From 099aaea447e13e186765dafe179f4b7aff948da6 Mon Sep 17 00:00:00 2001 From: Sergey Sharybin Date: Mon, 15 Jun 2015 23:02:42 +0200 Subject: [PATCH] Cycles: Move branched path tracking into own file Code there started becoming a bit too big, by splitting it up it'll make it easier to do improvements or extending the features in there. The layout is not totally final yet, would need to try de-duplicating parts of code from split kernel with non-split integrators, --- intern/cycles/kernel/CMakeLists.txt | 1 + intern/cycles/kernel/kernel_path.h | 510 ----------------- intern/cycles/kernel/kernel_path_branched.h | 529 ++++++++++++++++++ intern/cycles/kernel/kernels/cpu/kernel.cpp | 1 + .../cycles/kernel/kernels/cpu/kernel_avx.cpp | 1 + .../cycles/kernel/kernels/cpu/kernel_avx2.cpp | 1 + .../cycles/kernel/kernels/cpu/kernel_sse2.cpp | 1 + .../cycles/kernel/kernels/cpu/kernel_sse3.cpp | 1 + .../kernel/kernels/cpu/kernel_sse41.cpp | 1 + intern/cycles/kernel/kernels/cuda/kernel.cu | 1 + intern/cycles/kernel/kernels/opencl/kernel.cl | 1 + 11 files changed, 538 insertions(+), 510 deletions(-) create mode 100644 intern/cycles/kernel/kernel_path_branched.h diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt index fa8f36bad9a..18d1360542c 100644 --- a/intern/cycles/kernel/CMakeLists.txt +++ b/intern/cycles/kernel/CMakeLists.txt @@ -47,6 +47,7 @@ set(SRC_HEADERS kernel_montecarlo.h kernel_passes.h kernel_path.h + kernel_path_branched.h kernel_path_common.h kernel_path_state.h kernel_path_surface.h diff --git a/intern/cycles/kernel/kernel_path.h b/intern/cycles/kernel/kernel_path.h index 1804c8bfbe3..9794ad1d180 100644 --- a/intern/cycles/kernel/kernel_path.h +++ b/intern/cycles/kernel/kernel_path.h @@ -324,43 +324,6 @@ ccl_device void kernel_path_ao(KernelGlobals *kg, ShaderData *sd, PathRadiance * } } -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); - - float3 ao_D; - float ao_pdf; - - sample_cos_hemisphere(ao_N, bsdf_u, bsdf_v, &ao_D, &ao_pdf); - - if(dot(ccl_fetch(sd, Ng), ao_D) > 0.0f && ao_pdf != 0.0f) { - Ray light_ray; - float3 ao_shadow; - - light_ray.P = ray_offset(ccl_fetch(sd, P), ccl_fetch(sd, Ng)); - light_ray.D = ao_D; - light_ray.t = kernel_data.background.ao_distance; -#ifdef __OBJECT_MOTION__ - light_ray.time = ccl_fetch(sd, time); -#endif - light_ray.dP = ccl_fetch(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); - } - } -} - #ifdef __SUBSURFACE__ ccl_device bool kernel_path_subsurface_scatter(KernelGlobals *kg, ShaderData *sd, PathRadiance *L, PathState *state, RNG *rng, Ray *ray, float3 *throughput) @@ -708,446 +671,6 @@ ccl_device float4 kernel_path_integrate(KernelGlobals *kg, RNG *rng, int sample, return make_float4(L_sum.x, L_sum.y, L_sum.z, 1.0f - L_transparent); } -#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< ccl_fetch(sd, num_closure); i++) { - const ShaderClosure *sc = &ccl_fetch(sd, closure)[i]; - - if(!CLOSURE_IS_BSDF(sc->type)) - continue; - /* transparency is not handled here, but in outer loop */ - if(sc->type == CLOSURE_BSDF_TRANSPARENT_ID) - continue; - - int num_samples; - - 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 - num_samples = kernel_data.integrator.transmission_samples; - - num_samples = ceil_to_int(num_samples_adjust*num_samples); - - float num_samples_inv = num_samples_adjust/num_samples; - RNG bsdf_rng = cmj_hash(*rng, i); - - for(int j = 0; j < num_samples; j++) { - PathState ps = *state; - float3 tp = throughput; - Ray