2021-08-18 23:22:26 +00:00
|
|
|
#version 330 core
|
|
|
|
|
|
|
|
uniform sampler2D dfg_lut;
|
|
|
|
uniform samplerCube prefiltered_envmap;
|
|
|
|
uniform sampler2DShadow shadow_tex;
|
|
|
|
|
|
|
|
uniform mat4 fg_LightMatrix_csm0;
|
|
|
|
uniform mat4 fg_LightMatrix_csm1;
|
|
|
|
uniform mat4 fg_LightMatrix_csm2;
|
|
|
|
uniform mat4 fg_LightMatrix_csm3;
|
|
|
|
|
|
|
|
// Shadow mapping constants
|
|
|
|
const int sun_atlas_size = 8192;
|
|
|
|
const float DEPTH_BIAS = 2.0;
|
|
|
|
const float BAND_SIZE = 0.1;
|
|
|
|
const vec2 BAND_BOTTOM_LEFT = vec2(BAND_SIZE);
|
|
|
|
const vec2 BAND_TOP_RIGHT = vec2(1.0 - BAND_SIZE);
|
|
|
|
// Ideally these should be passed as an uniform, but we don't support uniform
|
|
|
|
// arrays yet
|
|
|
|
const vec2 uv_shifts[4] = vec2[4](
|
|
|
|
vec2(0.0, 0.0), vec2(0.5, 0.0),
|
|
|
|
vec2(0.0, 0.5), vec2(0.5, 0.5));
|
|
|
|
const vec2 uv_factor = vec2(0.5, 0.5);
|
|
|
|
|
|
|
|
// BRDF constants
|
|
|
|
const float PI = 3.14159265359;
|
|
|
|
const float RECIPROCAL_PI = 0.31830988618;
|
|
|
|
const float DIELECTRIC_SPECULAR = 0.04;
|
|
|
|
const float MAX_PREFILTERED_LOD = 4.0;
|
|
|
|
|
|
|
|
//------------------------------------------------------------------------------
|
|
|
|
// Shadow mapping related stuff
|
|
|
|
|
|
|
|
float sampleOffset(vec4 pos, vec2 offset, vec2 invTexelSize)
|
|
|
|
{
|
|
|
|
return texture(
|
|
|
|
shadow_tex, vec3(
|
|
|
|
pos.xy + offset * invTexelSize,
|
|
|
|
pos.z - DEPTH_BIAS * invTexelSize));
|
|
|
|
}
|
|
|
|
|
|
|
|
// OptimizedPCF from https://github.com/TheRealMJP/Shadows
|
|
|
|
// Original by Ignacio Castaño for The Witness
|
|
|
|
// Released under The MIT License
|
|
|
|
float sampleOptimizedPCF(vec4 pos)
|
|
|
|
{
|
|
|
|
vec2 invTexSize = vec2(1.0 / float(sun_atlas_size));
|
|
|
|
|
|
|
|
vec2 uv = pos.xy * sun_atlas_size;
|
|
|
|
vec2 base_uv = floor(uv + 0.5);
|
|
|
|
float s = (uv.x + 0.5 - base_uv.x);
|
|
|
|
float t = (uv.y + 0.5 - base_uv.y);
|
|
|
|
base_uv -= vec2(0.5);
|
|
|
|
base_uv *= invTexSize;
|
|
|
|
pos.xy = base_uv.xy;
|
|
|
|
|
|
|
|
float sum = 0.0;
|
|
|
|
|
|
|
|
float uw0 = (4.0 - 3.0 * s);
|
|
|
|
float uw1 = 7.0;
|
|
|
|
float uw2 = (1.0 + 3.0 * s);
|
|
|
|
|
|
|
|
float u0 = (3.0 - 2.0 * s) / uw0 - 2.0;
|
|
|
