Cycles: Implement preliminary test for volume stack update from SSS
[blender-staging.git] / intern / cycles / kernel / kernel_path.h
index cdf80aed4d1ae8e3579da62d5a918400eea3d45f..e68a837001220b706de601a18242a667b16bde2b 100644 (file)
 #include "osl_shader.h"
 #endif
 
-#include "kernel_differential.h"
-#include "kernel_montecarlo.h"
-#include "kernel_projection.h"
-#include "kernel_object.h"
-#include "kernel_triangle.h"
-#include "kernel_curve.h"
-#include "kernel_primitive.h"
-#include "kernel_projection.h"
 #include "kernel_random.h"
-#include "kernel_bvh.h"
-#include "kernel_accumulate.h"
+#include "kernel_projection.h"
+#include "kernel_montecarlo.h"
+#include "kernel_differential.h"
 #include "kernel_camera.h"
+
+#include "geom/geom.h"
+
+#include "kernel_accumulate.h"
 #include "kernel_shader.h"
 #include "kernel_light.h"
-#include "kernel_emission.h"
 #include "kernel_passes.h"
 
 #ifdef __SUBSURFACE__
 #include "kernel_subsurface.h"
 #endif
 
-CCL_NAMESPACE_BEGIN
-
-typedef struct PathState {
-       int flag;
-       int bounce;
+#ifdef __VOLUME__
+#include "kernel_volume.h"
+#endif
 
-       int diffuse_bounce;
-       int glossy_bounce;
-       int transmission_bounce;
-       int transparent_bounce;
-} PathState;
+#include "kernel_path_state.h"
+#include "kernel_shadow.h"
+#include "kernel_emission.h"
+#include "kernel_path_surface.h"
+#include "kernel_path_volume.h"
 
-__device_inline void path_state_init(PathState *state)
-{
-       state->flag = PATH_RAY_CAMERA|PATH_RAY_SINGULAR|PATH_RAY_MIS_SKIP;
-       state->bounce = 0;
-       state->diffuse_bounce = 0;
-       state->glossy_bounce = 0;
-       state->transmission_bounce = 0;
-       state->transparent_bounce = 0;
-}
+CCL_NAMESPACE_BEGIN
 
-__device_inline void path_state_next(KernelGlobals *kg, PathState *state, int label)
+ccl_device void kernel_path_indirect(KernelGlobals *kg, RNG *rng, Ray ray,
+       float3 throughput, int num_samples, PathState state, PathRadiance *L)
 {
-       /* ray through transparent keeps same flags from previous ray and is
-        * not counted as a regular bounce, transparent has separate max */
-       if(label & LABEL_TRANSPARENT) {
-               state->flag |= PATH_RAY_TRANSPARENT;
-               state->transparent_bounce++;
-
-               if(!kernel_data.integrator.transparent_shadows)
-                       state->flag |= PATH_RAY_MIS_SKIP;
-
-               return;
-       }
+       /* path iteration */
+       for(;;) {
+               /* intersect scene */
+               Intersection isect;
+               uint visibility = path_state_ray_visibility(kg, &state);
+               bool hit = scene_intersect(kg, &ray, visibility, &isect, NULL, 0.0f, 0.0f);
 
-       state->bounce++;
+#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;
 
-       /* reflection/transmission */
-       if(label & LABEL_REFLECT) {
-               state->flag |= PATH_RAY_REFLECT;
-               state->flag &= ~(PATH_RAY_TRANSMIT|PATH_RAY_CAMERA|PATH_RAY_TRANSPARENT);
+                       light_ray.P = ray.P - state.ray_t*ray.D;
+                       state.ray_t += isect.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;
 
-               if(label & LABEL_DIFFUSE)
-                       state->diffuse_bounce++;
-               else
-                       state->glossy_bounce++;
-       }
-       else {
-               kernel_assert(label & LABEL_TRANSMIT);
+                       /* intersect with lamp */
+                       float3 emission;
 
-               state->flag |= PATH_RAY_TRANSMIT;
-               state->flag &= ~(PATH_RAY_REFLECT|PATH_RAY_CAMERA|PATH_RAY_TRANSPARENT);
+                       if(indirect_lamp_emission(kg, &state, &light_ray, &emission))
+                               path_radiance_accum_emission(L, throughput, emission, state.bounce);
+               }
+#endif
 
-               state->transmission_bounce++;
-       }
+#ifdef __VOLUME__
+               /* volume attenuation, emission, scatter */
+               if(state.volume_stack[0].shader != SHADER_NONE) {
+                       Ray volume_ray = ray;
+                       volume_ray.t = (hit)? isect.t: FLT_MAX;
 
-       /* diffuse/glossy/singular */
-       if(label & LABEL_DIFFUSE) {
-               state->flag |= PATH_RAY_DIFFUSE|PATH_RAY_DIFFUSE_ANCESTOR;
-               state->flag &= ~(PATH_RAY_GLOSSY|PATH_RAY_SINGULAR|PATH_RAY_MIS_SKIP);
-       }
-       else if(label & LABEL_GLOSSY) {
-               state->flag |= PATH_RAY_GLOSSY|PATH_RAY_GLOSSY_ANCESTOR;
-               state->flag &= ~(PATH_RAY_DIFFUSE|PATH_RAY_SINGULAR|PATH_RAY_MIS_SKIP);
-       }
-       else {
-               kernel_assert(label & LABEL_SINGULAR);
+                       bool heterogeneous = volume_stack_is_heterogeneous(kg, state.volume_stack);
 
-               state->flag |= PATH_RAY_GLOSSY|PATH_RAY_SINGULAR|PATH_RAY_MIS_SKIP;
-               state->flag &= ~PATH_RAY_DIFFUSE;
-       }
-}
+#ifdef __VOLUME_DECOUPLED__
+                       int sampling_method = volume_stack_sampling_method(kg, state.volume_stack);
+                       bool decoupled = kernel_volume_use_decoupled(kg, heterogeneous, false, sampling_method);
 
-__device_inline uint path_state_ray_visibility(KernelGlobals *kg, PathState *state)
-{
-       uint flag = state->flag & PATH_RAY_ALL_VISIBILITY;
+                       if(decoupled) {
+                               /* cache steps along volume for repeated sampling */
+                               VolumeSegment volume_segment;
+                               ShaderData volume_sd;
 
-       /* for visibility, diffuse/glossy are for reflection only */
-       if(flag & PATH_RAY_TRANSMIT)
-               flag &= ~(PATH_RAY_DIFFUSE|PATH_RAY_GLOSSY);
-       /* for camera visibility, use render layer flags */
-       if(flag & PATH_RAY_CAMERA)
-               flag |= kernel_data.integrator.layer_flag;
+                               shader_setup_from_volume(kg, &volume_sd, &volume_ray, state.bounce, state.transparent_bounce);
+                               kernel_volume_decoupled_record(kg, &state,
+                                       &volume_ray, &volume_sd, &volume_segment, heterogeneous);
+                               
+                               volume_segment.sampling_method = sampling_method;
 
-       return flag;
-}
+                               /* emission */
+                               if(volume_segment.closure_flag & SD_EMISSION)
+                                       path_radiance_accum_emission(L, throughput, volume_segment.accum_emission, state.bounce);
 
-__device_inline float path_state_terminate_probability(KernelGlobals *kg, PathState *state, const float3 throughput)
-{
-       if(state->flag & PATH_RAY_TRANSPARENT) {
-               /* transparent rays treated separately */
-               if(state->transparent_bounce >= kernel_data.integrator.transparent_max_bounce)
-                       return 0.0f;
-               else if(state->transparent_bounce <= kernel_data.integrator.transparent_min_bounce)
-                       return 1.0f;
-       }
-       else {
-               /* other rays */
-               if((state->bounce >= kernel_data.integrator.max_bounce) ||
-                  (state->diffuse_bounce >= kernel_data.integrator.max_diffuse_bounce) ||
-                  (state->glossy_bounce >= kernel_data.integrator.max_glossy_bounce) ||
-                  (state->transmission_bounce >= kernel_data.integrator.max_transmission_bounce))
-               {
-                       return 0.0f;
-               }
-               else if(state->bounce <= kernel_data.integrator.min_bounce) {
-                       return 1.0f;
-               }
-       }
+                               /* scattering */
+                               VolumeIntegrateResult result = VOLUME_PATH_ATTENUATED;
 
-       /* probalistic termination */
-       return average(throughput); /* todo: try using max here */
-}
+                               if(volume_segment.closure_flag & SD_SCATTER) {
+                                       bool all = kernel_data.integrator.sample_all_lights_indirect;
 
-__device_inline bool shadow_blocked(KernelGlobals *kg, PathState *state, Ray *ray, float3 *shadow)
-{
-       *shadow = make_float3(1.0f, 1.0f, 1.0f);
+                                       /* direct light sampling */
+                                       kernel_branched_path_volume_connect_light(kg, rng, &volume_sd,
+                                               throughput, &state, L, 1.0f, all, &volume_ray, &volume_segment);
 
-       if(ray->t == 0.0f)
-               return false;
-       
-       Intersection isect;
-#ifdef __HAIR__
-       bool result = scene_intersect(kg, ray, PATH_RAY_SHADOW_OPAQUE, &isect, NULL, 0.0f, 0.0f);
-#else
-       bool result = scene_intersect(kg, ray, PATH_RAY_SHADOW_OPAQUE, &isect);
-#endif
+                                       /* indirect sample. if we use distance sampling and take just
+                                        * one sample for direct and indirect light, we could share
+                                        * this computation, but makes code a bit complex */
+                                       float rphase = path_state_rng_1D_for_decision(kg, rng, &state, PRNG_PHASE);
+                                       float rscatter = path_state_rng_1D_for_decision(kg, rng, &state, PRNG_SCATTER_DISTANCE);
 