bsdf_ray; - - if(!kernel_branched_path_surface_bounce(kg, &bsdf_rng, sd, sc, j, num_samples, &tp, &ps, L, &bsdf_ray)) - continue; - - 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< ccl_fetch(sd, num_closure); i++) { - ShaderClosure *sc = &ccl_fetch(sd, closure)[i]; - - if(!CLOSURE_IS_BSSRDF(sc->type)) - continue; - - /* 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); - - /* 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 && - ccl_fetch(sd, flag) & SD_OBJECT_INTERSECTS_VOLUME; -#endif - - /* 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_volume_stack_update_for_subsurface( - kg, - &volume_ray, - hit_state.volume_stack); - - /* Move volume ray forward. */ - volume_ray.P = P; - } -#endif - -#if defined(__EMISSION__) && defined(__BRANCHED_PATH__) - /* direct light */ - if(kernel_data.integrator.use_direct_light) { - 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 - - /* indirect light */ - kernel_branched_path_surface_indirect_light(kg, rng, - &bssrdf_sd[hit], throughput, num_samples_inv, - &hit_state, L); - } - } - } -} -#endif - -ccl_device float4 kernel_branched_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); - -#ifdef __KERNEL_DEBUG__ - DebugData debug_data; - debug_data_init(&debug_data); -#endif - - for(;;) { - /* 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_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, NULL, 0.0f, 0.0f); -#endif - -#ifdef __KERNEL_DEBUG__ - if(state.flag & PATH_RAY_CAMERA) { - debug_data.num_bvh_traversal_steps += isect.num_traversal_steps; - debug_data.num_bvh_traversed_instances += isect.num_traversed_instances; - } - debug_data.num_ray_bounces++; -#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, 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 - /* 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); - -#ifdef __PASSES__ - if(!(kernel_data.film.pass_flag & PASS_BACKGROUND)) -#endif - break; - } - -#ifdef __BACKGROUND__ - /* sample background shader */ - float3 L_background = indirect_background(kg, &state, &ray); - path_radiance_accum_background(&L, throughput, L_background, state.bounce); -#endif - - break; - } - - /* setup shading */ - ShaderData sd; - 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(&sd); - - /* holdout */ -#ifdef __HOLDOUT__ - if(sd.flag & (SD_HOLDOUT|SD_HOLDOUT_MASK)) { - 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 - - /* holdout mask objects do not write data passes */ - 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, state.ray_pdf); - path_radiance_accum_emission(&L, throughput, emission, state.bounce); - } -#endif - - /* transparency termination */ - if(state.flag & PATH_RAY_TRANSPARENT) { - /* 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); - - if(probability == 0.0f) { - break; - } - else if(probability != 1.0f) { - float terminate = path_state_rng_1D_for_decision(kg, rng, &state, PRNG_TERMINATE); - - if(terminate >= probability) - break; - - throughput /= probability; - } - } - -#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) { - kernel_branched_path_subsurface_scatter(kg, &sd, &L, &state, - rng, &ray, throughput); - } -#endif - - if(!