|
float u1 = (3.0 + s) / uw1;
|
|
|
|
float u2 = s / uw2 + 2.0;
|
|
|
|
|
|
|
|
float vw0 = (4.0 - 3.0 * t);
|
|
|
|
float vw1 = 7.0;
|
|
|
|
float vw2 = (1.0 + 3.0 * t);
|
|
|
|
|
|
|
|
float v0 = (3.0 - 2.0 * t) / vw0 - 2.0;
|
|
|
|
float v1 = (3.0 + t) / vw1;
|
|
|
|
float v2 = t / vw2 + 2.0;
|
|
|
|
|
|
|
|
sum += uw0 * vw0 * sampleOffset(pos, vec2(u0, v0), invTexSize);
|
|
|
|
sum += uw1 * vw0 * sampleOffset(pos, vec2(u1, v0), invTexSize);
|
|
|
|
sum += uw2 * vw0 * sampleOffset(pos, vec2(u2, v0), invTexSize);
|
|
|
|
|
|
|
|
sum += uw0 * vw1 * sampleOffset(pos, vec2(u0, v1), invTexSize);
|
|
|
|
sum += uw1 * vw1 * sampleOffset(pos, vec2(u1, v1), invTexSize);
|
|
|
|
sum += uw2 * vw1 * sampleOffset(pos, vec2(u2, v1), invTexSize);
|
|
|
|
|
|
|
|
sum += uw0 * vw2 * sampleOffset(pos, vec2(u0, v2), invTexSize);
|
|
|
|
sum += uw1 * vw2 * sampleOffset(pos, vec2(u1, v2), invTexSize);
|
|
|
|
sum += uw2 * vw2 * sampleOffset(pos, vec2(u2, v2), invTexSize);
|
|
|
|
|
|
|
|
return sum / 144.0;
|
|
|
|
}
|
|
|
|
|
|
|
|
float sampleCascade(vec4 p, vec2 shift)
|
|
|
|
{
|
|
|
|
vec4 pos = p;
|
|
|
|
pos.xy *= uv_factor;
|
|
|
|
pos.xy += shift;
|
|
|
|
return sampleOptimizedPCF(pos);
|
|
|
|
}
|
|
|
|
|
|
|
|
float sampleAndBlendBand(vec4 p1, vec4 p2, vec2 s1, vec2 s2)
|
|
|
|
{
|
|
|
|
vec2 s = smoothstep(vec2(0.0), BAND_BOTTOM_LEFT, p1.xy)
|
|
|
|
- smoothstep(BAND_TOP_RIGHT, vec2(1.0), p1.xy);
|
|
|
|
float blend = 1.0 - s.x * s.y;
|
|
|
|
return mix(sampleCascade(p1, s1),
|
|
|
|
sampleCascade(p2, s2),
|
|
|
|
blend);
|
|
|
|
}
|
|
|
|
|
|
|
|
bool checkWithinBounds(vec2 coords, vec2 bottomLeft, vec2 topRight)
|
|
|
|
{
|
|
|
|
vec2 r = step(bottomLeft, coords) - step(topRight, coords);
|
|
|
|
return bool(r.x * r.y);
|
|
|
|
}
|
|
|
|
|
|
|
|
bool isInsideCascade(vec4 p)
|
|
|
|
{
|
|
|
|
return checkWithinBounds(p.xy, vec2(0.0), vec2(1.0)) && ((p.z / p.w) <= 1.0);
|
|
|
|
}
|
|
|
|
|
|
|
|
bool isInsideBand(vec4 p)
|
|
|
|
{
|
|
|
|
return !checkWithinBounds(p.xy, BAND_BOTTOM_LEFT, BAND_TOP_RIGHT);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Get the light space position of point p.
|
|
|
|
* Both p and n must be in view space. The light matrix is also assumed to
|
|
|
|
* transform from view space to light space.
