Code cleanup: move rng into path state.

Also pass by value and don't write back now that it is just a hash for seeding
and no longer an LCG state. Together this makes CUDA a tiny bit faster in my
tests, but mainly simplifies code.

13 files changed, 33 insertions, 68 deletions

diff --git a/intern/cycles/kernel/split/kernel_branched.h b/intern/cycles/kernel/split/kernel_branched.h
index e2762a85fc8..9fe4ec18e9e 100644
--- a/intern/cycles/kernel/split/kernel_branched.h
+++ b/intern/cycles/kernel/split/kernel_branched.h
@@ -110,7 +110,6 @@ ccl_device_noinline bool kernel_split_branched_path_surface_indirect_light_iter(
 	SplitBranchedState *branched_state = &kernel_split_state.branched_state[ray_index];
 
 	ShaderData *sd = saved_sd;
-	RNG rng = kernel_split_state.rng[ray_index];
 	PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
 	float3 throughput = branched_state->throughput;
 	ccl_global PathState *ps = &kernel_split_state.path_state[ray_index];
@@ -157,20 +156,20 @@ ccl_device_noinline bool kernel_split_branched_path_surface_indirect_light_iter(
 		num_samples = ceil_to_int(num_samples_adjust*num_samples);
 
 		float num_samples_inv = num_samples_adjust/num_samples;
-		RNG bsdf_rng = cmj_hash(rng, i);
 
 		for(int j = branched_state->next_sample; j < num_samples; j++) {
 			if(reset_path_state) {
 				*ps = branched_state->path_state;
 			}
 
+			ps->rng_hash = cmj_hash(branched_state->path_state.rng_hash, i);
+
 			ccl_global float3 *tp = &kernel_split_state.throughput[ray_index];
 			*tp = throughput;
 
 			ccl_global Ray *bsdf_ray = &kernel_split_state.ray[ray_index];
 
 			if(!kernel_branched_path_surface_bounce(kg,
-			                                        &bsdf_rng,
 			                                        sd,
 			                                        sc,
 			                                        j,
@@ -184,6 +183,8 @@ ccl_device_noinline bool kernel_split_branched_path_surface_indirect_light_iter(
 				continue;
 			}
 
+			ps->rng_hash = branched_state->path_state.rng_hash;
+
 			/* update state for next iteration */
 			branched_state->next_closure = i;
 			branched_state->next_sample = j+1;
diff --git a/intern/cycles/kernel/split/kernel_buffer_update.h b/intern/cycles/kernel/split/kernel_buffer_update.h
index 6aa0d6948d0..de0c4160ca0 100644
--- a/intern/cycles/kernel/split/kernel_buffer_update.h
+++ b/intern/cycles/kernel/split/kernel_buffer_update.h
@@ -83,7 +83,6 @@ ccl_device void kernel_buffer_update(KernelGlobals *kg,
 	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];
-	RNG rng = kernel_split_state.rng[ray_index];
 	ccl_global float *buffer = kernel_split_params.buffer;
 
 	unsigned int work_index;
@@ -135,7 +134,8 @@ ccl_device void kernel_buffer_update(KernelGlobals *kg,
 			buffer += (kernel_split_params.offset + pixel_x + pixel_y*stride) * kernel_data.film.pass_stride;
 
 			/* Initialize random numbers and ray. */
-			kernel_path_trace_setup(kg, rng_state, sample, pixel_x, pixel_y, &rng, ray);
+			uint rng_hash;
+			kernel_path_trace_setup(kg, rng_state, sample, pixel_x, pixel_y, &rng_hash, ray);
 
 			if(ray->t != 0.0f) {
 				/* Initialize throughput, path radiance, Ray, PathState;
@@ -143,7 +143,7 @@ ccl_device void kernel_buffer_update(KernelGlobals *kg,
 				 */
 				*throughput = make_float3(1.0f, 1.0f, 1.0f);
 				path_radiance_init(L, kernel_data.film.use_light_pass);
-				path_state_init(kg, &kernel_split_state.sd_DL_shadow[ray_index], state, &rng, sample, ray);
+				path_state_init(kg, &kernel_split_state.sd_DL_shadow[ray_index], state, rng_hash, sample, ray);
 #ifdef __SUBSURFACE__
 				kernel_path_subsurface_init_indirect(&kernel_split_state.ss_rays[ray_index]);
 #endif
@@ -160,7 +160,6 @@ ccl_device void kernel_buffer_update(KernelGlobals *kg,
 			}
 		}
 	}
-	kernel_split_state.rng[ray_index] = rng;
 