-#ifdef __TRANSPARENT_SHADOWS__
-       if(result && kernel_data.integrator.transparent_shadows) {
-               /* transparent shadows work in such a way to try to minimize overhead
-                * in cases where we don't need them. after a regular shadow ray is
-                * cast we check if the hit primitive was potentially transparent, and
-                * only in that case start marching. this gives on extra ray cast for
-                * the cases were we do want transparency.
-                *
-                * also note that for this to work correct, multi close sampling must
-                * be used, since we don't pass a random number to shader_eval_surface */
-               if(shader_transparent_shadow(kg, &isect)) {
-                       float3 throughput = make_float3(1.0f, 1.0f, 1.0f);
-                       float3 Pend = ray->P + ray->D*ray->t;
-                       int bounce = state->transparent_bounce;
-
-                       for(;;) {
-                               if(bounce >= kernel_data.integrator.transparent_max_bounce) {
-                                       return true;
+                                       result = kernel_volume_decoupled_scatter(kg,
+                                               &state, &volume_ray, &volume_sd, &throughput,
+                                               rphase, rscatter, &volume_segment, NULL, true);
                                }
-                               else if(bounce >= kernel_data.integrator.transparent_min_bounce) {
-                                       /* todo: get random number somewhere for probabilistic terminate */
-#if 0
-                                       float probability = average(throughput);
-                                       float terminate = 0.0f;
 
-                                       if(terminate >= probability)
-                                               return true;
+                               if(result != VOLUME_PATH_SCATTERED)
+                                       throughput *= volume_segment.accum_transmittance;
 
-                                       throughput /= probability;
-#endif
-                               }
+                               /* free cached steps */
+                               kernel_volume_decoupled_free(kg, &volume_segment);
 
-#ifdef __HAIR__
-                               if(!scene_intersect(kg, ray, PATH_RAY_SHADOW_TRANSPARENT, &isect, NULL, 0.0f, 0.0f)) {
-#else
-                               if(!scene_intersect(kg, ray, PATH_RAY_SHADOW_TRANSPARENT, &isect)) {
-#endif
-                                       *shadow *= throughput;
-                                       return false;
+                               if(result == VOLUME_PATH_SCATTERED) {
+                                       if(kernel_path_volume_bounce(kg, rng, &volume_sd, &throughput, &state, L, &ray))
+                                               continue;
+                                       else
+                                               break;
                                }
-
-                               if(!shader_transparent_shadow(kg, &isect))
-                                       return true;
-
-                               ShaderData sd;
-                               shader_setup_from_ray(kg, &sd, &isect, ray, state->bounce+1);
-                               shader_eval_surface(kg, &sd, 0.0f, PATH_RAY_SHADOW, SHADER_CONTEXT_SHADOW);
-
-                               throughput *= shader_bsdf_transparency(kg, &sd);
-
-                               ray->P = ray_offset(sd.P, -sd.Ng);
-                               if(ray->t != FLT_MAX)
-                                       ray->D = normalize_len(Pend - ray->P, &ray->t);
-
-                               bounce++;
                        }
-               }
-       }
-#endif
-
-       return result;
-}
-
-__device float4 kernel_path_integrate(KernelGlobals *kg, RNG *rng, int sample, Ray ray, __global float *buffer)
-{
-       /* initialize */
-       PathRadiance L;
-       float3 throughput = make_float3(1.0f, 1.0f, 1.0f);
-       float L_transparent = 0.0f;
-
-       path_radiance_init(&L, kernel_data.film.use_light_pass);
-
-       float min_ray_pdf = FLT_MAX;
-       float ray_pdf = 0.0f;
-#ifdef __LAMP_MIS__
-       float ray_t = 0.0f;
+                       else
 #endif
-       PathState state;
-       int rng_offset = PRNG_BASE_NUM;
-#ifdef __CMJ__
-       int num_samples = kernel_data.integrator.aa_samples;
-#else
-       int num_samples = 0;
+                       {
+                               /* integrate along volume segment with distance sampling */
+                               ShaderData volume_sd;
+                               VolumeIntegrateResult result = kernel_volume_integrate(
+                                       kg, &state, &volume_sd, &volume_ray, L, &throughput, rng, heterogeneous);
+
+#ifdef __VOLUME_SCATTER__
+                               if(result == VOLUME_PATH_SCATTERED) {
+                                       /* direct lighting */
+                                       kernel_path_volume_connect_light(kg, rng, &volume_sd, throughput, &state, L);
+
+                                       /* indirect light bounce */
+                                       if(kernel_path_volume_bounce(kg, rng, &volume_sd, &throughput, &state, L, &ray))
+                                               continue;
+                                       else
+                                               break;
+                               }
 #endif
-
-       path_state_init(&state);
-
-       /* path iteration */
-       for(;; rng_offset += PRNG_BOUNCE_NUM) {
-               /* intersect scene */
-               Intersection isect;
-               uint visibility = path_state_ray_visibility(kg, &state);
-
-#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_init(*rng + rng_offset + sample*0x51633e2d);
-               }
-
-               bool hit = scene_intersect(kg, &ray, visibility, &isect, &lcg_state, difl, extmax);
-#else
-               bool hit = scene_intersect(kg, &ray, visibility, &isect);
-#endif
-
-#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 - ray_t*ray.D;
-                       ray_t += isect.t;
-                       light_ray.D = ray.D;
-                       light_ray.t = ray_t;
-                       light_ray.time = ray.time;
-                       light_ray.dD = ray.dD;
-                       light_ray.dP = ray.dP;
-
-                       /* intersect with lamp */
-                       float light_t = path_rng_1D(kg, rng, sample, num_samples, rng_offset + PRNG_LIGHT);
-                       float3 emission;
-
-                       if(indirect_lamp_emission(kg, &light_ray, state.flag, ray_pdf, light_t, &emission, state.bounce))
-                               path_radiance_accum_emission(&L, throughput, emission, state.bounce);
                }
 #endif
 
                if(!hit) {
-                       /* 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
-                                       break;
-                       }
-
 #ifdef __BACKGROUND__
                        /* sample background shader */
-                       float3 L_background = indirect_background(kg, &ray, state.flag, ray_pdf, state.bounce);
-                       path_radiance_accum_background(&L, throughput, L_background, state.bounce);
+                       float3 L_background = indirect_background(kg, &state, &ray);
+                       path_radiance_accum_background(L, throughput, L_background, state.bounce);
 #endif
 
                        break;
@@ -325,37 +175,17 @@ __device float4 kernel_path_integrate(KernelGlobals *kg, RNG *rng, int sample, R
 
                /* setup shading */
                ShaderData sd;
-               shader_setup_from_ray(kg, &sd, &isect, &ray, state.bounce);
-               float rbsdf = path_rng_1D(kg, rng, sample, num_samples, rng_offset + PRNG_BSDF);
-               shader_eval_surface(kg, &sd, rbsdf, state.flag, SHADER_CONTEXT_MAIN);
-
-               /* holdout */
-#ifdef __HOLDOUT__
-               if((sd.flag & (SD_HOLDOUT|SD_HOLDOUT_MASK)) && (state.flag & PATH_RAY_CAMERA)) {
-                       if(kernel_data.background.transparent) {
-                               float3 holdout_weight;
-                               
-                               if(sd.flag & SD_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 */
-                               L_transparent += average(holdout_weight*throughput);
-                       }
-
-                       if(sd.flag & SD_HOLDOUT_MASK)
-                               break;
-               }
+               shader_setup_from_ray(kg, &sd, &isect, &ray, state.bounce, state.transparent_bounce);
+               float rbsdf = path_state_rng_1D_for_decision(kg, rng, &state, PRNG_BSDF);
+               shader_eval_surface(kg, &sd, rbsdf, state.flag, SHADER_CONTEXT_INDIRECT);
+#ifdef __BRANCHED_PATH__
+               shader_merge_closures(&sd);
 #endif
 
-               /* holdout mask objects do not write data passes */
-               kernel_write_data_passes(kg, buffer, &L, &sd, sample, state.flag, throughput);
-
                /* blurring of bsdf after bounces, for rays that have a small likelihood
                 * of following this particular path (diffuse, rough glossy) */
                if(kernel_data.integrator.filter_glossy != FLT_MAX) {
-                       float blur_pdf = kernel_data.integrator.filter_glossy*min_ray_pdf;
+                       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;
@@ -366,22 +196,21 @@ __device float4 kernel_path_integrate(KernelGlobals *kg, RNG *rng, int sample, R
 #ifdef __EMISSION__
                /* emission */
                if(sd.flag & SD_EMISSION) {
-                       /* todo: is isect.t wrong here for transparent surfaces? */
-                       float3 emission = indirect_primitive_emission(kg, &sd, isect.t, state.flag, ray_pdf);
-                       path_radiance_accum_emission(&L, throughput, emission, state.bounce);
+                       float3 emission = indirect_primitive_emission(kg, &sd, isect.t, state.flag, state.ray_pdf);
+                       path_radiance_accum_emission(L, throughput, emission, state.bounce);
                }
 #endif
 
                /* 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 */
-               float probability = path_state_terminate_probability(kg, &state, throughput);
+               float probability = path_state_terminate_probability(kg, &state, throughput*num_samples);
 
                if(probability == 0.0f) {
                        break;
                }
                else if(probability != 1.0f) {
-                       float terminate = path_rng_1D(kg, rng, sample, num_samples, rng_offset + PRNG_TERMINATE);
+                       float terminate = path_state_rng_1D_for_decision(kg, rng, &state, PRNG_TERMINATE);
 
                        if(terminate >= probability)
                                break;
@@ -389,44 +218,18 @@ __device float4 kernel_path_integrate(KernelGlobals *kg, RNG *rng, int sample, R
                        throughput /= probability;
                }
 