(sd.flag & SD_HAS_ONLY_VOLUME)) { - PathState hit_state = state; - -#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 - - /* 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); - - 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 */ - - -#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); - -#ifdef __KERNEL_DEBUG__ - kernel_write_debug_passes(kg, buffer, &state, &debug_data, sample); -#endif - - return make_float4(L_sum.x, L_sum.y, L_sum.z, 1.0f - L_transparent); -} - -#endif - 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) @@ -1179,38 +702,5 @@ ccl_device void kernel_path_trace(KernelGlobals *kg, path_rng_end(kg, rng_state, rng); } -#ifdef __BRANCHED_PATH__ -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 */ - int index = offset + x + y*stride; - int pass_stride = kernel_data.film.pass_stride; - - rng_state += index; - buffer += index*pass_stride; - - /* initialize random numbers and ray */ - RNG rng; - Ray ray; - - kernel_path_trace_setup(kg, rng_state, sample, x, y, &rng, &ray); - - /* integrate */ - float4 L; - - 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); - - /* accumulate result in output buffer */ - kernel_write_pass_float4(buffer, sample, L); - - path_rng_end(kg, rng_state, rng); -} -#endif - CCL_NAMESPACE_END diff --git a/intern/cycles/kernel/kernel_path_branched.h b/intern/cycles/kernel/kernel_path_branched.h new file mode 100644 index 00000000000..6190cd57f17 --- /dev/null +++ b/intern/cycles/kernel/kernel_path_branched.h @@ -0,0 +1,529 @@ +/* + * Copyright 2011-2013 Blender Foundation + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +CCL_NAMESPACE_BEGIN + +#ifdef __BRANCHED_PATH__ + +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); + + float3 ao_D; + float ao_pdf; + + sample_cos_hemisphere(ao_N, bsdf_u, bsdf_v, &ao_D, &ao_pdf); + + if(dot(ccl_fetch(sd, Ng), ao_D) > 0.0f && ao_pdf != 0.0f) { + Ray light_ray; + float3 ao_shadow; + + light_ray.P = ray_offset(ccl_fetch(sd, P), ccl_fetch(sd, Ng)); + light_ray.D = ao_D; + light_ray.t = kernel_data.background.ao_distance; +#ifdef __OBJECT_MOTION__ + light_ray.time = ccl_fetch(sd, time); +#endif + light_ray.dP = ccl_fetch(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); + } + } +} + + +/* 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< ccl_fetch(sd, num_closure); i++) { + const ShaderClosure *sc = &ccl_fetch(sd, closure)[i]; + + if(!CLOSURE_IS_BSDF(sc->type)) + continue; + /* transparency is not handled here, but in outer loop */ + if(sc->type == CLOSURE_BSDF_TRANSPARENT_ID) + continue; + + int num_samples; + + 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 + num_samples = kernel_data.integrator.transmission_samples; + + num_samples = ceil_to_int(num_samples_adjust*num_samples); + + float num_samples_inv = num_samples_adjust/num_samples; + RNG bsdf_rng = cmj_hash(*rng, i); + + for(int j = 0; j < num_samples; j++) { + PathState ps = *state; + float3 tp = throughput; + Ray bsdf_ray; + + if(!kernel_branched_path_surface_bounce(kg, &bsdf_rng, sd, sc, j, num_samples, &tp, &ps, L, &bsdf_ray)) + continue; + + 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< ccl_fetch(sd, num_closure); i++) { + ShaderClosure *sc = &ccl_fetch(sd, closure)[i]; + + if(!CLOSURE_IS_BSSRDF(sc->type)) + continue; + + /* 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); + + /* 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 && + ccl_fetch(sd, flag) & SD_OBJECT_INTERSECTS_VOLUME; +#endif + + /* 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_volume_stack_update_for_subsurface( + kg, + &volume_ray, + hit_state.volume_stack); + + /* Move volume ray forward. */ + volume_ray.P = P; + } +#endif + +#if defined(__EMISSION__) && defined(__BRANCHED_PATH__) + /* direct light */ + if(kernel_data.integrator.use_direct_light) { + 