 #ifndef __COMPUTE_DEVICE_GPU__
 	}
diff --git a/intern/cycles/kernel/split/kernel_direct_lighting.h b/intern/cycles/kernel/split/kernel_direct_lighting.h
index 3336c968a44..8e3f7555550 100644
--- a/intern/cycles/kernel/split/kernel_direct_lighting.h
+++ b/intern/cycles/kernel/split/kernel_direct_lighting.h
@@ -62,8 +62,6 @@ ccl_device void kernel_direct_lighting(KernelGlobals *kg,
 
 		/* direct lighting */
 #ifdef __EMISSION__
-		RNG rng = kernel_split_state.rng[ray_index];
-
 		bool flag = (kernel_data.integrator.use_direct_light &&
 		             (sd->flag & SD_BSDF_HAS_EVAL));
 
@@ -83,10 +81,10 @@ ccl_device void kernel_direct_lighting(KernelGlobals *kg,
 
 		if(flag) {
 			/* Sample illumination from lights to find path contribution. */
-			float light_t = path_state_rng_1D(kg, &rng, state, PRNG_LIGHT);
+			float light_t = path_state_rng_1D(kg, state, PRNG_LIGHT);
 			float light_u, light_v;
-			path_state_rng_2D(kg, &rng, state, PRNG_LIGHT_U, &light_u, &light_v);
-			float terminate = path_state_rng_light_termination(kg, &rng, state);
+			path_state_rng_2D(kg, state, PRNG_LIGHT_U, &light_u, &light_v);
+			float terminate = path_state_rng_light_termination(kg, state);
 
 			LightSample ls;
 			if(light_sample(kg,
@@ -115,7 +113,6 @@ ccl_device void kernel_direct_lighting(KernelGlobals *kg,
 				}
 			}
 		}
-		kernel_split_state.rng[ray_index] = rng;
 #endif  /* __EMISSION__ */
 	}
 
diff --git a/intern/cycles/kernel/split/kernel_do_volume.h b/intern/cycles/kernel/split/kernel_do_volume.h
index 9f8dd2392d9..478d83d633e 100644
--- a/intern/cycles/kernel/split/kernel_do_volume.h
+++ b/intern/cycles/kernel/split/kernel_do_volume.h
@@ -30,7 +30,6 @@ ccl_device_noinline bool kernel_split_branched_path_volume_indirect_light_iter(K
 	SplitBranchedState *branched_state = &kernel_split_state.branched_state[ray_index];
 
 	ShaderData *sd = &kernel_split_state.sd[ray_index];
-	RNG rng = kernel_split_state.rng[ray_index];
 	PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
 	ShaderData *emission_sd = &kernel_split_state.sd_DL_shadow[ray_index];
 
@@ -58,15 +57,15 @@ ccl_device_noinline bool kernel_split_branched_path_volume_indirect_light_iter(K
 
 		/* integrate along volume segment with distance sampling */
 		VolumeIntegrateResult result = kernel_volume_integrate(
-			kg, ps, sd, &volume_ray, L, tp, &rng, heterogeneous);
+			kg, ps, sd, &volume_ray, L, tp, heterogeneous);
 
 #  ifdef __VOLUME_SCATTER__
 		if(result == VOLUME_PATH_SCATTERED) {
 			/* direct lighting */
-			kernel_path_volume_connect_light(kg, &rng, sd, emission_sd, *tp, &branched_state->path_state, L);
+			kernel_path_volume_connect_light(kg, sd, emission_sd, *tp, &branched_state->path_state, L);
 
 			/* indirect light bounce */
-			if(!kernel_path_volume_bounce(kg, &rng, sd, tp, ps, L, pray)) {
+			if(!kernel_path_volume_bounce(kg, sd, tp, ps, L, pray)) {
 				continue;
 			}
 