-#ifdef __SUBSURFACE__
-               /* bssrdf scatter to a different location on the same object, replacing
-                * the closures with a diffuse BSDF */
-               if(sd.flag & SD_BSSRDF) {
-                       float bssrdf_probability;
-                       ShaderClosure *sc = subsurface_scatter_pick_closure(kg, &sd, &bssrdf_probability);
-
-                       /* modify throughput for picking bssrdf or bsdf */
-                       throughput *= bssrdf_probability;
-
-                       /* do bssrdf scatter step if we picked a bssrdf closure */
-                       if(sc) {
-                               uint lcg_state = lcg_init(*rng + rng_offset + sample*0x68bc21eb);
-
-                               if(old_subsurface_scatter_use(&sd)) {
-                                       old_subsurface_scatter_step(kg, &sd, state.flag, sc, &lcg_state, false);
-                               }
-                               else {
-                                       float bssrdf_u, bssrdf_v;
-                                       path_rng_2D(kg, rng, sample, num_samples, rng_offset + PRNG_BSDF_U, &bssrdf_u, &bssrdf_v);
-                                       subsurface_scatter_step(kg, &sd, state.flag, sc, &lcg_state, bssrdf_u, bssrdf_v, false);
-                               }
-                       }
-               }
-#endif
-
 #ifdef __AO__
                /* ambient occlusion */
                if(kernel_data.integrator.use_ambient_occlusion || (sd.flag & SD_AO)) {
-                       /* todo: solve correlation */
                        float bsdf_u, bsdf_v;
-                       path_rng_2D(kg, rng, sample, num_samples, rng_offset + PRNG_BSDF_U, &bsdf_u, &bsdf_v);
+                       path_state_rng_2D(kg, rng, &state, PRNG_BSDF_U, &bsdf_u, &bsdf_v);
 
                        float ao_factor = kernel_data.background.ao_factor;
                        float3 ao_N;
                        float3 ao_bsdf = shader_bsdf_ao(kg, &sd, ao_factor, &ao_N);
                        float3 ao_D;
                        float ao_pdf;
+                       float3 ao_alpha = make_float3(0.0f, 0.0f, 0.0f);
 
                        sample_cos_hemisphere(ao_N, bsdf_u, bsdf_v, &ao_D, &ao_pdf);
 
@@ -444,124 +247,256 @@ __device float4 kernel_path_integrate(KernelGlobals *kg, RNG *rng, int sample, R
                                light_ray.dD = differential3_zero();
 
                                if(!shadow_blocked(kg, &state, &light_ray, &ao_shadow))
-                                       path_radiance_accum_ao(&L, throughput, ao_bsdf, ao_shadow, state.bounce);
+                                       path_radiance_accum_ao(L, throughput, ao_alpha, ao_bsdf, ao_shadow, state.bounce);
                        }
                }
 #endif
 
-#ifdef __EMISSION__
-               if(kernel_data.integrator.use_direct_light) {
-                       /* sample illumination from lights to find path contribution */
-                       if(sd.flag & SD_BSDF_HAS_EVAL) {
-                               float light_t = path_rng_1D(kg, rng, sample, num_samples, rng_offset + PRNG_LIGHT);
-#ifdef __MULTI_CLOSURE__
-                               float light_o = 0.0f;
-#else
-                               float light_o = path_rng_1D(kg, rng, sample, num_samples, rng_offset + PRNG_LIGHT_F);
-#endif
-                               float light_u, light_v;
-                               path_rng_2D(kg, rng, sample, num_samples, rng_offset + PRNG_LIGHT_U, &light_u, &light_v);
+#ifdef __SUBSURFACE__
+               /* bssrdf scatter to a different location on the same object, replacing
+                * the closures with a diffuse BSDF */
+               if(sd.flag & SD_BSSRDF) {
+                       float bssrdf_probability;
+                       ShaderClosure *sc = subsurface_scatter_pick_closure(kg, &sd, &bssrdf_probability);
 
-                               Ray light_ray;
-                               BsdfEval L_light;
-                               bool is_lamp;
+                       /* modify throughput for picking bssrdf or bsdf */
+                       throughput *= bssrdf_probability;
 
-#ifdef __OBJECT_MOTION__
-                               light_ray.time = sd.time;
-#endif
+                       /* do bssrdf scatter step if we picked a bssrdf closure */
+                       if(sc) {
+                               uint lcg_state = lcg_state_init(rng, &state, 0x68bc21eb);
 
-                               if(direct_emission(kg, &sd, -1, light_t, light_o, light_u, light_v, &light_ray, &L_light, &is_lamp, state.bounce)) {
-                                       /* trace shadow ray */
-                                       float3 shadow;
+                               float bssrdf_u, bssrdf_v;
+                               path_state_rng_2D(kg, rng, &state, PRNG_BSDF_U, &bssrdf_u, &bssrdf_v);
+                               subsurface_scatter_step(kg, &sd, state.flag, sc, &lcg_state, bssrdf_u, bssrdf_v, false);
 
-                                       if(!shadow_blocked(kg, &state, &light_ray, &shadow)) {
-                                               /* accumulate */
-                                               path_radiance_accum_light(&L, throughput, &L_light, shadow, 1.0f, state.bounce, is_lamp);
-                                       }
-                               }
+                               state.flag |= PATH_RAY_BSSRDF_ANCESTOR;
                        }
                }
 #endif
 
-               /* no BSDF? we can stop here */
-               if(!(sd.flag & SD_BSDF))
+#if defined(__EMISSION__) && defined(__BRANCHED_PATH__)
+               if(kernel_data.integrator.use_direct_light) {
+                       bool all = kernel_data.integrator.sample_all_lights_indirect;
+                       kernel_branched_path_surface_connect_light(kg, rng, &sd, &state, throughput, 1.0f, L, all);
+               }
+#endif
+
+               if(!kernel_path_surface_bounce(kg, rng, &sd, &throughput, &state, L, &ray))
                        break;
+       }
+}
 
-               /* sample BSDF */
-               float bsdf_pdf;
-               BsdfEval bsdf_eval;
-               float3 bsdf_omega_in;
-               differential3 bsdf_domega_in;
-               float bsdf_u, bsdf_v;
-               path_rng_2D(kg, rng, sample, num_samples, rng_offset + PRNG_BSDF_U, &bsdf_u, &bsdf_v);
-               int label;
+ccl_device void kernel_path_ao(KernelGlobals *kg, ShaderData *sd, PathRadiance *L, PathState *state, RNG *rng, float3 throughput)
+{
+       /* todo: solve correlation */
+       float bsdf_u, bsdf_v;
 
-               label = shader_bsdf_sample(kg, &sd, bsdf_u, bsdf_v, &bsdf_eval,
-                       &bsdf_omega_in, &bsdf_domega_in, &bsdf_pdf);
+       path_state_rng_2D(kg, rng, state, PRNG_BSDF_U, &bsdf_u, &bsdf_v);
 
-               if(bsdf_pdf == 0.0f || bsdf_eval_is_zero(&bsdf_eval))
-                       break;
+       float ao_factor = kernel_data.background.ao_factor;
+       float3 ao_N;
+       float3 ao_bsdf = shader_bsdf_ao(kg, sd, ao_factor, &ao_N);
+       float3 ao_D;
+       float ao_pdf;
+       float3 ao_alpha = shader_bsdf_alpha(kg, sd);
 
-               /* modify throughput */
-               path_radiance_bsdf_bounce(&L, &throughput, &bsdf_eval, bsdf_pdf, state.bounce, label);
+       sample_cos_hemisphere(ao_N, bsdf_u, bsdf_v, &ao_D, &ao_pdf);
 
-               /* set labels */
-               if(!(label & LABEL_TRANSPARENT)) {
-                       ray_pdf = bsdf_pdf;
-#ifdef __LAMP_MIS__
-                       ray_t = 0.0f;
+       if(dot(sd->Ng, ao_D) > 0.0f && ao_pdf != 0.0f) {
+               Ray light_ray;
+               float3 ao_shadow;
+
+               light_ray.P = ray_offset(sd->P, sd->Ng);
+               light_ray.D = ao_D;
+               light_ray.t = kernel_data.background.ao_distance;
+#ifdef __OBJECT_MOTION__
+               light_ray.time = sd->time;
 #endif
-                       min_ray_pdf = fminf(bsdf_pdf, min_ray_pdf);
-               }
+               light_ray.dP = sd->dP;
+               light_ray.dD = differential3_zero();
 
-               /* update path state */
-               path_state_next(kg, &state, label);
+               if(!shadow_blocked(kg, state, &light_ray, &ao_shadow))
+                       path_radiance_accum_ao(L, throughput, ao_alpha, ao_bsdf, ao_shadow, state->bounce);
+       }
+}
 
-               /* setup ray */
-               ray.P = ray_offset(sd.P, (label & LABEL_TRANSMIT)? -sd.Ng: sd.Ng);
-               ray.D = bsdf_omega_in;
+ccl_device void kernel_branched_path_ao(KernelGlobals *kg, ShaderData *sd, PathRadiance *L, PathState *state, RNG *rng, float3 throughput)
+{
+       int num_samples = kernel_data.integrator.ao_samples;
+       float num_samples_inv = 1.0f/num_samples;
+       float ao_factor = kernel_data.background.ao_factor;
+       float3 ao_N;
+       float3 ao_bsdf = shader_bsdf_ao(kg, sd, ao_factor, &ao_N);
+       float3 ao_alpha = shader_bsdf_alpha(kg, sd);
+
+       for(int j = 0; j < num_samples; j++) {
+               float bsdf_u, bsdf_v;
+               path_branched_rng_2D(kg, rng, state, j, num_samples, PRNG_BSDF_U, &bsdf_u, &bsdf_v);
 
-               if(state.bounce == 0)
-                       ray.t -= sd.ray_length; /* clipping works through transparent */
-               else
-                       ray.t = FLT_MAX;
+               float3 ao_D;
+               float ao_pdf;
 
