Code cleanup: deduplicate some branched and split kernel code.

Benchmarks peformance on GTX 1080 and RX 480 on Linux is the same for
bmw27, classroom, pabellon, and about 2% faster on fishy_cat and koro.

5 files changed, 28 insertions, 185 deletions

diff --git a/intern/cycles/kernel/split/kernel_direct_lighting.h b/intern/cycles/kernel/split/kernel_direct_lighting.h
index 5009828df92..2aac66ecb84 100644
--- a/intern/cycles/kernel/split/kernel_direct_lighting.h
+++ b/intern/cycles/kernel/split/kernel_direct_lighting.h
@@ -94,9 +94,7 @@ ccl_device void kernel_direct_lighting(KernelGlobals *kg,
 			                &ls)) {
 
 				Ray light_ray;
-#  ifdef __OBJECT_MOTION__
 				light_ray.time = sd->time;
-#  endif
 
 				BsdfEval L_light;
 				bool is_lamp;
diff --git a/intern/cycles/kernel/split/kernel_holdout_emission_blurring_pathtermination_ao.h b/intern/cycles/kernel/split/kernel_holdout_emission_blurring_pathtermination_ao.h
index 253b78526e7..733c25f5aa4 100644
--- a/intern/cycles/kernel/split/kernel_holdout_emission_blurring_pathtermination_ao.h
+++ b/intern/cycles/kernel/split/kernel_holdout_emission_blurring_pathtermination_ao.h
@@ -94,125 +94,42 @@ ccl_device void kernel_holdout_emission_blurring_pathtermination_ao(
 
 	ccl_global PathState *state = 0x0;
 	float3 throughput;
-	uint sample;
 
 	ccl_global char *ray_state = kernel_split_state.ray_state;
 	ShaderData *sd = &kernel_split_state.sd[ray_index];
-	ccl_global float *buffer = kernel_split_params.buffer;
 
 	if(IS_STATE(ray_state, ray_index, RAY_ACTIVE)) {
 		uint work_index = kernel_split_state.work_array[ray_index];
-		sample = get_work_sample(kg, work_index, ray_index) + kernel_split_params.start_sample;
-
 		uint pixel_x, pixel_y, tile_x, tile_y;
 		get_work_pixel_tile_position(kg, &pixel_x, &pixel_y,
 		                        &tile_x, &tile_y,
 		                        work_index,
 		                        ray_index);
 
+		ccl_global float *buffer = kernel_split_params.buffer;
 		buffer += (kernel_split_params.offset + pixel_x + pixel_y * stride) * kernel_data.film.pass_stride;
 
+		ccl_global Ray *ray = &kernel_split_state.ray[ray_index];
+		ShaderData *emission_sd = &kernel_split_state.sd_DL_shadow[ray_index];
+		PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
+
 		throughput = kernel_split_state.throughput[ray_index];
 		state = &kernel_split_state.path_state[ray_index];
 
-#ifdef __SHADOW_TRICKS__
-		if((sd->object_flag & SD_OBJECT_SHADOW_CATCHER)) {
-			if(state->flag & PATH_RAY_CAMERA) {
-				PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
-				state->flag |= (PATH_RAY_SHADOW_CATCHER |
-				                PATH_RAY_STORE_SHADOW_INFO);
-				if(!kernel_data.background.transparent) {
-					ccl_global Ray *ray = &kernel_split_state.ray[ray_index];
-					L->shadow_background_color = indirect_background(
-					        kg,
-					        &kernel_split_state.sd_DL_shadow[ray_index],
-					        state,
-					        ray);
-				}
-				L->shadow_radiance_sum = path_radiance_clamp_and_sum(kg, L);
-				L->shadow_throughput = average(throughput);
-			}
-		}
-		else if(state->flag & PATH_RAY_SHADOW_CATCHER) {
-			/* Only update transparency after shadow catcher bounce. */
-			PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
-			L->shadow_transparency *= average(shader_bsdf_transparency(kg, sd));
-		}
-#endif  /* __SHADOW_TRICKS__ */
-
-		/* holdout */
-#ifdef __HOLDOUT__
-		if(((sd->flag & SD_HOLDOUT) ||
-		    (sd->object_flag & SD_OBJECT_HOLDOUT_MASK)) &&
-		   (state->flag & PATH_RAY_CAMERA))
+		if(!kernel_path_shader_apply(kg,
+		                             sd,
+		                             state,
+		                             ray,
+		                             throughput,
+		                             emission_sd,
+		                             L,
+		                             buffer))
 		{
-			if(kernel_data.background.transparent) {
-				float3 holdout_weight;
-				if(sd->object_flag & SD_OBJECT_HOLDOUT_MASK) {
-					holdout_weight = make_float3(1.0f, 1.0f, 1.0f);
-				}
-				else {
-					holdout_weight = shader_holdout_eval(kg, sd);
-				}
-				/* any throughput is ok, should all be identical here */
-				PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
-				L->transparent += average(holdout_weight*throughput);
-			}
-			if(sd->object_flag & SD_OBJECT_HOLDOUT_MASK) {
-				kernel_split_path_end(kg, ray_index);
-			}
+			kernel_split_path_end(kg, ray_index);
 		}
-#endif  /* __HOLDOUT__ */
 	}
 
