[Bf-blender-cvs] [099aaea] master: Cycles: Move branched path tracking into own file
Sergey Sharybin
noreply at git.blender.org
Mon Jun 15 23:05:37 CEST 2015
Commit: 099aaea447e13e186765dafe179f4b7aff948da6
Author: Sergey Sharybin
Date: Mon Jun 15 23:02:42 2015 +0200
Branches: master
https://developer.blender.org/rB099aaea447e13e186765dafe179f4b7aff948da6
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,
===================================================================
M intern/cycles/kernel/CMakeLists.txt
M intern/cycles/kernel/kernel_path.h
A intern/cycles/kernel/kernel_path_branched.h
M intern/cycles/kernel/kernels/cpu/kernel.cpp
M intern/cycles/kernel/kernels/cpu/kernel_avx.cpp
M intern/cycles/kernel/kernels/cpu/kernel_avx2.cpp
M intern/cycles/kernel/kernels/cpu/kernel_sse2.cpp
M intern/cycles/kernel/kernels/cpu/kernel_sse3.cpp
M intern/cycles/kernel/kernels/cpu/kernel_sse41.cpp
M intern/cycles/kernel/kernels/cuda/kernel.cu
M intern/cycles/kernel/kernels/opencl/kernel.cl
===================================================================
diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt
index fa8f36b..18d1360 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 1804c8b..9794ad1 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_dec
@@ Diff output truncated at 10240 characters. @@
More information about the Bf-blender-cvs
mailing list