-#ifdef __RAY_DIFFERENTIALS__
-               ray.dP = sd.dP;
-               ray.dD = bsdf_domega_in;
+               sample_cos_hemisphere(ao_N, bsdf_u, bsdf_v, &ao_D, &ao_pdf);
+
+               if(dot(sd->Ng, ao_D) > 0.0f && ao_pdf != 0.0f) {
+                       Ray light_ray;
+                       float3 ao_shadow;
+
+                       light_ray.P = ray_offset(sd->P, sd->Ng);
+                       light_ray.D = ao_D;
+                       light_ray.t = kernel_data.background.ao_distance;
+#ifdef __OBJECT_MOTION__
+                       light_ray.time = sd->time;
 #endif
+                       light_ray.dP = sd->dP;
+                       light_ray.dD = differential3_zero();
+
+                       if(!shadow_blocked(kg, state, &light_ray, &ao_shadow))
+                               path_radiance_accum_ao(L, throughput*num_samples_inv, ao_alpha, ao_bsdf, ao_shadow, state->bounce);
+               }
        }
+}
 
-       float3 L_sum = path_radiance_sum(kg, &L);
+#ifdef __SUBSURFACE__
 
-#ifdef __CLAMP_SAMPLE__
-       path_radiance_clamp(&L, &L_sum, kernel_data.integrator.sample_clamp);
-#endif
+#ifdef __VOLUME__
+ccl_device void kernel_path_subsurface_update_volume_stack(KernelGlobals *kg,
+                                                           Ray *ray,
+                                                           VolumeStack *stack)
+{
+       kernel_assert(kernel_data.integrator.use_volumes);
 
-       kernel_write_light_passes(kg, buffer, &L, sample);
+       Ray volume_ray = *ray;
+       Intersection isect;
+       const float3 Pend = volume_ray.P + volume_ray.D*volume_ray.t;
 
-       return make_float4(L_sum.x, L_sum.y, L_sum.z, 1.0f - L_transparent);
+       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
 
-#ifdef __BRANCHED_PATH__
-
-__device void kernel_path_indirect(KernelGlobals *kg, RNG *rng, int sample, Ray ray, __global float *buffer,
-       float3 throughput, int num_samples, int num_total_samples,
-       float min_ray_pdf, float ray_pdf, PathState state, int rng_offset, PathRadiance *L)
+ccl_device bool kernel_path_subsurface_scatter(KernelGlobals *kg, ShaderData *sd, PathRadiance *L, PathState *state, RNG *rng, Ray *ray, float3 *throughput)
 {
+       float bssrdf_probability;
+       ShaderClosure *sc = subsurface_scatter_pick_closure(kg, sd, &bssrdf_probability);
+
+       /* modify throughput for picking bssrdf or bsdf */
+       *throughput *= bssrdf_probability;
+
+       /* do bssrdf scatter step if we picked a bssrdf closure */
+       if(sc) {
+               uint lcg_state = lcg_state_init(rng, state, 0x68bc21eb);
+
+               ShaderData bssrdf_sd[BSSRDF_MAX_HITS];
+               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;
+               bool need_update_volume_stack = kernel_data.integrator.use_volumes &&
+                                               sd->flag & SD_OBJECT_INTERSECTS_VOLUME;
+#endif
+
+               /* compute lighting with the BSDF closure */
+               for(int hit = 0; hit < num_hits; hit++) {
+                       float3 tp = *throughput;
+                       PathState hit_state = *state;
+                       Ray hit_ray = *ray;
+
+                       hit_state.flag |= PATH_RAY_BSSRDF_ANCESTOR;
+                       hit_state.rng_offset += PRNG_BOUNCE_NUM;
+                       
+                       kernel_path_surface_connect_light(kg, rng, &bssrdf_sd[hit], tp, state, L);
+
+                       if(kernel_path_surface_bounce(kg, rng, &bssrdf_sd[hit], &tp, &hit_state, L, &hit_ray)) {
 #ifdef __LAMP_MIS__
-       float ray_t = 0.0f;
+                               hit_state.ray_t = 0.0f;
+#endif
+
+#ifdef __VOLUME__
+                               if(need_update_volume_stack) {
+                                       /* 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
+                                * for the next samples */
+                               path_radiance_sum_indirect(L);
+                               path_radiance_reset_indirect(L);
+                       }
+               }
+               return true;
+       }
+       return false;
+}
 #endif
 
+ccl_device float4 kernel_path_integrate(KernelGlobals *kg, RNG *rng, int sample, Ray ray, ccl_global float *buffer)
+{
+       /* initialize */
+       PathRadiance L;
+       float3 throughput = make_float3(1.0f, 1.0f, 1.0f);
+       float L_transparent = 0.0f;
+
+       path_radiance_init(&L, kernel_data.film.use_light_pass);
+
+       PathState state;
+       path_state_init(kg, &state, rng, sample, &ray);
+
        /* path iteration */
-       for(;; rng_offset += PRNG_BOUNCE_NUM) {
+       for(;;) {
                /* intersect scene */
                Intersection isect;
                uint visibility = path_state_ray_visibility(kg, &state);
+
 #ifdef __HAIR__
-               bool hit = scene_intersect(kg, &ray, visibility, &isect, NULL, 0.0f, 0.0f);
+               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(rng, &state, 0x51633e2d);
+               }
+
+               bool hit = scene_intersect(kg, &ray, visibility, &isect, &lcg_state, difl, extmax);
 #else
-               bool hit = scene_intersect(kg, &ray, visibility, &isect);
+               bool hit = scene_intersect(kg, &ray, visibility, &isect, NULL, 0.0f, 0.0f);
 #endif
 
 #ifdef __LAMP_MIS__
@@ -569,44 +504,160 @@ __device void kernel_path_indirect(KernelGlobals *kg, RNG *rng, int sample, Ray
                        /* ray starting from previous non-transparent bounce */
                        Ray light_ray;
 
-                       light_ray.P = ray.P - ray_t*ray.D;
-                       ray_t += isect.t;
+                       light_ray.P = ray.P - state.ray_t*ray.D;
+                       state.ray_t += isect.t;
                        light_ray.D = ray.D;
-                       light_ray.t = ray_t;
+                       light_ray.t = state.ray_t;
                        light_ray.time = ray.time;
                        light_ray.dD = ray.dD;
                        light_ray.dP = ray.dP;
 
                        /* intersect with lamp */
-                       float light_t = path_rng_1D(kg, rng, sample, num_total_samples, rng_offset + PRNG_LIGHT);
                        float3 emission;
 
-                       if(indirect_lamp_emission(kg, &light_ray, state.flag, ray_pdf, light_t, &emission, state.bounce))
-                               path_radiance_accum_emission(L, throughput, emission, state.bounce);
+                       if(indirect_lamp_emission(kg, &state, &light_ray, &emission))
+                               path_radiance_accum_emission(&L, throughput, emission, state.bounce);
+               }
+#endif
+
+#ifdef __VOLUME__
+               /* volume attenuation, emission, scatter */
+               if(state.volume_stack[0].shader != SHADER_NONE) {
+                       Ray volume_ray = ray;
+                       volume_ray.t = (hit)? isect.t: FLT_MAX;
+
+                       bool heterogeneous = volume_stack_is_heterogeneous(kg, state.volume_stack);
+
+#ifdef __VOLUME_DECOUPLED__
+                       int sampling_method = volume_stack_sampling_method(kg, state.volume_stack);
+                       bool decoupled = kernel_volume_use_decoupled(kg, heterogeneous, true, sampling_method);
+
+                       if(decoupled) {
+                               /* cache steps along volume for repeated sampling */
+                               VolumeSegment volume_segment;
+                               ShaderData volume_sd;
+
+                               shader_setup_from_volume(kg, &volume_sd, &volume_ray, state.bounce, state.transparent_bounce);
+                               kernel_volume_decoupled_record(kg, &state,
+                                       &volume_ray, &volume_sd, &volume_segment, heterogeneous);
+
+                               volume_segment.sampling_method = sampling_method;
+
+                               /* emission */
+                               if(volume_segment.closure_flag & SD_EMISSION)
+                                       path_radiance_accum_emission(&L, throughput, volume_segment.accum_emission, state.bounce);
+
+                               /* scattering */
+                               VolumeIntegrateResult result = VOLUME_PATH_ATTENUATED;
+
+                               if(volume_segment.closure_flag & SD_SCATTER) {
+                                       bool all = false;
+
+                                       /* direct light sampling */
+                                       kernel_branched_path_volume_connect_light(kg, rng, &volume_sd,
+                                               throughput, &state, &L, 1.0f, all, &volume_ray, &volume_segment);
+
+                                       /* indirect sample. if we use distance sampling and take just
+                                        * one sample for direct and indirect light, we could share
+                                        * this computation, but makes code a bit complex */
+                                       float rphase = path_state_rng_1D_for_decision(kg, rng, &state, PRNG_PHASE);
+                                       float rscatter = path_state_rng_1D_for_decision(kg, rng, &state, PRNG_SCATTER_DISTANCE);
+
+                                       result = kernel_volume_decoupled_scatter(kg,
+                                               &state, &volume_ray, &volume_sd, &throughput,
+                                               rphase, rscatter, &volume_segment, NULL, true);
+                               }
+
+                               if(result != VOLUME_PATH_SCATTERED)
+                                       throughput *= volume_segment.accum_transmittance;
+
+                               /* free cached steps */
+                               kernel_volume_decoupled_free(kg, &volume_segment);
+
+                               if(result == VOLUME_PATH_SCATTERED) {
+                                       if(kernel_path_volume_bounce(kg, rng, &volume_sd, &throughput, &state, &L, &ray))
+                                               continue;
+                                       else
+                                               break;
+                               }
+                       }
+                       else 
+#endif
+                       {
+                               /* integrate along volume segment with distance sampling */
+                               ShaderData volume_sd;
+                               VolumeIntegrateResult result = kernel_volume_integrate(
+                                       kg, &state, &volume_sd, &volume_ray, &L, &throughput, rng, heterogeneous);
+
+#ifdef __VOLUME_SCATTER__
+                               if(result == VOLUME_PATH_SCATTERED) {
+                                       /* direct lighting */
+                                       kernel_path_volume_connect_light(kg, rng, &volume_sd, throughput, &state, &L);
+
+                                       /* indirect light bounce */
+                                       if(kernel_path_volume_bounce(kg, rng, &volume_sd, &throughput, &state, &L, &ray))
+                                               continue;
+                                       else
+                                               break;
+                               }
+#endif
+                       }
                }
 #endif
 
                if(!hit) {
+                       /* 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
+                                       break;
+                       }
+
 #ifdef __BACKGROUND__
                        /* sample background shader */
-                       float3 L_background = indirect_background(kg, &ray, state.flag, ray_pdf, state.bounce);
-                       path_radiance_accum_background(L, throughput, L_background, state.bounce);
+                       float3 L_background = indirect_background(kg, &state, &ray);
+                       path_radiance_accum_background(&L, throughput, L_background, state.bounce);
 #endif
 