@@ -141,7 +140,6 @@ ccl_device void kernel_do_volume(KernelGlobals *kg)
 	   IS_STATE(ray_state, ray_index, RAY_HIT_BACKGROUND)) {
 		ccl_global float3 *throughput = &kernel_split_state.throughput[ray_index];
 		ccl_global Ray *ray = &kernel_split_state.ray[ray_index];
-		RNG rng = kernel_split_state.rng[ray_index];
 		ccl_global Intersection *isect = &kernel_split_state.isect[ray_index];
 		ShaderData *sd = &kernel_split_state.sd[ray_index];
 		ShaderData *emission_sd = &kernel_split_state.sd_DL_shadow[ray_index];
@@ -165,15 +163,15 @@ ccl_device void kernel_do_volume(KernelGlobals *kg)
 				{
 					/* integrate along volume segment with distance sampling */
 					VolumeIntegrateResult result = kernel_volume_integrate(
-						kg, state, sd, &volume_ray, L, throughput, &rng, heterogeneous);
+						kg, state, sd, &volume_ray, L, throughput, heterogeneous);
 
 #  ifdef __VOLUME_SCATTER__
 					if(result == VOLUME_PATH_SCATTERED) {
 						/* direct lighting */
-						kernel_path_volume_connect_light(kg, &rng, sd, emission_sd, *throughput, state, L);
+						kernel_path_volume_connect_light(kg, sd, emission_sd, *throughput, state, L);
 
 						/* indirect light bounce */
-						if(kernel_path_volume_bounce(kg, &rng, sd, throughput, state, L, ray)) {
+						if(kernel_path_volume_bounce(kg, sd, throughput, state, L, ray)) {
 							ASSIGN_RAY_STATE(ray_state, ray_index, RAY_REGENERATED);
 						}
 						else {
@@ -194,8 +192,6 @@ ccl_device void kernel_do_volume(KernelGlobals *kg)
 			}
 #  endif  /* __BRANCHED_PATH__ */
 		}
-
-		kernel_split_state.rng[ray_index] = rng;
 	}
 
 #  ifdef __BRANCHED_PATH__
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 95f57fbff57..3fc45afbd92 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
@@ -100,7 +100,6 @@ ccl_device void kernel_holdout_emission_blurring_pathtermination_ao(
 	unsigned int tile_y;
 	unsigned int sample;
 
-	RNG rng = kernel_split_state.rng[ray_index];
 	ccl_global PathState *state = 0x0;
 	float3 throughput;
 
@@ -247,7 +246,7 @@ ccl_device void kernel_holdout_emission_blurring_pathtermination_ao(
 
 		if(IS_STATE(ray_state, ray_index, RAY_ACTIVE)) {
 			if(probability != 1.0f) {
-				float terminate = path_state_rng_1D_for_decision(kg, &rng, state, PRNG_TERMINATE);
+				float terminate = path_state_rng_1D_for_decision(kg, state, PRNG_TERMINATE);
 				if(terminate >= probability) {
 					kernel_split_path_end(kg, ray_index);
 				}
@@ -269,8 +268,6 @@ ccl_device void kernel_holdout_emission_blurring_pathtermination_ao(
 	}
 #endif  /* __AO__ */
 
