Cycles: perform clamping per light contribution instead of whole path

With upcoming light group passes, for them to sum up correctly to the combined
pass the clamping must be more fine grained.

This also has the advantage that if one light is particularly noisy, it does
not diminish the contribution from other lights which do not need as much
clamping.

Clamp values on existing scenes will need to be tweaked to get similar results,
there is no automatic conversion possible which would give the same results as
before.

Implemented by Lukas, with tweaks by Brecht.

Part of D4837

1 files changed, 81 insertions, 65 deletions

diff --git a/intern/cycles/kernel/kernel_accumulate.h b/intern/cycles/kernel/kernel_accumulate.h
index 46a51f5a560..606c288649a 100644
--- a/intern/cycles/kernel/kernel_accumulate.h
+++ b/intern/cycles/kernel/kernel_accumulate.h
@@ -174,13 +174,13 @@ ccl_device_inline float3 bsdf_eval_sum(const BsdfEval *eval)
  * visible as the first non-transparent hit, while indirectly visible are the
  * bounces after that. */
 
-ccl_device_inline void path_radiance_init(PathRadiance *L, int use_light_pass)
+ccl_device_inline void path_radiance_init(KernelGlobals *kg, PathRadiance *L)
 {
   /* clear all */
 #ifdef __PASSES__
-  L->use_light_pass = use_light_pass;
+  L->use_light_pass = kernel_data.film.use_light_pass;
 
-  if (use_light_pass) {
+  if (kernel_data.film.use_light_pass) {
     L->indirect = make_float3(0.0f, 0.0f, 0.0f);
     L->direct_emission = make_float3(0.0f, 0.0f, 0.0f);
 
@@ -285,7 +285,37 @@ ccl_device_inline void path_radiance_bsdf_bounce(KernelGlobals *kg,
   }
 }
 
-ccl_device_inline void path_radiance_accum_emission(PathRadiance *L,
+#ifdef __CLAMP_SAMPLE__
+ccl_device_forceinline void path_radiance_clamp(KernelGlobals *kg, float3 *L, int bounce)
+{
+  float limit = (bounce > 0) ? kernel_data.integrator.sample_clamp_indirect :
+                               kernel_data.integrator.sample_clamp_direct;
+  float sum = reduce_add(fabs(*L));
+  if (sum > limit) {
+    *L *= limit / sum;
+  }
+}
+
+ccl_device_forceinline void path_radiance_clamp_throughput(KernelGlobals *kg,
+                                                           float3 *L,
+                                                           float3 *throughput,
+                                                           int bounce)
+{
+  float limit = (bounce > 0) ? kernel_data.integrator.sample_clamp_indirect :
+                               kernel_data.integrator.sample_clamp_direct;
+
+  float sum = reduce_add(fabs(*L));
+  if (sum > limit) {
+    float clamp_factor = limit / sum;
+    *L *= clamp_factor;
+    *throughput *= clamp_factor;
+  }
+}
+
+#endif
+
+ccl_device_inline void path_radiance_accum_emission(KernelGlobals *kg,
+                                                    PathRadiance *L,
                                                     ccl_addr_space PathState *state,
                                                     float3 throughput,
                                                     float3 value)
@@ -296,23 +326,29 @@ ccl_device_inline void path_radiance_accum_emission(PathRadiance *L,
   }
 #endif
 
+  float3 contribution = throughput * value;
+#ifdef __CLAMP_SAMPLE__
+  path_radiance_clamp(kg, &contribution, state->bounce - 1);
+#endif
+
 #ifdef __PASSES__
   if (L->use_light_pass) {
     if (state->bounce == 0)
-      L->emission += throughput * value;
+      L->emission += contribution;
     else if (state->bounce == 1)
-      L->direct_emission += throughput * value;
+      L->direct_emission += contribution;
     else
-      L->indirect += throughput * value;
+      L->indirect += contribution;
   }
   else
 #endif
   {
-    L->emission += throughput * value;
+    L->emission += contribution;
   }
 }
 
-ccl_device_inline void path_radiance_accum_ao(PathRadiance *L,
+ccl_device_inline void path_radiance_accum_ao(KernelGlobals *kg,
+                                              PathRadiance *L,
                                               ccl_addr_space PathState *state,
                                               float3 throughput,
                                               float3 alpha,
@@ -339,21 +375,23 @@ ccl_device_inline void path_radiance_accum_ao(PathRadiance *L,
   }
 #endif
 
+  float3 contribution = throughput * bsdf * ao;
+
 #ifdef __PASSES__
   if (L->use_light_pass) {
     if (state->bounce == 0) {
       /* Directly visible lighting. */
-      L->direct_diffuse += throughput * bsdf * ao;
+      L->direct_diffuse += contribution;
     }
     else {
       /* Indirectly visible lighting after BSDF bounce. */
-      L->indirect += throughput * bsdf * ao;
+      L->indirect += contribution;
     }
   }
   else
 #endif
   {
-    L->emission += throughput * bsdf * ao;
+    L->emission += contribution;
   }
 }
 
@@ -374,7 +412,8 @@ ccl_device_inline void path_radiance_accum_total_ao(PathRadiance *L,
 #endif
 }
 
-ccl_device_inline void path_radiance_accum_light(PathRadiance *L,
+ccl_device_inline void path_radiance_accum_light(KernelGlobals *kg,
+                                                 PathRadiance *L,
                                                  ccl_addr_space PathState *state,
                                                  float3 throughput,
                                                  BsdfEval *bsdf_eval,
@@ -394,15 +433,24 @@ ccl_device_inline void path_radiance_accum_light(PathRadiance *L,
   }
 #endif
 