 	if(IS_STATE(ray_state, ray_index, RAY_ACTIVE)) {
-		PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
-
-#ifdef __BRANCHED_PATH__
-		if(!IS_FLAG(ray_state, ray_index, RAY_BRANCHED_INDIRECT))
-#endif  /* __BRANCHED_PATH__ */
-		{
-			/* Holdout mask objects do not write data passes. */
-			kernel_write_data_passes(kg,
-				                     buffer,
-				                     L,
-				                     sd,
-				                     sample,
-				                     state,
-				                     throughput);
-		}
-
-		/* Blurring of bsdf after bounces, for rays that have a small likelihood
-		 * of following this particular path (diffuse, rough glossy.
-		 */
-#ifndef __BRANCHED_PATH__
-		if(kernel_data.integrator.filter_glossy != FLT_MAX)
-#else
-		if(kernel_data.integrator.filter_glossy != FLT_MAX &&
-		   (!kernel_data.integrator.branched || IS_FLAG(ray_state, ray_index, RAY_BRANCHED_INDIRECT)))
-#endif  /* __BRANCHED_PATH__ */
-		{
-			float blur_pdf = kernel_data.integrator.filter_glossy*state->min_ray_pdf;
-			if(blur_pdf < 1.0f) {
-				float blur_roughness = sqrtf(1.0f - blur_pdf)*0.5f;
-				shader_bsdf_blur(kg, sd, blur_roughness);
-			}
-		}
-
-#ifdef __EMISSION__
-		/* emission */
-		if(sd->flag & SD_EMISSION) {
-			/* TODO(sergey): is isect.t wrong here for transparent surfaces? */
-			float3 emission = indirect_primitive_emission(
-			        kg,
-			        sd,
-			        kernel_split_state.isect[ray_index].t,
-			        state->flag,
-			        state->ray_pdf);
-			path_radiance_accum_emission(L, throughput, emission, state->bounce);
-		}
-#endif  /* __EMISSION__ */
-
 		/* Path termination. this is a strange place to put the termination, it's
 		 * mainly due to the mixed in MIS that we use. gives too many unneeded
 		 * shader evaluations, only need emission if we are going to terminate.
@@ -249,6 +166,7 @@ ccl_device void kernel_holdout_emission_blurring_pathtermination_ao(
 				}
 			}
 
+			PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
 			kernel_update_denoising_features(kg, sd, state, L);
 		}
 	}
diff --git a/intern/cycles/kernel/split/kernel_indirect_background.h b/intern/cycles/kernel/split/kernel_indirect_background.h
index 2e9231caa3e..437043a5971 100644
--- a/intern/cycles/kernel/split/kernel_indirect_background.h
+++ b/intern/cycles/kernel/split/kernel_indirect_background.h
@@ -50,32 +50,16 @@ ccl_device void kernel_indirect_background(KernelGlobals *kg)
 		return;
 	}
 
-	ccl_global PathState *state = &kernel_split_state.path_state[ray_index];
-	PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
-	ccl_global Ray *ray = &kernel_split_state.ray[ray_index];
-	ccl_global float3 *throughput = &kernel_split_state.throughput[ray_index];
-
 	if(IS_STATE(ray_state, ray_index, RAY_HIT_BACKGROUND)) {
-		/* eval background shader if nothing hit */
-		if(kernel_data.background.transparent && (state->flag & PATH_RAY_CAMERA)) {
-			L->transparent += average((*throughput));
-#ifdef __PASSES__
-			if(!(kernel_data.film.pass_flag & PASS_BACKGROUND))
-#endif
-				kernel_split_path_end(kg, ray_index);
-		}
+		ccl_global PathState *state = &kernel_split_state.path_state[ray_index];
+		PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
+		ccl_global Ray *ray = &kernel_split_state.ray[ray_index];
+		float3 throughput = kernel_split_state.throughput[ray_index];
+		ShaderData *emission_sd = &kernel_split_state.sd_DL_shadow[ray_index];
 