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 + + /* indirect light */ + kernel_branched_path_surface_indirect_light(kg, rng, + &bssrdf_sd[hit], throughput, num_samples_inv, + &hit_state, L); + } + } + } +} +#endif + +ccl_device float4 kernel_branched_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); + +#ifdef __KERNEL_DEBUG__ + DebugData debug_data; + debug_data_init(&debug_data); +#endif + + for(;;) { + /* 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_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, NULL, 0.0f, 0.0f); +#endif + +#ifdef __KERNEL_DEBUG__ + if(state.flag & PATH_RAY_CAMERA) { + debug_data.num_bvh_traversal_steps += isect.num_traversal_steps; + debug_data.num_bvh_traversed_instances += isect.num_traversed_instances; + } + debug_data.num_ray_bounces++; +#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, 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 + /* 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); + +#ifdef __PASSES__ + if(!(kernel_data.film.pass_flag & PASS_BACKGROUND)) +#endif + break; + } + +#ifdef __BACKGROUND__ + /* sample background shader */ + float3 L_background = indirect_background(kg, &state, &ray); + path_radiance_accum_background(&L, throughput, L_background, state.bounce); +#endif + + break; + } + + /* setup shading */ + ShaderData sd; + 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(&sd); + + /* holdout */ +#ifdef __HOLDOUT__ + if(sd.flag & (SD_HOLDOUT|SD_HOLDOUT_MASK)) { + 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 + + /* holdout mask objects do not write data passes */ + 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, state.ray_pdf); + path_radiance_accum_emission(&L, throughput, emission, state.bounce); + } +#endif + + /* transparency termination */ + if(state.flag & PATH_RAY_TRANSPARENT) { + /* 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); + + if(probability == 0.0f) { + break; + } + else if(probability != 1.0f) { + float terminate = path_state_rng_1D_for_decision(kg, rng, &state, PRNG_TERMINATE); + + if(terminate >= probability) + break; + + throughput /= probability; + } + } + +#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) { + kernel_branched_path_subsurface_scatter(kg, &sd, &L, &state, + rng, &ray, throughput); + } +#endif + + if(!(sd.flag & SD_HAS_ONLY_VOLUME)) { + PathState hit_state = state; + +#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 + + /* 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); + + 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 */ + + +#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); + +#ifdef __KERNEL_DEBUG__ + kernel_write_debug_passes(kg, buffer, &state, &debug_data, sample); +#endif + + return make_float4(L_sum.x, L_sum.y, L_sum.z, 1.0f - L_transparent); +} + +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 */ + int index = offset + x + y*stride; + int pass_stride = kernel_data.film.pass_stride; + + rng_state += index; + buffer += index*pass_stride; + + /* initialize random numbers and ray */ + RNG rng; + Ray ray; + + kernel_path_trace_setup(kg, rng_state, sample, x, y, &rng, &ray); + + /* integrate */ + float4 L; + + 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); + + /* accumulate result in output buffer */ + kernel_write_pass_float4(buffer, sample, L); + + path_rng_end(kg, rng_state, rng); +} + +#endif /* __BRANCHED_PATH__ */ + +CCL_NAMESPACE_END + diff --git a/intern/cycles/kernel/kernels/cpu/kernel.cpp