Cycles: bake transparent shadows for hair

These transparent shadows can be expansive to evaluate. Especially on the
GPU they can lead to poor occupancy when only some pixels require many kernel
launches to trace and evaluate many layers of transparency.

Baked transparency allows tracing a single ray in many cases by accumulating
the throughput directly in the intersection program without recording hits
or evaluating shaders. Transparency is baked at curve vertices and
interpolated, for most shaders this will look practically the same as actual
shader evaluation.

Fixes T91428, performance regression with spring demo file due to transparent
hair, and makes it render significantly faster than Blender 2.93.

Differential Revision: https://developer.blender.org/D12880

1 files changed, 75 insertions, 0 deletions

diff --git a/intern/cycles/kernel/geom/geom_shader_data.h b/intern/cycles/kernel/geom/geom_shader_data.h
index e6a5b8f7923..46bda2b656c 100644
--- a/intern/cycles/kernel/geom/geom_shader_data.h
+++ b/intern/cycles/kernel/geom/geom_shader_data.h
@@ -279,6 +279,81 @@ ccl_device void shader_setup_from_displace(KernelGlobals kg,
       LAMP_NONE);
 }
 
+/* ShaderData setup for point on curve. */
+
+ccl_device void shader_setup_from_curve(KernelGlobals kg,
+                                        ccl_private ShaderData *ccl_restrict sd,
+                                        int object,
+                                        int prim,
+                                        int segment,
+                                        float u)
+{
+  /* Primitive */
+  sd->type = PRIMITIVE_PACK_SEGMENT(PRIMITIVE_CURVE_THICK, segment);
+  sd->lamp = LAMP_NONE;
+  sd->prim = prim;
+  sd->u = u;
+  sd->v = 0.0f;
+  sd->time = 0.5f;
+  sd->ray_length = 0.0f;
+
+  /* Shader */
+  sd->shader = kernel_tex_fetch(__curves, prim).shader_id;
+  sd->flag = kernel_tex_fetch(__shaders, (sd->shader & SHADER_MASK)).flags;
+
+  /* Object */
+  sd->object = object;
+  sd->object_flag = kernel_tex_fetch(__object_flag, sd->object);
+#ifdef __OBJECT_MOTION__
+  shader_setup_object_transforms(kg, sd, sd->time);
+#endif
+
+  /* Get control points. */
+  KernelCurve kcurve = kernel_tex_fetch(__curves, prim);
+
+  int k0 = kcurve.first_key + PRIMITIVE_UNPACK_SEGMENT(sd->type);
+  int k1 = k0 + 1;
+  int ka = max(k0 - 1, kcurve.first_key);
+  int kb = min(k1 + 1, kcurve.first_key + kcurve.num_keys - 1);
+
+  float4 P_curve[4];
+
+  P_curve[0] = kernel_tex_fetch(__curve_keys, ka);
+  P_curve[1] = kernel_tex_fetch(__curve_keys, k0);
+  P_curve[2] = kernel_tex_fetch(__curve_keys, k1);
+  P_curve[3] = kernel_tex_fetch(__curve_keys, kb);
+
+  /* Interpolate position and tangent. */
+  sd->P = float4_to_float3(catmull_rom_basis_derivative(P_curve, sd->u));
+#ifdef __DPDU__
+  sd->dPdu = float4_to_float3(catmull_rom_basis_derivative(P_curve, sd->u));
+#endif
+
+  /* Transform into world space */
+  if (!(sd->object_flag & SD_OBJECT_TRANSFORM_APPLIED)) {
+    object_position_transform_auto(kg, sd, &sd->P);
+#ifdef __DPDU__
+    object_dir_transform_auto(kg, sd, &sd->dPdu);
+#endif
+  }
+
+  /* No view direction, normals or bitangent. */
+  sd->I = zero_float3();
+  sd->N = zero_float3();
+  sd->Ng = zero_float3();
+#ifdef __DPDU__
+  sd->dPdv = zero_float3();
+#endif
+
+  /* No ray differentials currently. */
+#ifdef __RAY_DIFFERENTIALS__
+  sd->dP = differential3_zero();
+  sd->dI = differential3_zero();
+  sd->du = differential_zero();
+  sd->dv = differential_zero();
+#endif
+}
+
 /* ShaderData setup from ray into background */
 
 ccl_device_inline void shader_setup_from_background(KernelGlobals kg,