-		if(IS_STATE(ray_state, ray_index, RAY_HIT_BACKGROUND)) {
-#ifdef __BACKGROUND__
-			/* sample background shader */
-			float3 L_background = indirect_background(kg, &kernel_split_state.sd_DL_shadow[ray_index], state, ray);
-			path_radiance_accum_background(L, state, (*throughput), L_background);
-#endif
-			kernel_split_path_end(kg, ray_index);
-		}
+		kernel_path_background(kg, state, ray, throughput, emission_sd, L);
+		kernel_split_path_end(kg, ray_index);
 	}
-
-
 }
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/kernel/split/kernel_lamp_emission.h b/intern/cycles/kernel/split/kernel_lamp_emission.h
index c669d79ddcd..448456d167d 100644
--- a/intern/cycles/kernel/split/kernel_lamp_emission.h
+++ b/intern/cycles/kernel/split/kernel_lamp_emission.h
@@ -57,27 +57,10 @@ ccl_device void kernel_lamp_emission(KernelGlobals *kg)
 
 		float3 throughput = kernel_split_state.throughput[ray_index];
 		Ray ray = kernel_split_state.ray[ray_index];
+		ccl_global Intersection *isect = &kernel_split_state.isect[ray_index];
+		ShaderData *emission_sd = &kernel_split_state.sd_DL_shadow[ray_index];
 
-#ifdef __LAMP_MIS__
-		if(kernel_data.integrator.use_lamp_mis && !(state->flag & PATH_RAY_CAMERA)) {
-			/* ray starting from previous non-transparent bounce */
-			Ray light_ray;
-
-			light_ray.P = ray.P - state->ray_t*ray.D;
-			state->ray_t += kernel_split_state.isect[ray_index].t;
-			light_ray.D = ray.D;
-			light_ray.t = state->ray_t;
-			light_ray.time = ray.time;
-			light_ray.dD = ray.dD;
-			light_ray.dP = ray.dP;
-			/* intersect with lamp */
-			float3 emission;
-
-			if(indirect_lamp_emission(kg, &kernel_split_state.sd_DL_shadow[ray_index], state, &light_ray, &emission)) {
-				path_radiance_accum_emission(L, throughput, emission, state->bounce);
-			}
-		}
-#endif  /* __LAMP_MIS__ */
+		kernel_path_lamp_emission(kg, state, &ray, throughput, isect, emission_sd, L);
 	}
 }
 
diff --git a/intern/cycles/kernel/split/kernel_scene_intersect.h b/intern/cycles/kernel/split/kernel_scene_intersect.h
index 99095f8fcbc..f5378bc172b 100644
--- a/intern/cycles/kernel/split/kernel_scene_intersect.h
+++ b/intern/cycles/kernel/split/kernel_scene_intersect.h
@@ -59,53 +59,13 @@ ccl_device void kernel_scene_intersect(KernelGlobals *kg)
 		return;
 	}
 
-	Intersection isect;
 	ccl_global PathState *state = &kernel_split_state.path_state[ray_index];
 	Ray ray = kernel_split_state.ray[ray_index];
-
-	/* intersect scene */
-	uint visibility = path_state_ray_visibility(kg, state);
-
-	if(path_state_ao_bounce(kg, state)) {
-		visibility = PATH_RAY_SHADOW;
-		ray.t = kernel_data.background.ao_distance;
-	}
-
-#ifdef __HAIR__
-	float difl = 0.0f, extmax = 0.0f;
-	uint lcg_state = 0;
-
-	if(kernel_data.bvh.have_curves) {
-		if((kernel_data.cam.resolution == 1) && (state->flag & PATH_RAY_CAMERA)) {
-			float3 pixdiff = ray.dD.dx + ray.dD.dy;
-			/*pixdiff = pixdiff - dot(pixdiff, ray.D)*ray.D;*/
-			difl = kernel_data.curve.minimum_width * len(pixdiff) * 0.5f;
-		}
-
-		extmax = kernel_data.curve.maximum_width;
-		lcg_state = lcg_state_init_addrspace(state, 0x51633e2d);
-	}
-
-	bool hit = scene_intersect(kg, ray, visibility, &isect, &lcg_state, difl, extmax);
-#else
-	if(path_state_ao_bounce(kg, state)) {
-		visibility = PATH_RAY_SHADOW;
-		ray.t = kernel_data.background.ao_distance;
-	}
-	bool hit = scene_intersect(kg, ray, visibility, &isect, NULL, 0.0f, 0.0f);
-#endif
-	kernel_split_state.isect[ray_index] = isect;
-
-#ifdef __KERNEL_DEBUG__
 	PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
 
-	if(state->flag & PATH_RAY_CAMERA) {
-		L->debug_data.num_bvh_traversed_nodes += isect.num_traversed_nodes;
-		L->debug_data.num_bvh_traversed_instances += isect.num_traversed_instances;
-		L->debug_data.num_bvh_intersections += isect.num_intersections;
-	}
-	L->debug_data.num_ray_bounces++;
-#endif
+	Intersection isect;
+	bool hit = kernel_path_scene_intersect(kg, state, &ray, &isect, L);
+	kernel_split_state.isect[ray_index] = isect;
 
 	if(!hit) {
 		/* Change the state of rays that hit the background;