1 files changed, 31 insertions, 13 deletions

diff --git a/intern/cycles/kernel/integrator/shade_surface.h b/intern/cycles/kernel/integrator/shade_surface.h
index 9f6077e5d66..10d3cbf7f57 100644
--- a/intern/cycles/kernel/integrator/shade_surface.h
+++ b/intern/cycles/kernel/integrator/shade_surface.h
@@ -182,23 +182,35 @@ ccl_device_forceinline void integrate_surface_direct_light(KernelGlobals kg,
 
   /* Write shadow ray and associated state to global memory. */
   integrator_state_write_shadow_ray(kg, shadow_state, &ray);
+  // Save memory by storing the light and object indices in the shadow_isect
+  INTEGRATOR_STATE_ARRAY_WRITE(shadow_state, shadow_isect, 0, object) = ray.self.object;
+  INTEGRATOR_STATE_ARRAY_WRITE(shadow_state, shadow_isect, 0, prim) = ray.self.prim;
+  INTEGRATOR_STATE_ARRAY_WRITE(shadow_state, shadow_isect, 1, object) = ray.self.light_object;
+  INTEGRATOR_STATE_ARRAY_WRITE(shadow_state, shadow_isect, 1, prim) = ray.self.light_prim;
 
   /* Copy state from main path to shadow path. */
   const uint16_t bounce = INTEGRATOR_STATE(state, path, bounce);
   const uint16_t transparent_bounce = INTEGRATOR_STATE(state, path, transparent_bounce);
   uint32_t shadow_flag = INTEGRATOR_STATE(state, path, flag);
   shadow_flag |= (is_light) ? PATH_RAY_SHADOW_FOR_LIGHT : 0;
-  shadow_flag |= (shadow_flag & PATH_RAY_ANY_PASS) ? 0 : PATH_RAY_SURFACE_PASS;
   const float3 throughput = INTEGRATOR_STATE(state, path, throughput) * bsdf_eval_sum(&bsdf_eval);
 
   if (kernel_data.kernel_features & KERNEL_FEATURE_LIGHT_PASSES) {
-    const packed_float3 pass_diffuse_weight =
-        (bounce == 0) ? packed_float3(bsdf_eval_pass_diffuse_weight(&bsdf_eval)) :
-                        INTEGRATOR_STATE(state, path, pass_diffuse_weight);
-    const packed_float3 pass_glossy_weight = (bounce == 0) ?
-                                                 packed_float3(
-                                                     bsdf_eval_pass_glossy_weight(&bsdf_eval)) :
-                                                 INTEGRATOR_STATE(state, path, pass_glossy_weight);
+    packed_float3 pass_diffuse_weight;
+    packed_float3 pass_glossy_weight;
+
+    if (shadow_flag & PATH_RAY_ANY_PASS) {
+      /* Indirect bounce, use weights from earlier surface or volume bounce. */
+      pass_diffuse_weight = INTEGRATOR_STATE(state, path, pass_diffuse_weight);
+      pass_glossy_weight = INTEGRATOR_STATE(state, path, pass_glossy_weight);
+    }
+    else {
+      /* Direct light, use BSDFs at this bounce. */
+      shadow_flag |= PATH_RAY_SURFACE_PASS;
+      pass_diffuse_weight = packed_float3(bsdf_eval_pass_diffuse_weight(&bsdf_eval));
+      pass_glossy_weight = packed_float3(bsdf_eval_pass_glossy_weight(&bsdf_eval));
+    }
+
     INTEGRATOR_STATE_WRITE(shadow_state, shadow_path, pass_diffuse_weight) = pass_diffuse_weight;
     INTEGRATOR_STATE_WRITE(shadow_state, shadow_path, pass_glossy_weight) = pass_glossy_weight;
   }
@@ -266,13 +278,11 @@ ccl_device_forceinline int integrate_surface_bsdf_bssrdf_bounce(
   }
 
   /* Setup ray. Note that clipping works through transparent bounces. */
-  INTEGRATOR_STATE_WRITE(state, ray, P) = ray_offset(sd->P,
-                                                     (label & LABEL_TRANSMIT) ? -sd->Ng : sd->Ng);
+  INTEGRATOR_STATE_WRITE(state, ray, P) = sd->P;
   INTEGRATOR_STATE_WRITE(state, ray, D) = normalize(bsdf_omega_in);
   INTEGRATOR_STATE_WRITE(state, ray, t) = (label & LABEL_TRANSPARENT) ?
                                               INTEGRATOR_STATE(state, ray, t) - sd->ray_length :
                                               FLT_MAX;
-
 #ifdef __RAY_DIFFERENTIALS__
   INTEGRATOR_STATE_WRITE(state, ray, dP) = differential_make_compact(sd->dP);
   INTEGRATOR_STATE_WRITE(state, ray, dD) = differential_make_compact(bsdf_domega_in);
@@ -316,7 +326,7 @@ ccl_device_forceinline bool integrate_surface_volume_only_bounce(IntegratorState
   }
 
   /* Setup ray position, direction stays unchanged. */
-  INTEGRATOR_STATE_WRITE(state, ray, P) = ray_offset(sd->P, -sd->Ng);
+  INTEGRATOR_STATE_WRITE(state, ray, P) = sd->P;
 
   /* Clipping works through transparent. */
   INTEGRATOR_STATE_WRITE(state, ray, t) -= sd->ray_length;
@@ -360,10 +370,14 @@ ccl_device_forceinline void integrate_surface_ao(KernelGlobals kg,
   }
 
   Ray ray ccl_optional_struct_init;
-  ray.P = ray_offset(sd->P, sd->Ng);
+  ray.P = sd->P;
   ray.D = ao_D;
   ray.t = kernel_data.integrator.ao_bounces_distance;
   ray.time = sd->time;
+  ray.self.object = sd->object;
+  ray.self.prim = sd->prim;
+  ray.self.light_object = OBJECT_NONE;
+  ray.self.light_prim = PRIM_NONE;
   ray.dP = differential_zero_compact();
   ray.dD = differential_zero_compact();
 
@@ -375,6 +389,10 @@ ccl_device_forceinline void integrate_surface_ao(KernelGlobals kg,
 
   /* Write shadow ray and associated state to global memory. */
   integrator_state_write_shadow_ray(kg, shadow_state, &ray);
+  INTEGRATOR_STATE_ARRAY_WRITE(shadow_state, shadow_isect, 0, object) = ray.self.object;
+  INTEGRATOR_STATE_ARRAY_WRITE(shadow_state, shadow_isect, 0, prim) = ray.self.prim;
+  INTEGRATOR_STATE_ARRAY_WRITE(shadow_state, shadow_isect, 1, object) = ray.self.light_object;
+  INTEGRATOR_STATE_ARRAY_WRITE(shadow_state, shadow_isect, 1, prim) = ray.self.light_prim;
 
   /* Copy state from main path to shadow path. */
   const uint16_t bounce = INTEGRATOR_STATE(state, path, bounce);