Eevee: Add support for new Hair geometry system.

This now can shade actual poly strips that mimics cylinders.
This makes hair coverage exact compared to the line method and result in
smoother fading hair.

This does make the sampling a bit more exact but needs more samples to
converge properly.

1 files changed, 30 insertions, 12 deletions

diff --git a/source/blender/draw/engines/eevee/shaders/lit_surface_frag.glsl b/source/blender/draw/engines/eevee/shaders/lit_surface_frag.glsl
index 737a2977146..accecaacbde 100644
--- a/source/blender/draw/engines/eevee/shaders/lit_surface_frag.glsl
+++ b/source/blender/draw/engines/eevee/shaders/lit_surface_frag.glsl
@@ -21,6 +21,15 @@ in vec3 worldNormal;
 in vec3 viewNormal;
 #endif
 
+#ifdef HAIR_SHADER
+in vec3 hairTangent; /* world space */
+in float hairThickTime;
+in float hairThickness;
+in float hairTime;
+
+uniform int hairThicknessRes = 1;
+#endif
+
 #endif /* LIT_SURFACE_UNIFORM */
 
 /** AUTO CONFIG
@@ -169,21 +178,30 @@ void CLOSURE_NAME(
 	vec4 rand = texelFetch(utilTex, ivec3(ivec2(gl_FragCoord.xy) % LUT_SIZE, 2.0), 0);
 
 #ifdef HAIR_SHADER
-	/* Random normal distribution on the hair surface. */
-	vec3 T = normalize(worldNormal); /* meh, TODO fix worldNormal misnaming. */
-	vec3 B = normalize(cross(V, T));
-	N = cross(T, B); /* Normal facing view */
-	/* We want a cosine distribution. */
-	float cos_theta = rand.x * 2.0 - 1.0;
-	float sin_theta = sqrt(max(0.0, 1.0f - cos_theta*cos_theta));;
-	N = N * sin_theta + B * cos_theta;
+	if (hairThicknessRes == 1) {
+		/* Random normal distribution on the hair surface. */
+		vec3 T = normalize(worldNormal); /* meh, TODO fix worldNormal misnaming. */
+		vec3 B = normalize(cross(V, T));
+		N = cross(T, B); /* Normal facing view */
+		/* We want a cosine distribution. */
+		float cos_theta = rand.x * 2.0 - 1.0;
+		float sin_theta = sqrt(max(0.0, 1.0f - cos_theta*cos_theta));;
+		N = N * sin_theta + B * cos_theta;
 
 #  ifdef CLOSURE_GLOSSY
-	/* Hair random normal does not work with SSR :(.
-	 * It just create self reflection feedback (which is beautifful btw)
-	 * but not correct. */
-	ssr_id = NO_SSR; /* Force bypass */
+		/* Hair random normal does not work with SSR :(.
+		 * It just create self reflection feedback (which is beautifful btw)
+		 * but not correct. */
+		ssr_id = NO_SSR; /* Force bypass */
 #  endif
+	}
+	else {
+		vec3 T = normalize(cross(hairTangent, worldNormal));
+		/* We want a cosine distribution. */
+		float cos_theta = hairThickTime / hairThickness;
+		float sin_theta = sqrt(max(0.0, 1.0f - cos_theta*cos_theta));;
+		N = normalize(hairTangent * cos_theta + T * sin_theta);
+	}
 #endif
 
 	/* ---------------------------------------------------------------- */