-                       break;
+                       break;
+               }
+
+               /* setup shading */
+               ShaderData sd;
+               shader_setup_from_ray(kg, &sd, &isect, &ray, state.bounce, state.transparent_bounce);
+               float rbsdf = path_state_rng_1D_for_decision(kg, rng, &state, PRNG_BSDF);
+               shader_eval_surface(kg, &sd, rbsdf, state.flag, SHADER_CONTEXT_MAIN);
+
+               /* holdout */
+#ifdef __HOLDOUT__
+               if((sd.flag & (SD_HOLDOUT|SD_HOLDOUT_MASK)) && (state.flag & PATH_RAY_CAMERA)) {
+                       if(kernel_data.background.transparent) {
+                               float3 holdout_weight;
+                               
+                               if(sd.flag & SD_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 */
+                               L_transparent += average(holdout_weight*throughput);
+                       }
+
+                       if(sd.flag & SD_HOLDOUT_MASK)
+                               break;
                }
+#endif
 
-               /* setup shading */
-               ShaderData sd;
-               shader_setup_from_ray(kg, &sd, &isect, &ray, state.bounce);
-               float rbsdf = path_rng_1D(kg, rng, sample, num_total_samples, rng_offset + PRNG_BSDF);
-               shader_eval_surface(kg, &sd, rbsdf, state.flag, SHADER_CONTEXT_INDIRECT);
-               shader_merge_closures(kg, &sd);
+               /* 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) */
                if(kernel_data.integrator.filter_glossy != FLT_MAX) {
-                       float blur_pdf = kernel_data.integrator.filter_glossy*min_ray_pdf;
+                       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;
@@ -617,21 +668,22 @@ __device void kernel_path_indirect(KernelGlobals *kg, RNG *rng, int sample, Ray
 #ifdef __EMISSION__
                /* emission */
                if(sd.flag & SD_EMISSION) {
-                       float3 emission = indirect_primitive_emission(kg, &sd, isect.t, state.flag, ray_pdf);
-                       path_radiance_accum_emission(L, throughput, emission, state.bounce);
+                       /* todo: is isect.t wrong here for transparent surfaces? */
+                       float3 emission = indirect_primitive_emission(kg, &sd, isect.t, state.flag, state.ray_pdf);
+                       path_radiance_accum_emission(&L, throughput, emission, state.bounce);
                }
 #endif
 
                /* 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 */
-               float probability = path_state_terminate_probability(kg, &state, throughput*num_samples);
+               float probability = path_state_terminate_probability(kg, &state, throughput);
 
                if(probability == 0.0f) {
                        break;
                }
                else if(probability != 1.0f) {
-                       float terminate = path_rng_1D(kg, rng, sample, num_total_samples, rng_offset + PRNG_TERMINATE);
+                       float terminate = path_state_rng_1D_for_decision(kg, rng, &state, PRNG_TERMINATE);
 
                        if(terminate >= probability)
                                break;
@@ -639,257 +691,44 @@ __device void kernel_path_indirect(KernelGlobals *kg, RNG *rng, int sample, Ray
                        throughput /= probability;
                }
 
-#ifdef __SUBSURFACE__
-               /* bssrdf scatter to a different location on the same object, replacing
-                * the closures with a diffuse BSDF */
-               if(sd.flag & SD_BSSRDF) {
-                       float bssrdf_probability;
-                       ShaderClosure *sc = subsurface_scatter_pick_closure(kg, &sd, &bssrdf_probability);
-
-                       /* modify throughput for picking bssrdf or bsdf */
-                       throughput *= bssrdf_probability;
-
-                       /* do bssrdf scatter step if we picked a bssrdf closure */
-                       if(sc) {
-                               uint lcg_state = lcg_init(*rng + rng_offset + sample*0x68bc21eb);
-
-                               if(old_subsurface_scatter_use(&sd)) {
-                                       old_subsurface_scatter_step(kg, &sd, state.flag, sc, &lcg_state, false);
-                               }
-                               else {
-                                       float bssrdf_u, bssrdf_v;
-                                       path_rng_2D(kg, rng, sample, num_total_samples, rng_offset + PRNG_BSDF_U, &bssrdf_u, &bssrdf_v);
-                                       subsurface_scatter_step(kg, &sd, state.flag, sc, &lcg_state, bssrdf_u, bssrdf_v, false);
-                               }
-                       }
-               }
-#endif
-
 #ifdef __AO__
                /* ambient occlusion */
                if(kernel_data.integrator.use_ambient_occlusion || (sd.flag & SD_AO)) {
-                       float bsdf_u, bsdf_v;
-                       path_rng_2D(kg, rng, sample, num_total_samples, rng_offset + PRNG_BSDF_U, &bsdf_u, &bsdf_v);
-
-                       float ao_factor = kernel_data.background.ao_factor;
-                       float3 ao_N;
-                       float3 ao_bsdf = shader_bsdf_ao(kg, &sd, ao_factor, &ao_N);
-                       float3 ao_D;
-                       float ao_pdf;
-
-                       sample_cos_hemisphere(ao_N, bsdf_u, bsdf_v, &ao_D, &ao_pdf);
-
-                       if(dot(sd.Ng, ao_D) > 0.0f && ao_pdf != 0.0f) {
-                               Ray light_ray;
-                               float3 ao_shadow;
-
-                               light_ray.P = ray_offset(sd.P, sd.Ng);
-                               light_ray.D = ao_D;
-                               light_ray.t = kernel_data.background.ao_distance;
-#ifdef __OBJECT_MOTION__
-                               light_ray.time = sd.time;
-#endif
-                               light_ray.dP = sd.dP;
-                               light_ray.dD = differential3_zero();
-
-                               if(!shadow_blocked(kg, &state, &light_ray, &ao_shadow))
-                                       path_radiance_accum_ao(L, throughput, ao_bsdf, ao_shadow, state.bounce);
-                       }
+                       kernel_path_ao(kg, &sd, &L, &state, rng, throughput);
                }
 #endif
 
-#ifdef __EMISSION__
-               if(kernel_data.integrator.use_direct_light) {
-                       /* sample illumination from lights to find path contribution */
-                       if(sd.flag & SD_BSDF_HAS_EVAL) {
-                               float light_t = path_rng_1D(kg, rng, sample, num_total_samples, rng_offset + PRNG_LIGHT);
-#ifdef __MULTI_CLOSURE__
-                               float light_o = 0.0f;
-#else
-                               float light_o = path_rng_1D(kg, rng, sample, num_total_samples, rng_offset + PRNG_LIGHT_F);
-#endif
-                               float light_u, light_v;
-                               path_rng_2D(kg, rng, sample, num_total_samples, rng_offset + PRNG_LIGHT_U, &light_u, &light_v);
-
-                               Ray light_ray;
-                               BsdfEval L_light;
-                               bool is_lamp;
-
-#ifdef __OBJECT_MOTION__
-                               light_ray.time = sd.time;
-#endif
-
-                               /* sample random light */
-                               if(direct_emission(kg, &sd, -1, light_t, light_o, light_u, light_v, &light_ray, &L_light, &is_lamp, state.bounce)) {
-                                       /* trace shadow ray */
-                                       float3 shadow;
-
-                                       if(!shadow_blocked(kg, &state, &light_ray, &shadow)) {
-                                               /* accumulate */
-                                               path_radiance_accum_light(L, throughput, &L_light, shadow, 1.0f, state.bounce, is_lamp);
-                                       }
-                               }
-                       }
+#ifdef __SUBSURFACE__
+               /* bssrdf scatter to a different location on the same object, replacing
+                * the closures with a diffuse BSDF */
+               if(sd.flag & SD_BSSRDF) {
+                       if(kernel_path_subsurface_scatter(kg, &sd, &L, &state, rng, &ray, &throughput))
+                               break;
                }
 #endif
 
-               /* no BSDF? we can stop here */
-               if(!(sd.flag & SD_BSDF))
-                       break;
-
-               /* sample BSDF */
-               float bsdf_pdf;
-               BsdfEval bsdf_eval;
-               float3 bsdf_omega_in;
-               differential3 bsdf_domega_in;
-               float bsdf_u, bsdf_v;
-               path_rng_2D(kg, rng, sample, num_total_samples, rng_offset + PRNG_BSDF_U, &bsdf_u, &bsdf_v);
-               int label;
-
-               label = shader_bsdf_sample(kg, &sd, bsdf_u, bsdf_v, &bsdf_eval,
-                       &bsdf_omega_in, &bsdf_domega_in, &bsdf_pdf);
+               /* direct lighting */
+               kernel_path_surface_connect_light(kg, rng, &sd, throughput, &state, &L);
 
