diff options
| author | Brecht Van Lommel <brechtvanlommel@gmail.com> | 2016-05-22 23:35:47 +0300 |
|---|---|---|
| committer | Brecht Van Lommel <brechtvanlommel@gmail.com> | 2016-05-23 23:29:24 +0300 |
| commit | 999d5a67852b5958b9361c9888734ebc889e4a22 (patch) | |
| tree | 5f3c5ad0409c77fc6ae3486420b3888fa1e2fea8 /intern/cycles/kernel/split | |
| parent | af4a04eae07184f7437a8c51858a4ddb8a2e3e4c (diff) | |
Cycles CUDA: reduce stack memory by reusing ShaderData.
57% less for path and 48% less for branched path.
Diffstat (limited to 'intern/cycles/kernel/split')
5 files changed, 6 insertions, 4 deletions
diff --git a/intern/cycles/kernel/split/kernel_background_buffer_update.h b/intern/cycles/kernel/split/kernel_background_buffer_update.h index 3d12a3dd993..f42d0a985bb 100644 --- a/intern/cycles/kernel/split/kernel_background_buffer_update.h +++ b/intern/cycles/kernel/split/kernel_background_buffer_update.h @@ -157,7 +157,7 @@ ccl_device char kernel_background_buffer_update( if(IS_STATE(ray_state, ray_index, RAY_HIT_BACKGROUND)) { #ifdef __BACKGROUND__ /* sample background shader */ - float3 L_background = indirect_background(kg, state, ray); + float3 L_background = indirect_background(kg, kg->sd_input, state, ray); path_radiance_accum_background(L, (*throughput), L_background, state->bounce); #endif ASSIGN_RAY_STATE(ray_state, ray_index, RAY_UPDATE_BUFFER); @@ -226,7 +226,7 @@ ccl_device char kernel_background_buffer_update( *throughput = make_float3(1.0f, 1.0f, 1.0f); *L_transparent = 0.0f; path_radiance_init(L, kernel_data.film.use_light_pass); - path_state_init(kg, state, rng, sample, ray); + path_state_init(kg, kg->sd_input, state, rng, sample, ray); #ifdef __KERNEL_DEBUG__ debug_data_init(debug_data); #endif diff --git a/intern/cycles/kernel/split/kernel_data_init.h b/intern/cycles/kernel/split/kernel_data_init.h index 9891391a3a3..e3dbc43757e 100644 --- a/intern/cycles/kernel/split/kernel_data_init.h +++ b/intern/cycles/kernel/split/kernel_data_init.h @@ -207,6 +207,7 @@ ccl_device void kernel_data_init( L_transparent_coop[ray_index] = 0.0f; path_radiance_init(&PathRadiance_coop[ray_index], kernel_data.film.use_light_pass); path_state_init(kg, + kg->sd_input, &PathState_coop[ray_index], &rng_coop[ray_index], my_sample, diff --git a/intern/cycles/kernel/split/kernel_direct_lighting.h b/intern/cycles/kernel/split/kernel_direct_lighting.h index c7a2aa6426c..ebe91097496 100644 --- a/intern/cycles/kernel/split/kernel_direct_lighting.h +++ b/intern/cycles/kernel/split/kernel_direct_lighting.h @@ -88,7 +88,7 @@ ccl_device char kernel_direct_lighting( BsdfEval L_light; bool is_lamp; - if(direct_emission(kg, sd, &ls, state, &light_ray, &L_light, &is_lamp)) { + if(direct_emission(kg, sd, kg->sd_input, &ls, state, &light_ray, &L_light, &is_lamp)) { /* Write intermediate data to global memory to access from * the next kernel. */ diff --git a/intern/cycles/kernel/split/kernel_lamp_emission.h b/intern/cycles/kernel/split/kernel_lamp_emission.h index dc3b4b34d4e..3bd0e361078 100644 --- a/intern/cycles/kernel/split/kernel_lamp_emission.h +++ b/intern/cycles/kernel/split/kernel_lamp_emission.h @@ -74,7 +74,7 @@ ccl_device void kernel_lamp_emission( /* intersect with lamp */ float3 emission; - if(indirect_lamp_emission(kg, state, &light_ray, &emission)) { + if(indirect_lamp_emission(kg, kg->sd_input, state, &light_ray, &emission)) { path_radiance_accum_emission(L, throughput, emission, state->bounce); } } diff --git a/intern/cycles/kernel/split/kernel_shadow_blocked.h b/intern/cycles/kernel/split/kernel_shadow_blocked.h index 0c989861eef..6153af47f96 100644 --- a/intern/cycles/kernel/split/kernel_shadow_blocked.h +++ b/intern/cycles/kernel/split/kernel_shadow_blocked.h @@ -71,6 +71,7 @@ ccl_device void kernel_shadow_blocked( float3 shadow; update_path_radiance = !(shadow_blocked(kg, + kg->sd_input, state, light_ray_global, &shadow)); |