|
|
|
|
*/
|
|
|
|
vec4 getLightSpacePosition(vec3 p, vec3 n, float NdotL, float bias,
|
|
|
|
mat4 lightMatrix)
|
|
|
|
{
|
|
|
|
float sinTheta = sqrt(1.0 - NdotL * NdotL);
|
|
|
|
vec3 offset = p + n * (sinTheta * bias);
|
|
|
|
return lightMatrix * vec4(offset, 1.0);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Get shadowing factor for a given position. 1.0 corresponds to a fragment
|
|
|
|
* being completely lit, and 0.0 to a fragment being completely in shadow.
|
|
|
|
* Both p and n must be in view space.
|
|
|
|
*/
|
|
|
|
float getShadowing(vec3 p, vec3 n, float NdotL)
|
|
|
|
{
|
|
|
|
// Ignore fragments that don't face the light
|
|
|
|
if (NdotL <= 0.0)
|
|
|
|
return 0.0;
|
|
|
|
|
|
|
|
float shadow = 1.0;
|
|
|
|
|
|
|
|
vec4 lightSpacePos[4];
|
|
|
|
lightSpacePos[0] = getLightSpacePosition(p, n, NdotL, 0.05, fg_LightMatrix_csm0);
|
2021-08-19 16:20:37 +00:00
|
|
|
lightSpacePos[1] = getLightSpacePosition(p, n, NdotL, 0.1, fg_LightMatrix_csm1);
|
|
|
|
lightSpacePos[2] = getLightSpacePosition(p, n, NdotL, 0.5, fg_LightMatrix_csm2);
|
|
|
|
lightSpacePos[3] = getLightSpacePosition(p, n, NdotL, 1.0, fg_LightMatrix_csm3);
|
2021-08-18 23:22:26 +00:00
|
|
|
|
|
|
|
for (int i = 0; i < 4; ++i) {
|
|
|
|
// Map-based cascade selection
|
|
|
|
// We test if we are inside the cascade bounds to find the tightest
|
|
|
|
// map that contains the fragment.
|
|
|
|
if (isInsideCascade(lightSpacePos[i])) {
|
|
|
|
if (isInsideBand(lightSpacePos[i]) && ((i+1) < 4)) {
|
|
|
|
// Blend between cascades if the fragment is near the
|
|
|
|
// next cascade to avoid abrupt transitions.
|
|
|
|
shadow = clamp(sampleAndBlendBand(lightSpacePos[i],
|
|
|
|
lightSpacePos[i+1],
|
|
|
|
uv_shifts[i],
|
|
|
|
uv_shifts[i+1]),
|
|
|
|
0.0, 1.0);
|
|
|
|
} else {
|
|
|
|
// We are far away from the borders of the cascade, so
|
|
|
|
// we skip the blending to avoid the performance cost
|
|
|
|
// of sampling the shadow map twice.
|
|
|
|
shadow = clamp(sampleCascade(lightSpacePos[i], uv_shifts[i]),
|
|
|
|
0.0, 1.0);
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return shadow;
|
|
|
|
}
|
|
|
|
|
|
|
|
//------------------------------------------------------------------------------
|
|
|
|
// BRDF utility functions
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Fresnel term with included roughness to get a pleasant visual result.
|
|
|
|
* See https://seblagarde.wordpress.com/2011/08/17/hello-world/
|
|
|
|
*/
|
|
|
|
vec3 F_SchlickRoughness(float NdotV, vec3 F0, float r)
|
|
|
|
{
|
|
|
|
return F0 + (max(vec3(1.0 - r), F0) - F0) * pow(max(1.0 - NdotV, 0.0), 5.0);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Fresnel (specular F)
|
|
|
|
* Schlick's approximation for the Cook-Torrance BRDF.