-               if(bsdf_pdf == 0.0f || bsdf_eval_is_zero(&bsdf_eval))
+               /* compute direct lighting and next bounce */
+               if(!kernel_path_surface_bounce(kg, rng, &sd, &throughput, &state, &L, &ray))
                        break;
-
-               /* modify throughput */
-               path_radiance_bsdf_bounce(L, &throughput, &bsdf_eval, bsdf_pdf, state.bounce, label);
-
-               /* set labels */
-               if(!(label & LABEL_TRANSPARENT)) {
-                       ray_pdf = bsdf_pdf;
-#ifdef __LAMP_MIS__
-                       ray_t = 0.0f;
-#endif
-                       min_ray_pdf = fminf(bsdf_pdf, min_ray_pdf);
-               }
-
-               /* update path state */
-               path_state_next(kg, &state, label);
-
-               /* setup ray */
-               ray.P = ray_offset(sd.P, (label & LABEL_TRANSMIT)? -sd.Ng: sd.Ng);
-               ray.D = bsdf_omega_in;
-               ray.t = FLT_MAX;
-#ifdef __RAY_DIFFERENTIALS__
-               ray.dP = sd.dP;
-               ray.dD = bsdf_domega_in;
-#endif
-       }
-}
-
-__device_noinline void kernel_branched_path_integrate_lighting(KernelGlobals *kg, RNG *rng,
-       int sample, int aa_samples,
-       ShaderData *sd, float3 throughput, float num_samples_adjust,
-       float min_ray_pdf, float ray_pdf, PathState state,
-       int rng_offset, PathRadiance *L, __global float *buffer)
-{
-#ifdef __AO__
-       /* ambient occlusion */
-       if(kernel_data.integrator.use_ambient_occlusion || (sd->flag & SD_AO)) {
-               int num_samples = ceil_to_int(kernel_data.integrator.ao_samples*num_samples_adjust);
-               float num_samples_inv = num_samples_adjust/num_samples;
-               float ao_factor = kernel_data.background.ao_factor;
-               float3 ao_N;
-               float3 ao_bsdf = shader_bsdf_ao(kg, sd, ao_factor, &ao_N);
-
-               for(int j = 0; j < num_samples; j++) {
-                       float bsdf_u, bsdf_v;
-                       path_rng_2D(kg, rng, sample*num_samples + j, aa_samples*num_samples, rng_offset + PRNG_BSDF_U, &bsdf_u, &bsdf_v);
-
-                       float3 ao_D;
-                       float ao_pdf;
-
-                       sample_cos_hemisphere(ao_N, bsdf_u, bsdf_v, &ao_D, &ao_pdf);
-
-                       if(dot(sd->Ng, ao_D) > 0.0f && ao_pdf != 0.0f) {
-                               Ray light_ray;
-                               float3 ao_shadow;
-
-                               light_ray.P = ray_offset(sd->P, sd->Ng);
-                               light_ray.D = ao_D;
-                               light_ray.t = kernel_data.background.ao_distance;
-#ifdef __OBJECT_MOTION__
-                               light_ray.time = sd->time;
-#endif
-                               light_ray.dP = sd->dP;
-                               light_ray.dD = differential3_zero();
-
-                               if(!shadow_blocked(kg, &state, &light_ray, &ao_shadow))
-                                       path_radiance_accum_ao(L, throughput*num_samples_inv, ao_bsdf, ao_shadow, state.bounce);
-                       }
-               }
        }
-#endif
-
-
-#ifdef __EMISSION__
-       /* sample illumination from lights to find path contribution */
-       if(sd->flag & SD_BSDF_HAS_EVAL) {
-               Ray light_ray;
-               BsdfEval L_light;
-               bool is_lamp;
-
-#ifdef __OBJECT_MOTION__
-               light_ray.time = sd->time;
-#endif
 
-               /* lamp sampling */
-               for(int i = 0; i < kernel_data.integrator.num_all_lights; i++) {
-                       int num_samples = ceil_to_int(num_samples_adjust*light_select_num_samples(kg, i));
-                       float num_samples_inv = num_samples_adjust/(num_samples*kernel_data.integrator.num_all_lights);
-                       RNG lamp_rng = cmj_hash(*rng, i);
+       float3 L_sum = path_radiance_clamp_and_sum(kg, &L);
 
-                       if(kernel_data.integrator.pdf_triangles != 0.0f)
-                               num_samples_inv *= 0.5f;
-
-                       for(int j = 0; j < num_samples; j++) {
-                               float light_u, light_v;
-                               path_rng_2D(kg, &lamp_rng, sample*num_samples + j, aa_samples*num_samples, rng_offset + PRNG_LIGHT_U, &light_u, &light_v);
-
-                               if(direct_emission(kg, sd, i, 0.0f, 0.0f, light_u, light_v, &light_ray, &L_light, &is_lamp, state.bounce)) {
-                                       /* trace shadow ray */
-                                       float3 shadow;
-
-                                       if(!shadow_blocked(kg, &state, &light_ray, &shadow)) {
-                                               /* accumulate */
-                                               path_radiance_accum_light(L, throughput*num_samples_inv, &L_light, shadow, num_samples_inv, state.bounce, is_lamp);
-                                       }
-                               }
-                       }
-               }
-
-               /* mesh light sampling */
-               if(kernel_data.integrator.pdf_triangles != 0.0f) {
-                       int num_samples = ceil_to_int(num_samples_adjust*kernel_data.integrator.mesh_light_samples);
-                       float num_samples_inv = num_samples_adjust/num_samples;
-
-                       if(kernel_data.integrator.num_all_lights)
-                               num_samples_inv *= 0.5f;
-
-                       for(int j = 0; j < num_samples; j++) {
-                               float light_t = path_rng_1D(kg, rng, sample*num_samples + j, aa_samples*num_samples, rng_offset + PRNG_LIGHT);
-                               float light_u, light_v;
-                               path_rng_2D(kg, rng, sample*num_samples + j, aa_samples*num_samples, rng_offset + PRNG_LIGHT_U, &light_u, &light_v);
-
-                               /* only sample triangle lights */
-                               if(kernel_data.integrator.num_all_lights)
-                                       light_t = 0.5f*light_t;
+       kernel_write_light_passes(kg, buffer, &L, sample);
 
-                               if(direct_emission(kg, sd, -1, light_t, 0.0f, light_u, light_v, &light_ray, &L_light, &is_lamp, state.bounce)) {
-                                       /* trace shadow ray */
-                                       float3 shadow;
+       return make_float4(L_sum.x, L_sum.y, L_sum.z, 1.0f - L_transparent);
+}
 
-                                       if(!shadow_blocked(kg, &state, &light_ray, &shadow)) {
-                                               /* accumulate */
-                                               path_radiance_accum_light(L, throughput*num_samples_inv, &L_light, shadow, num_samples_inv, state.bounce, is_lamp);
-                                       }
-                               }
-                       }
-               }
-       }
-#endif
+#ifdef __BRANCHED_PATH__
 
+/* branched path tracing: bounce off surface and integrate indirect light */
+ccl_device_noinline void kernel_branched_path_surface_indirect_light(KernelGlobals *kg,
+       RNG *rng, ShaderData *sd, float3 throughput, float num_samples_adjust,
+       PathState *state, PathRadiance *L)
+{
        for(int i = 0; i< sd->num_closure; i++) {
                const ShaderClosure *sc = &sd->closure[i];
 
@@ -901,8 +740,10 @@ __device_noinline void kernel_branched_path_integrate_lighting(KernelGlobals *kg
 
                int num_samples;
 
-               if(CLOSURE_IS_BSDF_DIFFUSE(sc->type) || CLOSURE_IS_BSDF_BSSRDF(sc->type))
+               if(CLOSURE_IS_BSDF_DIFFUSE(sc->type))
                        num_samples = kernel_data.integrator.diffuse_samples;
+               else if(CLOSURE_IS_BSDF_BSSRDF(sc->type))
+                       num_samples = 1;
                else if(CLOSURE_IS_BSDF_GLOSSY(sc->type))
                        num_samples = kernel_data.integrator.glossy_samples;
                else
@@ -914,59 +755,104 @@ __device_noinline void kernel_branched_path_integrate_lighting(KernelGlobals *kg
                RNG bsdf_rng = cmj_hash(*rng, i);
 
                for(int j = 0; j < num_samples; j++) {
-                       /* sample BSDF */
-                       float bsdf_pdf;
-                       BsdfEval bsdf_eval;
-                       float3 bsdf_omega_in;
-                       differential3 bsdf_domega_in;
-                       float bsdf_u, bsdf_v;
-                       path_rng_2D(kg, &bsdf_rng, sample*num_samples + j, aa_samples*num_samples, rng_offset + PRNG_BSDF_U, &bsdf_u, &bsdf_v);
-                       int label;
-
-                       label = shader_bsdf_sample_closure(kg, sd, sc, bsdf_u, bsdf_v, &bsdf_eval,
-                               &bsdf_omega_in, &bsdf_domega_in, &bsdf_pdf);
+                       PathState ps = *state;
+                       float3 tp = throughput;
+                       Ray bsdf_ray;
 
-                       if(bsdf_pdf == 0.0f || bsdf_eval_is_zero(&bsdf_eval))
+                       if(!kernel_branched_path_surface_bounce(kg, &bsdf_rng, sd, sc, j, num_samples, &tp, &ps, L, &bsdf_ray))
                                continue;
 
-                       /* modify throughput */
-                       float3 tp = throughput;
-                       path_radiance_bsdf_bounce(L, &tp, &bsdf_eval, bsdf_pdf, state.bounce, label);
+                       kernel_path_indirect(kg, rng, bsdf_ray, tp*num_samples_inv, num_samples, ps, L);
+
+                       /* for render passes, sum and reset indirect light pass variables
+                        * for the next samples */
+                       path_radiance_sum_indirect(L);
+                       path_radiance_reset_indirect(L);
+               }
+       }
+}
+
+#ifdef __SUBSURFACE__
+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];
 
-                       /* set labels */
-                       float min_ray_pdf = fminf(bsdf_pdf, FLT_MAX);
+               if(!CLOSURE_IS_BSSRDF(sc->type))
+                       continue;
 
-                       /* modify path state */
-                       PathState ps = state;
-                       path_state_next(kg, &ps, label);
+               /* set up random number generator */
+               uint lcg_state = lcg_state_init(rng, 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);
 
