V[2].xyz, vpos_s1);
float2 uv_s1_01 = clamp((spos_s1.xy * rcp(zdist) + float(1.0)) * float(0.5), float(0.0), float(1.0));
#endif
// Depth at the sample point
float depth_s1 = SampleAndGetLinearEyeDepth(uv_s1_01);
// Relative position of the sample point
half3 vpos_s2 = ReconstructViewPos(uv_s1_01, depth_s1);
half3 v_s2 = vpos_s2 - vpos_o;
// Estimate the obscurance value
half dotVal = dot(v_s2, norm_o);
#if defined(_ORTHOGRAPHIC)
dotVal -= half(2.0 * kBeta * depth_o);
#else
dotVal -= half(kBeta * depth_o);
#endif
half a1 = max(dotVal, half(0.0));
half a2 = dot(v_s2, v_s2) + kEpsilon;
ao += a1 * rcp(a2);
}
// Intensity normalization
ao *= RADIUS;
// Apply contrast
ao = PositivePow(ao * INTENSITY * rcpSampleCount, kContrast);
return PackAONormal(ao, norm_o);
}
// Geometry-aware separable bilateral filter
half4 Blur(float2 uv, float2 delta) : SV_Target
{
half4 p0 = (half4) SAMPLE_BASEMAP(uv );
half4 p1a = (half4) SAMPLE_BASEMAP(uv - delta * 1.3846153846);
half4 p1b = (half4) SAMPLE_BASEMAP(uv + delta * 1.3846153846);
half4 p2a = (half4) SAMPLE_BASEMAP(uv - delta * 3.2307692308);
half4 p2b = (half4) SAMPLE_BASEMAP(uv + delta * 3.2307692308);
#if defined(BLUR_SAMPLE_CENTER_NORMAL)
#if defined(_SOURCE_DEPTH_NORMALS)
half3 n0 = half3(SampleSceneNormals(uv));
#else
half3 n0 = SampleNormal(uv);
#endif
#else
half3 n0 = GetPackedNormal(p0);
#endif
half w0 = half(0.2270270270);
half w1a = CompareNormal(n0, GetPackedNormal(p1a)) * half(0.3162162162);
half w1b = CompareNormal(n0, GetPackedNormal(p1b)) * half(0.3162162162);
half w2a = CompareNormal(n0, GetPackedNormal(p2a)) * half(0.0702702703);
half w2b = CompareNormal(n0, GetPackedNormal(p2b)) * half(0.0702702703);
half s = half(0.0);
s += GetPackedAO(p0) * w0;
s += GetPackedAO(p1a) * w1a;
s += GetPackedAO(p1b) * w1b;
s += GetPackedAO(p2a) * w2a;
s += GetPackedAO(p2b) * w2b;
s *= rcp(w0 + w1a + w1b + w2a + w2b);
return PackAONormal(s, n0);
}
// Geometry-aware bilateral filter (single pass/small kernel)
half BlurSmall(float2 uv, float2 delta)
{
half4 p0 = (half4) SAMPLE_BASEMAP(uv );
half4 p1 = (half4) SAMPLE_BASEMAP(uv + float2(-delta.x, -delta.y));
half4 p2 = (half4) SAMPLE_BASEMAP(uv + float2( delta.x, -delta.y));
half4 p3 = (half4) SAMPLE_BASEMAP(uv + float2(-delta.x, delta.y));
half4 p4 = (half4) SAMPLE_BASEMAP(uv + float2( delta.x, delta.y));
half3 n0 = GetPackedNormal(p0);
half w0 = half(1.0);
half w1 = CompareNormal(n0, GetPackedNormal(p1));
half w2 = CompareNormal(n0, GetPackedNormal(p2));
half w3 = CompareNormal(n0, GetPackedNormal(p3));
half w4 = CompareNormal(n0, GetPackedNormal(p4));
half s = half(0.0);
s += GetPackedAO(p0) * w0;
s += GetPackedAO(p1) * w1;
s += GetPackedAO(p2) * w2;
s += GetPackedAO(p3) * w3;
s += GetPackedAO(p4) * w4;
return s *= rcp(w0 + w1 + w2 + w3 + w4);
}
half4 HorizontalBlur(Varyings input) : SV_Target
{
UNITY_SETUP_STEREO_EYE_INDEX_POST_VERTEX(input);
const float2 uv = input.uv;
const float2 delta = float2(_SourceSize.z, 0.0);
return Blur(uv, delta);
}
half4 VerticalBlur(Varyings input) : SV_Target
{
UNITY_SETUP_STEREO_EYE_INDEX_POST_VERTEX(input);
const float2 uv = input.uv;
const float2 delta = float2(0.0, _SourceSize.w * rcp(DOWNSAMPLE));
return Blur(uv, delta);
}
half4 FinalBlur(Varyings input) : SV_Target
{
UNITY_SETUP_STEREO_EYE_INDEX_POST_VERTEX(input);
const float2 uv = input.uv;
const float2 delta = _SourceSize.zw;
return half(1.0) - BlurSmall(uv, delta );
}
#endif //UNIVERSAL_SSAO_INCLUDED