|
|
|
|
*/
|
|
|
|
vec3 F_Schlick(float VdotH, vec3 F0)
|
|
|
|
{
|
|
|
|
return F0 + (vec3(1.0) - F0) * pow(clamp(1.0 - VdotH, 0.0, 1.0), 5.0);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Normal distribution function (NDF) (specular D)
|
|
|
|
* Trowbridge-Reitz/GGX microfacet distribution. Includes Disney's
|
|
|
|
* reparametrization of a=roughness*roughness
|
|
|
|
*/
|
|
|
|
float D_GGX(float NdotH, float a2)
|
|
|
|
{
|
|
|
|
float f = (NdotH * a2 - NdotH) * NdotH + 1.0;
|
|
|
|
return a2 / (PI * f * f);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Geometric attenuation (specular G)
|
|
|
|
* Smith-GGX formulation.
|
|
|
|
*/
|
|
|
|
float G_SmithGGX(float NdotV, float NdotL, float a2)
|
|
|
|
{
|
|
|
|
float attV = 2.0 * NdotV / (NdotV + sqrt(a2 + (1.0 - a2) * (NdotV * NdotV)));
|
|
|
|
float attL = 2.0 * NdotL / (NdotL + sqrt(a2 + (1.0 - a2) * (NdotL * NdotL)));
|
|
|
|
return attV * attL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Basic Lambertian diffuse BRDF
|
|
|
|
*/
|
|
|
|
vec3 Fd_Lambert(vec3 c_diff)
|
|
|
|
{
|
|
|
|
return c_diff * RECIPROCAL_PI;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Get the fresnel reflectance at 0 degrees (light hitting the surface
|
|
|
|
* perpendicularly).
|
|
|
|
*/
|
|
|
|
vec3 getF0Reflectance(vec3 baseColor, float metallic)
|
|
|
|
{
|
|
|
|
return mix(vec3(DIELECTRIC_SPECULAR), baseColor, metallic);
|
|
|
|
}
|
|
|
|
|
|
|
|
//------------------------------------------------------------------------------
|
|
|
|
// IBL evaluation
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Indirect diffuse irradiance
|
|
|
|
* To get better results we should be precomputing the irradiance into a cubemap
|
|
|
|
* or calculating spherical harmonics coefficients on the CPU.
|
|
|
|
* Sampling the roughness=1 mipmap level of the prefiltered specular map
|
|
|
|
* works too. :)
|
|
|
|
*/
|
|
|
|
vec3 evaluateDiffuseIrradianceIBL(vec3 n)
|
|
|
|
{
|
|
|
|
int roughnessOneLevel = int(MAX_PREFILTERED_LOD);
|
|
|
|
ivec2 s = textureSize(prefiltered_envmap, roughnessOneLevel);
|
|
|
|
float du = 1.0 / float(s.x);
|
|
|
|
float dv = 1.0 / float(s.y);
|
|
|
|
vec3 m0 = normalize(cross(n, vec3(0.0, 1.0, 0.0)));
|
|
|
|
vec3 m1 = cross(m0, n);
|
|
|
|
vec3 m0du = m0 * du;
|
|
|
|
vec3 m1dv = m1 * dv;
|
|
|
|
|
|
|
|
vec3 c;
|
|
|
|
c = textureLod(prefiltered_envmap, n - m0du - m1dv, roughnessOneLevel).rgb;
|
|
|
|
c += textureLod(prefiltered_envmap, n + m0du - m1dv, roughnessOneLevel).rgb;
|
|
|
|
c += textureLod(prefiltered_envmap, n + m0du + m1dv, roughnessOneLevel).rgb;
|
|
|
|
c += textureLod(prefiltered_envmap, n - m0du + m1dv, roughnessOneLevel).rgb;
|
|
|
|
return c * 0.25;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Indirect specular (ambient specular)
|
|
|
|
* Sample from the prefiltered environment map.