-	kernel_split_state.rng[ray_index] = rng;
-
 #ifndef __COMPUTE_DEVICE_GPU__
 	}
 #endif
diff --git a/intern/cycles/kernel/split/kernel_next_iteration_setup.h b/intern/cycles/kernel/split/kernel_next_iteration_setup.h
index 7758e35fd32..4e0c966cca9 100644
--- a/intern/cycles/kernel/split/kernel_next_iteration_setup.h
+++ b/intern/cycles/kernel/split/kernel_next_iteration_setup.h
@@ -126,7 +126,6 @@ ccl_device void kernel_next_iteration_setup(KernelGlobals *kg,
 	if(active) {
 		ccl_global float3 *throughput = &kernel_split_state.throughput[ray_index];
 		ccl_global Ray *ray = &kernel_split_state.ray[ray_index];
-		RNG rng = kernel_split_state.rng[ray_index];
 		ShaderData *sd = &kernel_split_state.sd[ray_index];
 		ccl_global PathState *state = &kernel_split_state.path_state[ray_index];
 		PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
@@ -135,7 +134,7 @@ ccl_device void kernel_next_iteration_setup(KernelGlobals *kg,
 		if(!kernel_data.integrator.branched || IS_FLAG(ray_state, ray_index, RAY_BRANCHED_INDIRECT)) {
 #endif
 			/* Compute direct lighting and next bounce. */
-			if(!kernel_path_surface_bounce(kg, &rng, sd, throughput, state, L, ray)) {
+			if(!kernel_path_surface_bounce(kg, sd, throughput, state, L, ray)) {
 				kernel_split_path_end(kg, ray_index);
 			}
 #ifdef __BRANCHED_PATH__
@@ -157,8 +156,6 @@ ccl_device void kernel_next_iteration_setup(KernelGlobals *kg,
 			}
 		}
 #endif  /* __BRANCHED_PATH__ */
-
-		kernel_split_state.rng[ray_index] = rng;
 	}
 
 	/* Enqueue RAY_UPDATE_BUFFER rays. */
diff --git a/intern/cycles/kernel/split/kernel_path_init.h b/intern/cycles/kernel/split/kernel_path_init.h
index 8315b0b2bd3..8b70df16d2a 100644
--- a/intern/cycles/kernel/split/kernel_path_init.h
+++ b/intern/cycles/kernel/split/kernel_path_init.h
@@ -60,14 +60,14 @@ ccl_device void kernel_path_init(KernelGlobals *kg) {
 	ccl_global float *buffer = kernel_split_params.buffer;
 	buffer += (kernel_split_params.offset + pixel_x + pixel_y * kernel_split_params.stride) * kernel_data.film.pass_stride;
 
-	RNG rng = kernel_split_state.rng[ray_index];
+	uint rng_hash;
 
 	/* Initialize random numbers and ray. */
 	kernel_path_trace_setup(kg,
 	                        rng_state,
 	                        my_sample,
 	                        pixel_x, pixel_y,
-	                        &rng,
+	                        &rng_hash,
 	                        &kernel_split_state.ray[ray_index]);
 
 	if(kernel_split_state.ray[ray_index].t != 0.0f) {
@@ -79,7 +79,7 @@ ccl_device void kernel_path_init(KernelGlobals *kg) {
 		path_state_init(kg,
 		                &kernel_split_state.sd_DL_shadow[ray_index],
 		                &kernel_split_state.path_state[ray_index],
-		                &rng,
+		                rng_hash,
 		                my_sample,
 		                &kernel_split_state.ray[ray_index]);
 #ifdef __SUBSURFACE__
@@ -93,7 +93,6 @@ ccl_device void kernel_path_init(KernelGlobals *kg) {
 		kernel_write_pass_float4(buffer, my_sample, L_rad);
 		ASSIGN_RAY_STATE(kernel_split_state.ray_state, ray_index, RAY_TO_REGENERATE);
 	}
-	kernel_split_state.rng[ray_index] = rng;
 }
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/kernel/split/kernel_scene_intersect.h b/intern/cycles/kernel/split/kernel_scene_intersect.h
index 5c6d90eecc4..d0afd39ef29 100644
--- a/intern/cycles/kernel/split/kernel_scene_intersect.h
+++ b/intern/cycles/kernel/split/kernel_scene_intersect.h
@@ -74,7 +74,6 @@ ccl_device void kernel_scene_intersect(KernelGlobals *kg)
 #ifdef __HAIR__
 	float difl = 0.0f, extmax = 0.0f;
 	uint lcg_state = 0;
-	RNG rng = kernel_split_state.rng[ray_index];
 
 	if(kernel_data.bvh.have_curves) {
 		if((kernel_data.cam.resolution == 1) && (state.flag & PATH_RAY_CAMERA)) {
@@ -84,7 +83,7 @@ ccl_device void kernel_scene_intersect(KernelGlobals *kg)
 		}
 
 		extmax = kernel_data.curve.maximum_width;
-		lcg_state = lcg_state_init(&rng, state.rng_offset, state.sample, 0x51633e2d);
+		lcg_state = lcg_state_init(&state, 0x51633e2d);
 	}
 