-                       /* setup ray */
-                       Ray bsdf_ray;
+               state->flag |= PATH_RAY_BSSRDF_ANCESTOR;
 
-                       bsdf_ray.P = ray_offset(sd->P, (label & LABEL_TRANSMIT)? -sd->Ng: sd->Ng);
-                       bsdf_ray.D = bsdf_omega_in;
-                       bsdf_ray.t = FLT_MAX;
-#ifdef __RAY_DIFFERENTIALS__
-                       bsdf_ray.dP = sd->dP;
-                       bsdf_ray.dD = bsdf_domega_in;
+               /* do subsurface scatter step with copy of shader data, this will
+                * replace the BSSRDF with a diffuse BSDF closure */
+               for(int j = 0; j < num_samples; j++) {
+                       ShaderData bssrdf_sd[BSSRDF_MAX_HITS];
+                       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;
+                       bool need_update_volume_stack = kernel_data.integrator.use_volumes &&
+                                                       sd->flag & SD_OBJECT_INTERSECTS_VOLUME;
 #endif
-#ifdef __OBJECT_MOTION__
-                       bsdf_ray.time = sd->time;
+
+                       /* compute lighting with the BSDF closure */
+                       for(int hit = 0; hit < num_hits; hit++) {
+                               PathState hit_state = *state;
+
+                               path_state_branch(&hit_state, j, num_samples);
+
+#ifdef __VOLUME__
+                               if(need_update_volume_stack) {
+                                       /* 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
 
-                       kernel_path_indirect(kg, rng, sample*num_samples + j, bsdf_ray, buffer,
-                               tp*num_samples_inv, num_samples, aa_samples*num_samples,
-                               min_ray_pdf, bsdf_pdf, ps, rng_offset+PRNG_BOUNCE_NUM, L);
+#if defined(__EMISSION__) && defined(__BRANCHED_PATH__)
+                               /* direct light */
+                               if(kernel_data.integrator.use_direct_light) {
+                                       bool all = kernel_data.integrator.sample_all_lights_direct;
+                                       kernel_branched_path_surface_connect_light(kg, rng,
+                                               &bssrdf_sd[hit], &hit_state, throughput, num_samples_inv, L, all);
+                               }
+#endif
 
-                       /* for render passes, sum and reset indirect light pass variables
-                        * for the next samples */
-                       path_radiance_sum_indirect(L);
-                       path_radiance_reset_indirect(L);
+                               /* indirect light */
+                               kernel_branched_path_surface_indirect_light(kg, rng,
+                                       &bssrdf_sd[hit], throughput, num_samples_inv,
+                                       &hit_state, L);
+                       }
                }
+
+               state->flag &= ~PATH_RAY_BSSRDF_ANCESTOR;
        }
 }
+#endif
 
-__device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, int sample, Ray ray, __global float *buffer)
+ccl_device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, int sample, Ray ray, ccl_global float *buffer)
 {
        /* initialize */
        PathRadiance L;
@@ -975,18 +861,10 @@ __device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, int
 
        path_radiance_init(&L, kernel_data.film.use_light_pass);
 
-       float ray_pdf = 0.0f;
        PathState state;
-       int rng_offset = PRNG_BASE_NUM;
-#ifdef __CMJ__
-       int aa_samples = kernel_data.integrator.aa_samples;
-#else
-       int aa_samples = 0;
-#endif
-
-       path_state_init(&state);
+       path_state_init(kg, &state, rng, sample, &ray);
 
-       for(;; rng_offset += PRNG_BOUNCE_NUM) {
+       for(;;) {
                /* intersect scene */
                Intersection isect;
                uint visibility = path_state_ray_visibility(kg, &state);
@@ -1003,13 +881,134 @@ __device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, int
                        }
 
                        extmax = kernel_data.curve.maximum_width;
-                       lcg_state = lcg_init(*rng + rng_offset + sample*0x51633e2d);
+                       lcg_state = lcg_state_init(rng, &state, 0x51633e2d);
                }
 
-               if(!scene_intersect(kg, &ray, visibility, &isect, &lcg_state, difl, extmax)) {
+               bool hit = scene_intersect(kg, &ray, visibility, &isect, &lcg_state, difl, extmax);
+#else
+               bool hit = scene_intersect(kg, &ray, visibility, &isect, NULL, 0.0f, 0.0f);
+#endif
+
+#ifdef __VOLUME__
+               /* volume attenuation, emission, scatter */
+               if(state.volume_stack[0].shader != SHADER_NONE) {
+                       Ray volume_ray = ray;
+                       volume_ray.t = (hit)? isect.t: FLT_MAX;
+                       
+                       bool heterogeneous = volume_stack_is_heterogeneous(kg, state.volume_stack);
+
+#ifdef __VOLUME_DECOUPLED__
+                       /* decoupled ray marching only supported on CPU */
+
+                       /* cache steps along volume for repeated sampling */
+                       VolumeSegment volume_segment;
+                       ShaderData volume_sd;
+
+                       shader_setup_from_volume(kg, &volume_sd, &volume_ray, state.bounce, state.transparent_bounce);
+                       kernel_volume_decoupled_record(kg, &state,
+                               &volume_ray, &volume_sd, &volume_segment, heterogeneous);
+
+                       /* direct light sampling */
+                       if(volume_segment.closure_flag & SD_SCATTER) {
+                               volume_segment.sampling_method = volume_stack_sampling_method(kg, state.volume_stack);
+
+                               bool all = kernel_data.integrator.sample_all_lights_direct;
+
+                               kernel_branched_path_volume_connect_light(kg, rng, &volume_sd,
+                                       throughput, &state, &L, 1.0f, all, &volume_ray, &volume_segment);
+
+                               /* indirect light sampling */
+                               int num_samples = kernel_data.integrator.volume_samples;
+                               float num_samples_inv = 1.0f/num_samples;
+
+                               for(int j = 0; j < num_samples; j++) {
+                                       /* workaround to fix correlation bug in T38710, can find better solution
+                                        * in random number generator later, for now this is done here to not impact
+                                        * performance of rendering without volumes */
+                                       RNG tmp_rng = cmj_hash(*rng, state.rng_offset);
+
+                                       PathState ps = state;
+                                       Ray pray = ray;
+                                       float3 tp = throughput;
+
+                                       /* branch RNG state */
+                                       path_state_branch(&ps, j, num_samples);
+
+                                       /* scatter sample. if we use distance sampling and take just one
+                                        * sample for direct and indirect light, we could share this
+                                        * computation, but makes code a bit complex */
+                                       float rphase = path_state_rng_1D_for_decision(kg, &tmp_rng, &ps, PRNG_PHASE);
+                                       float rscatter = path_state_rng_1D_for_decision(kg, &tmp_rng, &ps, PRNG_SCATTER_DISTANCE);
+
+                                       VolumeIntegrateResult result = kernel_volume_decoupled_scatter(kg,
+                                               &ps, &pray, &volume_sd, &tp, rphase, rscatter, &volume_segment, NULL, false);
+                                               
+                                       (void)result;
+                                       kernel_assert(result == VOLUME_PATH_SCATTERED);
+
+                                       if(kernel_path_volume_bounce(kg, rng, &volume_sd, &tp, &ps, &L, &pray)) {
+                                               kernel_path_indirect(kg, rng, pray, tp*num_samples_inv, num_samples, ps, &L);
+
+                                               /* for render passes, sum and reset indirect light pass variables
+                                                * for the next samples */
+                                               path_radiance_sum_indirect(&L);
+                                               path_radiance_reset_indirect(&L);
+                                       }
+                               }
+                       }
+
+                       /* emission and transmittance */
+                       if(volume_segment.closure_flag & SD_EMISSION)
+                               path_radiance_accum_emission(&L, throughput, volume_segment.accum_emission, state.bounce);
+                       throughput *= volume_segment.accum_transmittance;
+
+                       /* free cached steps */
+                       kernel_volume_decoupled_free(kg, &volume_segment);
 #else
-               if(!scene_intersect(kg, &ray, visibility, &isect)) {
+                       /* GPU: no decoupled ray marching, scatter probalistically */
+                       int num_samples = kernel_data.integrator.volume_samples;
+                       float num_samples_inv = 1.0f/num_samples;
+
+                       /* todo: we should cache the shader evaluations from stepping
+                        * through the volume, for now we redo them multiple times */
+
+                       for(int j = 0; j < num_samples; j++) {
+                               PathState ps = state;
+                               Ray pray = ray;
+                               ShaderData volume_sd;
+                               float3 tp = throughput * num_samples_inv;
+
+                               /* branch RNG state */
+                               path_state_branch(&ps, j, num_samples);
+
+                               VolumeIntegrateResult result = kernel_volume_integrate(
+                                       kg, &ps, &volume_sd, &volume_ray, &L, &tp, rng, heterogeneous);
+                               
+#ifdef __VOLUME_SCATTER__
+                               if(result == VOLUME_PATH_SCATTERED) {
+                                       /* todo: support equiangular, MIS and all light sampling.
+                                        * alternatively get decoupled ray marching working on the GPU */
+                                       kernel_path_volume_connect_light(kg, rng, &volume_sd, tp, &state, &L);
+
+                                       if(kernel_path_volume_bounce(kg, rng, &volume_sd, &tp, &ps, &L, &pray)) {
+                                               kernel_path_indirect(kg, rng, pray, tp, num_samples, ps, &L);
+
+                                               /* for render passes, sum and reset indirect light pass variables
+                                                * for the next samples */
+                                               path_radiance_sum_indirect(&L);
+                                               path_radiance_reset_indirect(&L);
+                                       }
+                               }
+#endif
+                       }
+
+                       /* todo: avoid this calculation using decoupled ray marching */
+                       kernel_volume_shadow(kg, &state, &volume_ray, &throughput);
+#endif
+               }
 #endif
+
+               if(!hit) {
                        /* eval background shader if nothing hit */
                        if(kernel_data.background.transparent) {
                                L_transparent += average(throughput);
@@ -1022,7 +1021,7 @@ __device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, int
 