|
|
|
|
*/
|
|
|
|
vec3 evaluateSpecularIBL(float NdotV, vec3 reflected, float roughness, vec3 f)
|
|
|
|
{
|
|
|
|
vec3 prefilteredColor = textureLod(prefiltered_envmap,
|
|
|
|
reflected,
|
|
|
|
roughness * MAX_PREFILTERED_LOD).rgb;
|
|
|
|
vec2 envBRDF = texture(dfg_lut, vec2(NdotV, roughness)).rg;
|
|
|
|
return prefilteredColor * (f * envBRDF.x + envBRDF.y);
|
|
|
|
}
|
|
|
|
|
|
|
|
vec3 evaluateIBL(
|
|
|
|
vec3 baseColor,
|
|
|
|
float metallic,
|
|
|
|
float roughness,
|
|
|
|
vec3 f0, // Use getF0Reflectance() to obtain this
|
|
|
|
float occlusion,
|
|
|
|
vec3 nWorldSpace, // Normal in world space
|
|
|
|
float NdotV, // Must be positive and non-zero
|
|
|
|
vec3 reflected // Reflected vector in world space: reflect(-v, n)
|
|
|
|
)
|
|
|
|
{
|
|
|
|
vec3 f = F_SchlickRoughness(NdotV, f0, roughness);
|
|
|
|
|
|
|
|
vec3 specular = evaluateSpecularIBL(NdotV, reflected, roughness, f);
|
|
|
|
vec3 diffuse = evaluateDiffuseIrradianceIBL(nWorldSpace) * baseColor
|
|
|
|
* (vec3(1.0) - f) * (1.0 - metallic);
|
|
|
|
|
|
|
|
return (diffuse + specular) * occlusion;
|
|
|
|
}
|
|
|
|
|
|
|
|
//------------------------------------------------------------------------------
|
|
|
|
// Analytical light source evaluation
|
|
|
|
|
|
|
|
vec3 evaluateLight(
|
|
|
|
vec3 baseColor,
|
|
|
|
float metallic,
|
|
|
|
float roughness,
|
|
|
|
float clearcoat,
|
|
|
|
float clearcoatRoughness,
|
|
|
|
vec3 f0, // Use getF0Reflectance() to obtain this
|
|
|
|
vec3 intensity,
|
|
|
|
float occlusion,
|
|
|
|
vec3 n,
|
|
|
|
vec3 l,
|
|
|
|
vec3 v,
|
|
|
|
float NdotL, // Must not be clamped to [0,1]
|
|
|
|
float NdotV // Must be positive and non-zero
|
|
|
|
)
|
|
|
|
{
|
|
|
|
// Skip fragments that are completely occluded or that are not facing the light
|
|
|
|
if (occlusion <= 0.0 || NdotL <= 0.0)
|
|
|
|
return vec3(0.0);
|
|
|
|
|
|
|
|
NdotL = clamp(NdotL, 0.001, 1.0);
|
|
|
|
|
|
|
|
vec3 h = normalize(v + l);
|
|
|
|
float NdotH = clamp(dot(n, h), 0.0, 1.0);
|
|
|
|
float VdotH = clamp(dot(v, h), 0.0, 1.0);
|
|
|
|
|
|
|
|
vec3 c_diff = mix(baseColor * (1.0 - DIELECTRIC_SPECULAR), vec3(0.0), metallic);
|
|
|
|
|
|
|
|
float a = roughness * roughness;
|
|
|
|
float a2 = a * a;
|
|
|
|
|
|
|
|
vec3 F = F_Schlick(VdotH, f0);
|
|
|
|
float D = D_GGX(NdotH, a2);
|
|
|
|
float G = G_SmithGGX(NdotV, NdotL, a2);
|
|
|
|
|
|
|
|
// Diffuse term
|
|
|
|
// Lambertian diffuse model
|
|
|
|
vec3 diffuse = (vec3(1.0) - F) * Fd_Lambert(c_diff);
|
|
|
|
// Specular term
|
|
|
|
// Cook-Torrance specular microfacet model
|
|
|
|
vec3 specular = (F * D * G) / (4.0 * NdotV * NdotL);
|
|
|
|
|
|
|
|
vec3 material = diffuse + specular;
|
|
|
|
|
|
|
|
vec3 color = material * intensity * occlusion;
|
|
|
|
return color;
|
|
|
|
}
|