 	bool hit = scene_intersect(kg, ray, visibility, &isect, &lcg_state, difl, extmax);
diff --git a/intern/cycles/kernel/split/kernel_shader_eval.h b/intern/cycles/kernel/split/kernel_shader_eval.h
index 2801b32f285..d0fa29ef3f8 100644
--- a/intern/cycles/kernel/split/kernel_shader_eval.h
+++ b/intern/cycles/kernel/split/kernel_shader_eval.h
@@ -48,18 +48,17 @@ ccl_device void kernel_shader_eval(KernelGlobals *kg)
 
 	ccl_global char *ray_state = kernel_split_state.ray_state;
 	if(IS_STATE(ray_state, ray_index, RAY_ACTIVE)) {
-		RNG rng = kernel_split_state.rng[ray_index];
 		ccl_global PathState *state = &kernel_split_state.path_state[ray_index];
 
 #ifndef __BRANCHED_PATH__
-		float rbsdf = path_state_rng_1D_for_decision(kg, &rng, state, PRNG_BSDF);
-		shader_eval_surface(kg, &kernel_split_state.sd[ray_index], &rng, state, rbsdf, state->flag, SHADER_CONTEXT_MAIN);
+		float rbsdf = path_state_rng_1D_for_decision(kg, state, PRNG_BSDF);
+		shader_eval_surface(kg, &kernel_split_state.sd[ray_index], state, rbsdf, state->flag, SHADER_CONTEXT_MAIN);
 #else
 		ShaderContext ctx = SHADER_CONTEXT_MAIN;
 		float rbsdf = 0.0f;
 
 		if(!kernel_data.integrator.branched || IS_FLAG(ray_state, ray_index, RAY_BRANCHED_INDIRECT)) {
-			rbsdf = path_state_rng_1D_for_decision(kg, &rng, state, PRNG_BSDF);
+			rbsdf = path_state_rng_1D_for_decision(kg, state, PRNG_BSDF);
 
 		}
 
@@ -67,11 +66,9 @@ ccl_device void kernel_shader_eval(KernelGlobals *kg)
 			ctx = SHADER_CONTEXT_INDIRECT;
 		}
 
-		shader_eval_surface(kg, &kernel_split_state.sd[ray_index], &rng, state, rbsdf, state->flag, ctx);
+		shader_eval_surface(kg, &kernel_split_state.sd[ray_index], state, rbsdf, state->flag, ctx);
 		shader_merge_closures(&kernel_split_state.sd[ray_index]);
 #endif  /* __BRANCHED_PATH__ */
-
-		kernel_split_state.rng[ray_index] = rng;
 	}
 }
 
diff --git a/intern/cycles/kernel/split/kernel_shadow_blocked_ao.h b/intern/cycles/kernel/split/kernel_shadow_blocked_ao.h
index 474286285a9..79aa2c9435b 100644
--- a/intern/cycles/kernel/split/kernel_shadow_blocked_ao.h
+++ b/intern/cycles/kernel/split/kernel_shadow_blocked_ao.h
@@ -37,21 +37,18 @@ ccl_device void kernel_shadow_blocked_ao(KernelGlobals *kg)
 	ShaderData *emission_sd = &kernel_split_state.sd_DL_shadow[ray_index];
 	PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
 	ccl_global PathState *state = &kernel_split_state.path_state[ray_index];
-	RNG rng = kernel_split_state.rng[ray_index];
 	float3 throughput = kernel_split_state.throughput[ray_index];
 
