Fix T54105: random walk SSS missing in branched indirect paths.

Unify the path and branched path indirect SSS code. No performance impact
found on CUDA, for AMD split kernel the extra code was already there.

3 files changed, 22 insertions, 42 deletions

diff --git a/intern/cycles/kernel/split/kernel_indirect_subsurface.h b/intern/cycles/kernel/split/kernel_indirect_subsurface.h
index e9fe5552e8c..b65f3d1a940 100644
--- a/intern/cycles/kernel/split/kernel_indirect_subsurface.h
+++ b/intern/cycles/kernel/split/kernel_indirect_subsurface.h
@@ -49,28 +49,22 @@ ccl_device void kernel_indirect_subsurface(KernelGlobals *kg)
 	ccl_global Ray *ray = &kernel_split_state.ray[ray_index];
 	ccl_global float3 *throughput = &kernel_split_state.throughput[ray_index];
 
-#ifdef __BRANCHED_PATH__
-	if(!kernel_data.integrator.branched) {
-#endif
-		if(IS_STATE(ray_state, ray_index, RAY_UPDATE_BUFFER)) {
-			ccl_addr_space SubsurfaceIndirectRays *ss_indirect = &kernel_split_state.ss_rays[ray_index];
+	if(IS_STATE(ray_state, ray_index, RAY_UPDATE_BUFFER)) {
+		ccl_addr_space SubsurfaceIndirectRays *ss_indirect = &kernel_split_state.ss_rays[ray_index];
 
-			/* Trace indirect subsurface rays by restarting the loop. this uses less
-			 * stack memory than invoking kernel_path_indirect.
-			 */
-			if(ss_indirect->num_rays) {
-				kernel_path_subsurface_setup_indirect(kg,
-					                                  ss_indirect,
-					                                  state,
-					                                  ray,
-					                                  L,
-					                                  throughput);
-				ASSIGN_RAY_STATE(ray_state, ray_index, RAY_REGENERATED);
-			}
+		/* Trace indirect subsurface rays by restarting the loop. this uses less
+		 * stack memory than invoking kernel_path_indirect.
+		 */
+		if(ss_indirect->num_rays) {
+			kernel_path_subsurface_setup_indirect(kg,
+												  ss_indirect,
+												  state,
+												  ray,
+												  L,
+												  throughput);
+			ASSIGN_RAY_STATE(ray_state, ray_index, RAY_REGENERATED);
 		}
-#ifdef __BRANCHED_PATH__
 	}
-#endif
 
 #endif  /* __SUBSURFACE__ */
 
diff --git a/intern/cycles/kernel/split/kernel_split_common.h b/intern/cycles/kernel/split/kernel_split_common.h
index 21886ee62ee..b52e7bddc82 100644
--- a/intern/cycles/kernel/split/kernel_split_common.h
+++ b/intern/cycles/kernel/split/kernel_split_common.h
@@ -59,7 +59,12 @@ ccl_device_inline void kernel_split_path_end(KernelGlobals *kg, int ray_index)
 	ccl_global char *ray_state = kernel_split_state.ray_state;
 
 #ifdef __BRANCHED_PATH__
-	if(IS_FLAG(ray_state, ray_index, RAY_BRANCHED_INDIRECT_SHARED)) {
+	ccl_addr_space SubsurfaceIndirectRays *ss_indirect = &kernel_split_state.ss_rays[ray_index];
+
+	if(ss_indirect->num_rays) {
+		ASSIGN_RAY_STATE(ray_state, ray_index, RAY_UPDATE_BUFFER);
+	}
+	else if(IS_FLAG(ray_state, ray_index, RAY_BRANCHED_INDIRECT_SHARED)) {
 		int orig_ray = kernel_split_state.branched_state[ray_index].original_ray;
 
 		PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
diff --git a/intern/cycles/kernel/split/kernel_subsurface_scatter.h b/intern/cycles/kernel/split/kernel_subsurface_scatter.h
index e50d63ea3bc..af0303d8608 100644
--- a/intern/cycles/kernel/split/kernel_subsurface_scatter.h
+++ b/intern/cycles/kernel/split/kernel_subsurface_scatter.h
@@ -228,7 +228,9 @@ ccl_device void kernel_subsurface_scatter(KernelGlobals *kg)
 		if(sd->flag & SD_BSSRDF) {
 
 #ifdef __BRANCHED_PATH__
-			if(!kernel_data.integrator.branched) {
+			if(!kernel_data.integrator.branched ||
+			   IS_FLAG(ray_state, ray_index, RAY_BRANCHED_INDIRECT))
+			{
 #endif
 				if(kernel_path_subsurface_scatter(kg,
 				                                  sd,
@@ -243,27 +245,6 @@ ccl_device void kernel_subsurface_scatter(KernelGlobals *kg)
 				}
 #ifdef __BRANCHED_PATH__
 			}
-			else if(IS_FLAG(ray_state, ray_index, RAY_BRANCHED_INDIRECT)) {
-				float bssrdf_u, bssrdf_v;
-				path_state_rng_2D(kg,
-				                  state,
-				                  PRNG_BSDF_U,
-				                  &bssrdf_u, &bssrdf_v);
-
-				const ShaderClosure *sc = shader_bssrdf_pick(sd, throughput, &bssrdf_u);
-
-				/* do bssrdf scatter step if we picked a bssrdf closure */
-				if(sc) {
-					uint lcg_state = lcg_state_init_addrspace(state, 0x68bc21eb);
-					subsurface_scatter_step(kg,
-					                        sd,
-					                        state,
-					                        sc,
-					                        &lcg_state,
-					                        bssrdf_u, bssrdf_v,
-					                        false);
-				}
-			}
 			else {
 				kernel_split_branched_path_subsurface_indirect_light_init(kg, ray_index);