Fix T39823: SSS scatter doesn't update volume stack, causing shading artifacts

Basically the title says it all, we need to update volume stack when doing ray
scatter for SSS. This leads to speed regressions in cases scene does have both
volume and SSS (performance in case there's no SSS or no volume should be the
same).

We might try optimizing kernel_path_subsurface_update_volume_stack() a bit by
either recording all intersections or using some more appropriate visibility
flags.

Reviewers: brecht, juicyfruit, dingto

Differential Revision: https://developer.blender.org/D795

2 files changed, 83 insertions, 3 deletions

diff --git a/intern/cycles/kernel/kernel_bake.h b/intern/cycles/kernel/kernel_bake.h
index dfbb49db7e6..822ad14039b 100644
--- a/intern/cycles/kernel/kernel_bake.h
+++ b/intern/cycles/kernel/kernel_bake.h
@@ -96,7 +96,7 @@ ccl_device void compute_light_pass(KernelGlobals *kg, ShaderData *sd, PathRadian
 		/* sample subsurface scattering */
 		if((is_combined || is_sss_sample) && (sd->flag & SD_BSSRDF)) {
 			/* when mixing BSSRDF and BSDF closures we should skip BSDF lighting if scattering was successful */
-			kernel_branched_path_subsurface_scatter(kg, sd, &L_sample, &state, &rng, throughput);
+			kernel_branched_path_subsurface_scatter(kg, sd, &L_sample, &state, &rng, &ray, throughput);
 		}
 #endif
 
diff --git a/intern/cycles/kernel/kernel_path.h b/intern/cycles/kernel/kernel_path.h
index 8e9fc743694..0515489f77f 100644
--- a/intern/cycles/kernel/kernel_path.h
+++ b/intern/cycles/kernel/kernel_path.h
@@ -359,6 +359,40 @@ ccl_device void kernel_branched_path_ao(KernelGlobals *kg, ShaderData *sd, PathR
 }
 
 #ifdef __SUBSURFACE__
+
+#  ifdef __VOLUME__
+ccl_device void kernel_path_subsurface_update_volume_stack(KernelGlobals *kg,
+                                                           Ray *ray,
+                                                           VolumeStack *stack)
+{
+	kernel_assert(kernel_data.integrator.use_volumes);
+
+	Ray volume_ray = *ray;
+	Intersection isect;
+	const float3 Pend = volume_ray.P + volume_ray.D*volume_ray.t;
+
+	while(
+		scene_intersect(kg, &volume_ray, PATH_RAY_ALL_VISIBILITY,
+		                &isect, NULL, 0.0f, 0.0f)) {
+		ShaderData sd;
+		shader_setup_from_ray(kg, &sd, &isect, &volume_ray, 0, 0);
+		kernel_volume_stack_enter_exit(kg, &sd, stack);
+
+		/* Move ray forward. */
+		volume_ray.P = ray_offset(sd.P, -sd.Ng);
+		volume_ray.D = normalize_len(Pend - volume_ray.P,
+		                             &volume_ray.t);
+
+		/* TODO(sergey): Find a faster way detecting that ray_offset moved
+		 * us pass through the end point.
+		 */
+		if(dot(ray->D, volume_ray.D) < 0.0f) {
+			break;
+		}
+	}
+}
+#  endif
+
 ccl_device bool kernel_path_subsurface_scatter(KernelGlobals *kg, ShaderData *sd, PathRadiance *L, PathState *state, RNG *rng, Ray *ray, float3 *throughput)
 {
 	float bssrdf_probability;
@@ -375,6 +409,9 @@ ccl_device bool kernel_path_subsurface_scatter(KernelGlobals *kg, ShaderData *sd
 		float bssrdf_u, bssrdf_v;
 		path_state_rng_2D(kg, rng, state, PRNG_BSDF_U, &bssrdf_u, &bssrdf_v);
 		int num_hits = subsurface_scatter_multi_step(kg, sd, bssrdf_sd, state->flag, sc, &lcg_state, bssrdf_u, bssrdf_v, false);
+#ifdef __VOLUME__
+		Ray volume_ray = *ray;
+#endif
 