b/intern/cycles/kernel/kernels/cpu/kernel.cpp index a7eaa758f5d..37a73ab2f04 100644 --- a/intern/cycles/kernel/kernels/cpu/kernel.cpp +++ b/intern/cycles/kernel/kernels/cpu/kernel.cpp @@ -23,6 +23,7 @@ #include "kernel_globals.h" #include "kernel_film.h" #include "kernel_path.h" +#include "kernel_path_branched.h" #include "kernel_bake.h" CCL_NAMESPACE_BEGIN diff --git a/intern/cycles/kernel/kernels/cpu/kernel_avx.cpp b/intern/cycles/kernel/kernels/cpu/kernel_avx.cpp index f1027ad413d..df77bedc729 100644 --- a/intern/cycles/kernel/kernels/cpu/kernel_avx.cpp +++ b/intern/cycles/kernel/kernels/cpu/kernel_avx.cpp @@ -38,6 +38,7 @@ #include "kernel_globals.h" #include "kernel_film.h" #include "kernel_path.h" +#include "kernel_path_branched.h" #include "kernel_bake.h" CCL_NAMESPACE_BEGIN diff --git a/intern/cycles/kernel/kernels/cpu/kernel_avx2.cpp b/intern/cycles/kernel/kernels/cpu/kernel_avx2.cpp index b2f16ff54d8..b3192369794 100644 --- a/intern/cycles/kernel/kernels/cpu/kernel_avx2.cpp +++ b/intern/cycles/kernel/kernels/cpu/kernel_avx2.cpp @@ -39,6 +39,7 @@ #include "kernel_globals.h" #include "kernel_film.h" #include "kernel_path.h" +#include "kernel_path_branched.h" #include "kernel_bake.h" CCL_NAMESPACE_BEGIN diff --git a/intern/cycles/kernel/kernels/cpu/kernel_sse2.cpp b/intern/cycles/kernel/kernels/cpu/kernel_sse2.cpp index cc8c603e8f8..f9c5134e442 100644 --- a/intern/cycles/kernel/kernels/cpu/kernel_sse2.cpp +++ b/intern/cycles/kernel/kernels/cpu/kernel_sse2.cpp @@ -34,6 +34,7 @@ #include "kernel_globals.h" #include "kernel_film.h" #include "kernel_path.h" +#include "kernel_path_branched.h" #include "kernel_bake.h" CCL_NAMESPACE_BEGIN diff --git a/intern/cycles/kernel/kernels/cpu/kernel_sse3.cpp b/intern/cycles/kernel/kernels/cpu/kernel_sse3.cpp index 20919a4f26e..2dbe4b81821 100644 --- a/intern/cycles/kernel/kernels/cpu/kernel_sse3.cpp +++ b/intern/cycles/kernel/kernels/cpu/kernel_sse3.cpp @@ -36,6 +36,7 @@ #include "kernel_globals.h" #include "kernel_film.h" #include "kernel_path.h" +#include "kernel_path_branched.h" #include "kernel_bake.h" CCL_NAMESPACE_BEGIN diff --git a/intern/cycles/kernel/kernels/cpu/kernel_sse41.cpp b/intern/cycles/kernel/kernels/cpu/kernel_sse41.cpp index 48579d3b7e5..5c57ad01181 100644 --- a/intern/cycles/kernel/kernels/cpu/kernel_sse41.cpp +++ b/intern/cycles/kernel/kernels/cpu/kernel_sse41.cpp @@ -37,6 +37,7 @@ #include "kernel_globals.h" #include "kernel_film.h" #include "kernel_path.h" +#include "kernel_path_branched.h" #include "kernel_bake.h" CCL_NAMESPACE_BEGIN diff --git a/intern/cycles/kernel/kernels/cuda/kernel.cu b/intern/cycles/kernel/kernels/cuda/kernel.cu index 29bf67d9750..bcd55b8c676 100644 --- a/intern/cycles/kernel/kernels/cuda/kernel.cu +++ b/intern/cycles/kernel/kernels/cuda/kernel.cu @@ -22,6 +22,7 @@ #include "../../kernel_globals.h" #include "../../kernel_film.h" #include "../../kernel_path.h" +#include "../../kernel_path_branched.h" #include "../../kernel_bake.h" /* device data taken from CUDA occupancy calculator */ diff --git a/intern/cycles/kernel/kernels/opencl/kernel.cl b/intern/cycles/kernel/kernels/opencl/kernel.cl index bffcd53bab3..15fb34cfe3b 100644 --- a/intern/cycles/kernel/kernels/opencl/kernel.cl +++ b/intern/cycles/kernel/kernels/opencl/kernel.cl @@ -23,6 +23,7 @@ #include "../../kernel_film.h" #include "../../kernel_path.h" +#include "../../kernel_path_branched.h" #include "../../kernel_bake.h" #ifdef __COMPILE_ONLY_MEGAKERNEL__