 #ifdef __BACKGROUND__
                        /* sample background shader */
-                       float3 L_background = indirect_background(kg, &ray, state.flag, ray_pdf, state.bounce);
+                       float3 L_background = indirect_background(kg, &state, &ray);
                        path_radiance_accum_background(&L, throughput, L_background, state.bounce);
 #endif
 
@@ -1031,13 +1030,13 @@ __device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, int
 
                /* setup shading */
                ShaderData sd;
-               shader_setup_from_ray(kg, &sd, &isect, &ray, state.bounce);
+               shader_setup_from_ray(kg, &sd, &isect, &ray, state.bounce, state.transparent_bounce);
                shader_eval_surface(kg, &sd, 0.0f, state.flag, SHADER_CONTEXT_MAIN);
-               shader_merge_closures(kg, &sd);
+               shader_merge_closures(&sd);
 
                /* holdout */
 #ifdef __HOLDOUT__
-               if((sd.flag & (SD_HOLDOUT|SD_HOLDOUT_MASK))) {
+               if(sd.flag & (SD_HOLDOUT|SD_HOLDOUT_MASK)) {
                        if(kernel_data.background.transparent) {
                                float3 holdout_weight;
                                
@@ -1056,12 +1055,12 @@ __device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, int
 #endif
 
                /* holdout mask objects do not write data passes */
-               kernel_write_data_passes(kg, buffer, &L, &sd, sample, state.flag, throughput);
+               kernel_write_data_passes(kg, buffer, &L, &sd, sample, &state, throughput);
 
 #ifdef __EMISSION__
                /* emission */
                if(sd.flag & SD_EMISSION) {
-                       float3 emission = indirect_primitive_emission(kg, &sd, isect.t, state.flag, ray_pdf);
+                       float3 emission = indirect_primitive_emission(kg, &sd, isect.t, state.flag, state.ray_pdf);
                        path_radiance_accum_emission(&L, throughput, emission, state.bounce);
                }
 #endif
@@ -1077,7 +1076,7 @@ __device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, int
                                break;
                        }
                        else if(probability != 1.0f) {
-                               float terminate = path_rng_1D(kg, rng, sample, aa_samples, rng_offset + PRNG_TERMINATE);
+                               float terminate = path_state_rng_1D_for_decision(kg, rng, &state, PRNG_TERMINATE);
 
                                if(terminate >= probability)
                                        break;
@@ -1086,72 +1085,62 @@ __device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, int
                        }
                }
 
+#ifdef __AO__
+               /* ambient occlusion */
+               if(kernel_data.integrator.use_ambient_occlusion || (sd.flag & SD_AO)) {
+                       kernel_branched_path_ao(kg, &sd, &L, &state, rng, throughput);
+               }
+#endif
+
 #ifdef __SUBSURFACE__
                /* bssrdf scatter to a different location on the same object */
                if(sd.flag & SD_BSSRDF) {
-                       for(int i = 0; i< sd.num_closure; i++) {
-                               ShaderClosure *sc = &sd.closure[i];
-
-                               if(!CLOSURE_IS_BSSRDF(sc->type))
-                                       continue;
+                       kernel_branched_path_subsurface_scatter(kg, &sd, &L, &state,
+                                                               rng, &ray, throughput);
+               }
+#endif
 
-                               /* set up random number generator */
-                               uint lcg_state = lcg_init(*rng + rng_offset + sample*0x68bc21eb);
-                               int num_samples = kernel_data.integrator.subsurface_samples;
-                               float num_samples_inv = 1.0f/num_samples;
-                               RNG bssrdf_rng = cmj_hash(*rng, i);
+               if(!(sd.flag & SD_HAS_ONLY_VOLUME)) {
+                       PathState hit_state = state;
 
-                               /* do subsurface scatter step with copy of shader data, this will
-                                * replace the BSSRDF with a diffuse BSDF closure */
-                               for(int j = 0; j < num_samples; j++) {
-                                       if(old_subsurface_scatter_use(&sd)) {
-                                               ShaderData bssrdf_sd = sd;
-                                               old_subsurface_scatter_step(kg, &bssrdf_sd, state.flag, sc, &lcg_state, true);
-
-                                               /* compute lighting with the BSDF closure */
-                                               kernel_branched_path_integrate_lighting(kg, rng, sample*num_samples + j,
-                                                       aa_samples*num_samples,
-                                                       &bssrdf_sd, throughput, num_samples_inv,
-                                                       ray_pdf, ray_pdf, state, rng_offset, &L, buffer);
-                                       }
-                                       else {
-                                               ShaderData bssrdf_sd[BSSRDF_MAX_HITS];
-                                               float bssrdf_u, bssrdf_v;
-                                               path_rng_2D(kg, &bssrdf_rng, sample*num_samples + j, aa_samples*num_samples, rng_offset + 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);
-
-                                               /* compute lighting with the BSDF closure */
-                                               for(int hit = 0; hit < num_hits; hit++)
-                                                       kernel_branched_path_integrate_lighting(kg, rng, sample*num_samples + j,
-                                                               aa_samples*num_samples,
-                                                               &bssrdf_sd[hit], throughput, num_samples_inv,
-                                                               ray_pdf, ray_pdf, state, rng_offset, &L, buffer);
-                                       }
-                               }
+#ifdef __EMISSION__
+                       /* direct light */
+                       if(kernel_data.integrator.use_direct_light) {
+                               bool all = kernel_data.integrator.sample_all_lights_direct;
+                               kernel_branched_path_surface_connect_light(kg, rng,
+                                       &sd, &hit_state, throughput, 1.0f, &L, all);
                        }
-               }
 #endif
 
-               /* lighting */
-               kernel_branched_path_integrate_lighting(kg, rng, sample, aa_samples,
-                       &sd, throughput, 1.0f, ray_pdf, ray_pdf, state, rng_offset, &L, buffer);
+                       /* indirect light */
+                       kernel_branched_path_surface_indirect_light(kg, rng,
+                               &sd, throughput, 1.0f, &hit_state, &L);
 
-               /* continue in case of transparency */
-               throughput *= shader_bsdf_transparency(kg, &sd);
+                       /* continue in case of transparency */
+                       throughput *= shader_bsdf_transparency(kg, &sd);
 
-               if(is_zero(throughput))
-                       break;
+                       if(is_zero(throughput))
+                               break;
+               }
 
                path_state_next(kg, &state, LABEL_TRANSPARENT);
                ray.P = ray_offset(sd.P, -sd.Ng);
                ray.t -= sd.ray_length; /* clipping works through transparent */
-       }
 
-       float3 L_sum = path_radiance_sum(kg, &L);
 
-#ifdef __CLAMP_SAMPLE__
-       path_radiance_clamp(&L, &L_sum, kernel_data.integrator.sample_clamp);
+#ifdef __RAY_DIFFERENTIALS__
+               ray.dP = sd.dP;
+               ray.dD.dx = -sd.dI.dx;
+               ray.dD.dy = -sd.dI.dy;
+#endif
+
+#ifdef __VOLUME__
+               /* enter/exit volume */
+               kernel_volume_stack_enter_exit(kg, &sd, state.volume_stack);
 #endif
+       }
+
+       float3 L_sum = path_radiance_clamp_and_sum(kg, &L);
 
        kernel_write_light_passes(kg, buffer, &L, sample);
 
@@ -1160,15 +1149,12 @@ __device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, int
 
 #endif
 
-__device_inline void kernel_path_trace_setup(KernelGlobals *kg, __global uint *rng_state, int sample, int x, int y, RNG *rng, Ray *ray)
+ccl_device_inline void kernel_path_trace_setup(KernelGlobals *kg, ccl_global uint *rng_state, int sample, int x, int y, RNG *rng, Ray *ray)
 {
        float filter_u;
        float filter_v;
-#ifdef __CMJ__
+
        int num_samples = kernel_data.integrator.aa_samples;
-#else
-       int num_samples = 0;
-#endif
 
        path_rng_init(kg, rng_state, sample, num_samples, rng, x, y, &filter_u, &filter_v);
 
@@ -1189,8 +1175,8 @@ __device_inline void kernel_path_trace_setup(KernelGlobals *kg, __global uint *r
        camera_sample(kg, x, y, filter_u, filter_v, lens_u, lens_v, time, ray);
 }
 
-__device void kernel_path_trace(KernelGlobals *kg,
-       __global float *buffer, __global uint *rng_state,
+ccl_device void kernel_path_trace(KernelGlobals *kg,
+       ccl_global float *buffer, ccl_global uint *rng_state,
        int sample, int x, int y, int offset, int stride)
 {
        /* buffer offset */
@@ -1209,7 +1195,7 @@ __device void kernel_path_trace(KernelGlobals *kg,
        /* integrate */
        float4 L;
 
-       if (ray.t != 0.0f)
+       if(ray.t != 0.0f)
                L = kernel_path_integrate(kg, &rng, sample, ray, buffer);
        else
                L = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
@@ -1221,8 +1207,8 @@ __device void kernel_path_trace(KernelGlobals *kg,
 }
 
 #ifdef __BRANCHED_PATH__
-__device void kernel_branched_path_trace(KernelGlobals *kg,
-       __global float *buffer, __global uint *rng_state,
+ccl_device void kernel_branched_path_trace(KernelGlobals *kg,
+       ccl_global float *buffer, ccl_global uint *rng_state,
        int sample, int x, int y, int offset, int stride)
 {
        /* buffer offset */
@@ -1241,7 +1227,7 @@ __device void kernel_branched_path_trace(KernelGlobals *kg,
        /* integrate */
        float4 L;
 
-       if (ray.t != 0.0f)
+       if(ray.t != 0.0f)
                L = kernel_branched_path_integrate(kg, &rng, sample, ray, buffer);
        else
                L = make_float4(0.0f, 0.0f, 0.0f, 0.0f);