+  float3 shaded_throughput = throughput * shadow;
+
 #ifdef __PASSES__
   if (L->use_light_pass) {
+    /* Compute the clamping based on the total contribution.
+     * The resulting scale is then be applied to all individual components. */
+    float3 full_contribution = shaded_throughput * bsdf_eval_sum(bsdf_eval);
+#  ifdef __CLAMP_SAMPLE__
+    path_radiance_clamp_throughput(kg, &full_contribution, &shaded_throughput, state->bounce);
+#  endif
+
     if (state->bounce == 0) {
       /* directly visible lighting */
-      L->direct_diffuse += throughput * bsdf_eval->diffuse * shadow;
-      L->direct_glossy += throughput * bsdf_eval->glossy * shadow;
-      L->direct_transmission += throughput * bsdf_eval->transmission * shadow;
-      L->direct_subsurface += throughput * bsdf_eval->subsurface * shadow;
-      L->direct_scatter += throughput * bsdf_eval->scatter * shadow;
+      L->direct_diffuse += shaded_throughput * bsdf_eval->diffuse;
+      L->direct_glossy += shaded_throughput * bsdf_eval->glossy;
+      L->direct_transmission += shaded_throughput * bsdf_eval->transmission;
+      L->direct_subsurface += shaded_throughput * bsdf_eval->subsurface;
+      L->direct_scatter += shaded_throughput * bsdf_eval->scatter;
 
       if (is_lamp) {
         L->shadow.x += shadow.x * shadow_fac;
@@ -412,13 +460,15 @@ ccl_device_inline void path_radiance_accum_light(PathRadiance *L,
     }
     else {
       /* indirectly visible lighting after BSDF bounce */
-      L->indirect += throughput * bsdf_eval_sum(bsdf_eval) * shadow;
+      L->indirect += full_contribution;
     }
   }
   else
 #endif
   {
-    L->emission += throughput * bsdf_eval->diffuse * shadow;
+    float3 contribution = shaded_throughput * bsdf_eval->diffuse;
+    path_radiance_clamp(kg, &contribution, state->bounce);
+    L->emission += contribution;
   }
 }
 
@@ -439,7 +489,8 @@ ccl_device_inline void path_radiance_accum_total_light(PathRadiance *L,
 #endif
 }
 
-ccl_device_inline void path_radiance_accum_background(PathRadiance *L,
+ccl_device_inline void path_radiance_accum_background(KernelGlobals *kg,
+                                                      PathRadiance *L,
                                                       ccl_addr_space PathState *state,
                                                       float3 throughput,
                                                       float3 value)
@@ -456,19 +507,24 @@ ccl_device_inline void path_radiance_accum_background(PathRadiance *L,
   }
 #endif
 
+  float3 contribution = throughput * value;
+#ifdef __CLAMP_SAMPLE__
+  path_radiance_clamp(kg, &contribution, state->bounce - 1);
+#endif
+
 #ifdef __PASSES__
   if (L->use_light_pass) {
     if (state->flag & PATH_RAY_TRANSPARENT_BACKGROUND)
-      L->background += throughput * value;
+      L->background += contribution;
     else if (state->bounce == 1)
-      L->direct_emission += throughput * value;
+      L->direct_emission += contribution;
     else
-      L->indirect += throughput * value;
+      L->indirect += contribution;
   }
   else
 #endif
   {
-    L->emission += throughput * value;
+    L->emission += contribution;
   }
 
 #ifdef __DENOISING_FEATURES__
@@ -587,8 +643,6 @@ ccl_device_inline float3 path_radiance_clamp_and_sum(KernelGlobals *kg,
   /* Light Passes are used */
 #ifdef __PASSES__
   float3 L_direct, L_indirect;
-  float clamp_direct = kernel_data.integrator.sample_clamp_direct;
-  float clamp_indirect = kernel_data.integrator.sample_clamp_indirect;
   if (L->use_light_pass) {
     path_radiance_sum_indirect(L);
 
@@ -622,44 +676,6 @@ ccl_device_inline float3 path_radiance_clamp_and_sum(KernelGlobals *kg,
 
       L->emission = make_float3(0.0f, 0.0f, 0.0f);
     }
-
-    /* Clamp direct and indirect samples */
-#  ifdef __CLAMP_SAMPLE__
-    else if (sum > clamp_direct || sum > clamp_indirect) {
-      float scale;
-
-      /* Direct */
-      float sum_direct = fabsf(L_direct.x) + fabsf(L_direct.y) + fabsf(L_direct.z);
-      if (sum_direct > clamp_direct) {
-        scale = clamp_direct / sum_direct;
-        L_direct *= scale;
-
-        L->direct_diffuse *= scale;
-        L->direct_glossy *= scale;
-        L->direct_transmission *= scale;
-        L->direct_subsurface *= scale;
-        L->direct_scatter *= scale;
-        L->emission *= scale;
-        L->background *= scale;
-      }
-
-      /* Indirect */
-      float sum_indirect = fabsf(L_indirect.x) + fabsf(L_indirect.y) + fabsf(L_indirect.z);
-      if (sum_indirect > clamp_indirect) {
-        scale = clamp_indirect / sum_indirect;
-        L_indirect *= scale;
-
-        L->indirect_diffuse *= scale;
-        L->indirect_glossy *= scale;
-        L->indirect_transmission *= scale;
-        L->indirect_subsurface *= scale;
-        L->indirect_scatter *= scale;
-      }
-
-      /* Sum again, after clamping */
-      L_sum = L_direct + L_indirect;
-    }
-#  endif
   }
 
   /* No Light Passes */