 #ifdef __BRANCHED_PATH__
 	if(!kernel_data.integrator.branched || IS_FLAG(kernel_split_state.ray_state, ray_index, RAY_BRANCHED_INDIRECT)) {
 #endif
-		kernel_path_ao(kg, sd, emission_sd, L, state, &rng, throughput, shader_bsdf_alpha(kg, sd));
+		kernel_path_ao(kg, sd, emission_sd, L, state, throughput, shader_bsdf_alpha(kg, sd));
 #ifdef __BRANCHED_PATH__
 	}
 	else {
-		kernel_branched_path_ao(kg, sd, emission_sd, L, state, &rng, throughput);
+		kernel_branched_path_ao(kg, sd, emission_sd, L, state, throughput);
 	}
 #endif
-
-	kernel_split_state.rng[ray_index] = rng;
 }
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/kernel/split/kernel_shadow_blocked_dl.h b/intern/cycles/kernel/split/kernel_shadow_blocked_dl.h
index 78e61709b01..50c9acec4ee 100644
--- a/intern/cycles/kernel/split/kernel_shadow_blocked_dl.h
+++ b/intern/cycles/kernel/split/kernel_shadow_blocked_dl.h
@@ -45,7 +45,6 @@ ccl_device void kernel_shadow_blocked_dl(KernelGlobals *kg)
 	PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
 	ShaderData *sd = &kernel_split_state.sd[ray_index];
 	float3 throughput = kernel_split_state.throughput[ray_index];
-	RNG rng = kernel_split_state.rng[ray_index];
 
 	BsdfEval L_light = kernel_split_state.bsdf_eval[ray_index];
 	ShaderData *emission_sd = &kernel_split_state.sd_DL_shadow[ray_index];
@@ -75,7 +74,6 @@ ccl_device void kernel_shadow_blocked_dl(KernelGlobals *kg)
 
 	if(use_branched) {
 		kernel_branched_path_surface_connect_light(kg,
-		                                           &rng,
 		                                           sd,
 		                                           emission_sd,
 		                                           state,
@@ -103,8 +101,6 @@ ccl_device void kernel_shadow_blocked_dl(KernelGlobals *kg)
 			path_radiance_accum_total_light(L, state, throughput, &L_light);
 		}
 	}
-
-	kernel_split_state.rng[ray_index] = rng;
 }
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/kernel/split/kernel_split_data_types.h b/intern/cycles/kernel/split/kernel_split_data_types.h
index 4f32c68d630..3eae884d479 100644
--- a/intern/cycles/kernel/split/kernel_split_data_types.h
+++ b/intern/cycles/kernel/split/kernel_split_data_types.h
@@ -114,7 +114,6 @@ typedef ccl_global struct SplitBranchedState {
 #endif /* __VOLUME__ */
 
 #define SPLIT_DATA_ENTRIES \
-	SPLIT_DATA_ENTRY(ccl_global RNG, rng, 1) \
 	SPLIT_DATA_ENTRY(ccl_global float3, throughput, 1) \
 	SPLIT_DATA_ENTRY(PathRadiance, path_radiance, 1) \
 	SPLIT_DATA_ENTRY(ccl_global Ray, ray, 1) \
@@ -133,7 +132,6 @@ typedef ccl_global struct SplitBranchedState {
 
 /* entries to be copied to inactive rays when sharing branched samples (TODO: which are actually needed?) */
 #define SPLIT_DATA_ENTRIES_BRANCHED_SHARED \
-	SPLIT_DATA_ENTRY(ccl_global RNG, rng, 1) \
 	SPLIT_DATA_ENTRY(ccl_global float3, throughput, 1) \
 	SPLIT_DATA_ENTRY(PathRadiance, path_radiance, 1) \
 	SPLIT_DATA_ENTRY(ccl_global Ray, ray, 1) \
diff --git a/intern/cycles/kernel/split/kernel_subsurface_scatter.h b/intern/cycles/kernel/split/kernel_subsurface_scatter.h
index d5083b23f80..76d198abc8a 100644
--- a/intern/cycles/kernel/split/kernel_subsurface_scatter.h
+++ b/intern/cycles/kernel/split/kernel_subsurface_scatter.h
@@ -38,7 +38,6 @@ ccl_device_noinline bool kernel_split_branched_path_subsurface_indirect_light_it
 	SplitBranchedState *branched_state = &kernel_split_state.branched_state[ray_index];
 
 	ShaderData *sd = &branched_state->sd;
-	RNG rng = kernel_split_state.rng[ray_index];
 	PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
 	ShaderData *emission_sd = &kernel_split_state.sd_DL_shadow[ray_index];
 