 		/* compute lighting with the BSDF closure */
 		for(int hit = 0; hit < num_hits; hit++) {
@@ -392,6 +429,22 @@ ccl_device bool kernel_path_subsurface_scatter(KernelGlobals *kg, ShaderData *sd
 				hit_state.ray_t = 0.0f;
 #endif
 
+#ifdef __VOLUME__
+				if(kernel_data.integrator.use_volumes) {
+					/* Setup ray from previous surface point to the new one. */
+					volume_ray.D = normalize_len(hit_ray.P - volume_ray.P,
+					                             &volume_ray.t);
+
+					kernel_path_subsurface_update_volume_stack(
+					    kg,
+					    &volume_ray,
+					    hit_state.volume_stack);
+
+					/* Move volume ray forward. */
+					volume_ray.P = hit_ray.P;
+				}
+#endif
+
 				kernel_path_indirect(kg, rng, hit_ray, tp, state->num_samples, hit_state, L);
 
 				/* for render passes, sum and reset indirect light pass variables
@@ -718,7 +771,13 @@ ccl_device_noinline void kernel_branched_path_surface_indirect_light(KernelGloba
 }
 
 #ifdef __SUBSURFACE__
-ccl_device void kernel_branched_path_subsurface_scatter(KernelGlobals *kg, ShaderData *sd, PathRadiance *L, PathState *state, RNG *rng, float3 throughput)
+ccl_device void kernel_branched_path_subsurface_scatter(KernelGlobals *kg,
+                                                        ShaderData *sd,
+                                                        PathRadiance *L,
+                                                        PathState *state,
+                                                        RNG *rng,
+                                                        Ray *ray,
+                                                        float3 throughput)
 {
 	for(int i = 0; i< sd->num_closure; i++) {
 		ShaderClosure *sc = &sd->closure[i];
@@ -741,6 +800,9 @@ ccl_device void kernel_branched_path_subsurface_scatter(KernelGlobals *kg, Shade
 			float bssrdf_u, bssrdf_v;
 			path_branched_rng_2D(kg, &bssrdf_rng, state, j, num_samples, PRNG_BSDF_U, &bssrdf_u, &bssrdf_v);
 			int num_hits = subsurface_scatter_multi_step(kg, sd, bssrdf_sd, state->flag, sc, &lcg_state, bssrdf_u, bssrdf_v, true);
+#  ifdef __VOLUME__
+			Ray volume_ray = *ray;
+#  endif
 
 			/* compute lighting with the BSDF closure */
 			for(int hit = 0; hit < num_hits; hit++) {
@@ -748,6 +810,23 @@ ccl_device void kernel_branched_path_subsurface_scatter(KernelGlobals *kg, Shade
 
 				path_state_branch(&hit_state, j, num_samples);
 
+#ifdef __VOLUME__
+				if(kernel_data.integrator.use_volumes) {
+					/* Setup ray from previous surface point to the new one. */
+					float3 P = ray_offset(bssrdf_sd[hit].P, -bssrdf_sd[hit].Ng);
+					volume_ray.D = normalize_len(P - volume_ray.P,
+					                             &volume_ray.t);
+
+					kernel_path_subsurface_update_volume_stack(
+					    kg,
+					    &volume_ray,
+					    hit_state.volume_stack);
+
+					/* Move volume ray forward. */
+					volume_ray.P = P;
+				}
+#endif
+
 #if defined(__EMISSION__) && defined(__BRANCHED_PATH__)
 				/* direct light */
 				if(kernel_data.integrator.use_direct_light) {
@@ -1012,7 +1091,8 @@ ccl_device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, in
 #ifdef __SUBSURFACE__
 		/* bssrdf scatter to a different location on the same object */
 		if(sd.flag & SD_BSSRDF) {
-			kernel_branched_path_subsurface_scatter(kg, &sd, &L, &state, rng, throughput);
+			kernel_branched_path_subsurface_scatter(kg, &sd, &L, &state,
+			                                        rng, &ray, throughput);
 		}
 #endif