@@ -52,14 +51,12 @@ ccl_device_noinline bool kernel_split_branched_path_subsurface_indirect_light_it
 		if(branched_state->ss_next_sample == 0 && branched_state->next_hit == 0 &&
 		   branched_state->next_closure == 0 && branched_state->next_sample == 0)
 		{
-			branched_state->lcg_state = lcg_state_init(&rng,
-			                                           branched_state->path_state.rng_offset,
-			                                           branched_state->path_state.sample,
+			branched_state->lcg_state = lcg_state_init(&branched_state->path_state,
 			                                           0x68bc21eb);
 		}
 		int num_samples = kernel_data.integrator.subsurface_samples;
 		float num_samples_inv = 1.0f/num_samples;
-		RNG bssrdf_rng = cmj_hash(rng, i);
+		uint bssrdf_rng_hash = cmj_hash(branched_state->path_state.rng_hash, i);
 
 		/* do subsurface scatter step with copy of shader data, this will
 		 * replace the BSSRDF with a diffuse BSDF closure */
@@ -67,7 +64,7 @@ ccl_device_noinline bool kernel_split_branched_path_subsurface_indirect_light_it
 			ccl_global SubsurfaceIntersection *ss_isect = &branched_state->ss_isect;
 			float bssrdf_u, bssrdf_v;
 			path_branched_rng_2D(kg,
-			                     &bssrdf_rng,
+			                     bssrdf_rng_hash,
 			                     &branched_state->path_state,
 			                     j,
 			                     num_samples,
@@ -77,7 +74,7 @@ ccl_device_noinline bool kernel_split_branched_path_subsurface_indirect_light_it
 
 			/* intersection is expensive so avoid doing multiple times for the same input */
 			if(branched_state->next_hit == 0 && branched_state->next_closure == 0 && branched_state->next_sample == 0) {
-				RNG lcg_state = branched_state->lcg_state;
+				uint lcg_state = branched_state->lcg_state;
 				SubsurfaceIntersection ss_isect_private;
 
 				branched_state->num_hits = subsurface_scatter_multi_intersect(kg,
@@ -152,7 +149,6 @@ ccl_device_noinline bool kernel_split_branched_path_subsurface_indirect_light_it
 						int all = (kernel_data.integrator.sample_all_lights_direct) ||
 							      (branched_state->path_state.flag & PATH_RAY_SHADOW_CATCHER);
 						kernel_branched_path_surface_connect_light(kg,
-						                                           &rng,
 						                                           bssrdf_sd,
 						                                           emission_sd,
 						                                           hit_state,
@@ -229,7 +225,6 @@ ccl_device void kernel_subsurface_scatter(KernelGlobals *kg)
 	if(IS_STATE(ray_state, ray_index, RAY_ACTIVE)) {
 		ccl_global PathState *state = &kernel_split_state.path_state[ray_index];
 		PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
-		RNG rng = kernel_split_state.rng[ray_index];
 		ccl_global Ray *ray = &kernel_split_state.ray[ray_index];
 		ccl_global float3 *throughput = &kernel_split_state.throughput[ray_index];
 		ccl_global SubsurfaceIndirectRays *ss_indirect = &kernel_split_state.ss_rays[ray_index];
@@ -246,7 +241,6 @@ ccl_device void kernel_subsurface_scatter(KernelGlobals *kg)
 				                                  emission_sd,
 				                                  L,
 				                                  state,
-				                                  &rng,
 				                                  ray,
 				                                  throughput,
 				                                  ss_indirect))
@@ -264,10 +258,9 @@ ccl_device void kernel_subsurface_scatter(KernelGlobals *kg)
 
 				/* do bssrdf scatter step if we picked a bssrdf closure */
 				if(sc) {
-					uint lcg_state = lcg_state_init(&rng, state->rng_offset, state->sample, 0x68bc21eb);
+					uint lcg_state = lcg_state_init(state, 0x68bc21eb);
 					float bssrdf_u, bssrdf_v;
 					path_state_rng_2D(kg,
-					                  &rng,
 					                  state,
 					                  PRNG_BSDF_U,
 					                  &bssrdf_u, &bssrdf_v);
@@ -290,7 +283,6 @@ ccl_device void kernel_subsurface_scatter(KernelGlobals *kg)
 			}
 #endif
 		}
-		kernel_split_state.rng[ray_index] = rng;
 	}
 
 #  ifdef __BRANCHED_PATH__