Cycles: Stop lamp sampling if the lamp isn't visible

Both spot and area light have large areas where they're not visible.
Therefore, this patch stops the light sampling code when one of these cases (outside of the spotlight cone or behind the area light) occurs, before the lamp shader is evaluated.
In the case of the area light, the solid angle sampling can also be skipped.

In a test scene with Sample All Lights and 18 Area lamps and 9 Spot lamps that all point away from the area that the camera sees, render time drops from 12sec to 5sec.

Reviewers: brecht, sergey, dingto, juicyfruit

Differential Revision: https://developer.blender.org/D2216

4 files changed, 110 insertions, 114 deletions

diff --git a/intern/cycles/kernel/kernel_light.h b/intern/cycles/kernel/kernel_light.h
index 1e99f2cb99f..fffa9afb342 100644
--- a/intern/cycles/kernel/kernel_light.h
+++ b/intern/cycles/kernel/kernel_light.h
@@ -510,7 +510,7 @@ ccl_device float lamp_light_pdf(KernelGlobals *kg, const float3 Ng, const float3
 	return t*t/cos_pi;
 }
 
-ccl_device_inline void lamp_light_sample(KernelGlobals *kg,
+ccl_device_inline bool lamp_light_sample(KernelGlobals *kg,
                                          int lamp,
                                          float randu, float randv,
                                          float3 P,
@@ -581,6 +581,9 @@ ccl_device_inline void lamp_light_sample(KernelGlobals *kg,
 				/* spot light attenuation */
 				float4 data2 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 2);
 				ls->eval_fac *= spot_light_attenuation(data1, data2, ls);
+				if(ls->eval_fac == 0.0f) {
+					return false;
+				}
 			}
 			ls->pdf *= lamp_light_pdf(kg, ls->Ng, -ls->D, ls->t);
 		}
@@ -593,6 +596,10 @@ ccl_device_inline void lamp_light_sample(KernelGlobals *kg,
 			float3 axisv = make_float3(data2.y, data2.z, data2.w);
 			float3 D = make_float3(data3.y, data3.z, data3.w);
 
+			if(dot(ls->P - P, D) > 0.0f) {
+				return false;
+			}
+
 			ls->pdf = area_light_sample(P, &ls->P,
 			                          axisu, axisv,
 			                          randu, randv,
@@ -603,13 +610,12 @@ ccl_device_inline void lamp_light_sample(KernelGlobals *kg,
 
 			float invarea = data2.x;
 			ls->eval_fac = 0.25f*invarea;
-
-			if(dot(ls->D, D) > 0.0f)
-				ls->pdf = 0.0f;
 		}
 
 		ls->eval_fac *= kernel_data.integrator.inv_pdf_lights;
 	}
+
+	return (ls->pdf > 0.0f);
 }
 
 ccl_device bool lamp_light_eval(KernelGlobals *kg, int lamp, float3 P, float3 D, float t, LightSample *ls)
@@ -836,7 +842,7 @@ ccl_device bool light_select_reached_max_bounces(KernelGlobals *kg, int index, i
 	return (bounce > __float_as_int(data4.x));
 }
 
-ccl_device_noinline void light_sample(KernelGlobals *kg,
+ccl_device_noinline bool light_sample(KernelGlobals *kg,
                                       float randt,
                                       float randu,
                                       float randv,
@@ -857,21 +863,20 @@ ccl_device_noinline void light_sample(KernelGlobals *kg,
 		int shader_flag = __float_as_int(l.z);
 
 		triangle_light_sample(kg, prim, object, randu, randv, time, ls);
-
 		/* compute incoming direction, distance and pdf */
 		ls->D = normalize_len(ls->P - P, &ls->t);
 		ls->pdf = triangle_light_pdf(kg, ls->Ng, -ls->D, ls->t);
 		ls->shader |= shader_flag;
+		return (ls->pdf > 0.0f);
 	}
 	else {
 		int lamp = -prim-1;
 
 		if(UNLIKELY(light_select_reached_max_bounces(kg, lamp, bounce))) {
-			ls->pdf = 0.0f;
-			return;
+			return false;
 		}
 
-		lamp_light_sample(kg, lamp, randu, randv, P, ls);
+		return lamp_light_sample(kg, lamp, randu, randv, P, ls);
 	}
 }
 
diff --git a/intern/cycles/kernel/kernel_path_surface.h b/intern/cycles/kernel/kernel_path_surface.h
index 250b8e92a45..27acf8166b3 100644
--- a/intern/cycles/kernel/kernel_path_surface.h
+++ b/intern/cycles/kernel/kernel_path_surface.h
@@ -54,15 +54,15 @@ ccl_device_noinline void kernel_branched_path_surface_connect_light(KernelGlobal
 				path_branched_rng_2D(kg, &lamp_rng, state, j, num_samples, PRNG_LIGHT_U, &light_u, &light_v);
 
 				LightSample ls;
-				lamp_light_sample(kg, i, light_u, light_v, ccl_fetch(sd, P), &ls);
-
-				if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
-					/* trace shadow ray */
-					float3 shadow;
-
-					if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
-						/* accumulate */
-						path_radiance_accum_light(L, throughput*num_samples_inv, &L_light, shadow, num_samples_inv, state->bounce, is_lamp);
+				if(lamp_light_sample(kg, i, light_u, light_v, ccl_fetch(sd, P), &ls)) {
+					if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
+						/* trace shadow ray */
+						float3 shadow;
+
+						if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
+							/* accumulate */
+							path_radiance_accum_light(L, throughput*num_samples_inv, &L_light, shadow, num_samples_inv, state->bounce, is_lamp);
+						}
 					}
 				}
 			}
@@ -86,15 +86,15 @@ ccl_device_noinline void kernel_branched_path_surface_connect_light(KernelGlobal
 					light_t = 0.5f*light_t;
 
 				LightSample ls;
-				light_sample(kg, light_t, light_u, light_v, ccl_fetch(sd, time), ccl_fetch(sd, P), state->bounce, &ls);
-
-				if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
-					/* trace shadow ray */
-					float3 shadow;
-
-					if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
-						/* accumulate */
-						path_radiance_accum_light(L, throughput*num_samples_inv, &L_light, shadow, num_samples_inv, state->bounce, is_lamp);
+				if(light_sample(kg, light_t, light_u, light_v, ccl_fetch(sd, time), ccl_fetch(sd, P), state->bounce, &ls)) {
+					if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
+						/* trace shadow ray */
+						float3 shadow;
+
+						if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
+							/* accumulate */
+							path_radiance_accum_light(L, throughput*num_samples_inv, &L_light, shadow, num_samples_inv, state->bounce, is_lamp);
+						}
 					}
 				}
 			}
@@ -107,16 +107,16 @@ ccl_device_noinline void kernel_branched_path_surface_connect_light(KernelGlobal
 		path_state_rng_2D(kg, rng, state, PRNG_LIGHT_U, &light_u, &light_v);
 
 		LightSample ls;
-		light_sample(kg, light_t, light_u, light_v, ccl_fetch(sd, time), ccl_fetch(sd, P), state->bounce, &ls);
-
-		/* sample random light */
-		if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
-			/* trace shadow ray */
-			float3 shadow;
-
-			if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
-				/* accumulate */
-				path_radiance_accum_light(L, throughput*num_samples_adjust, &L_light, shadow, num_samples_adjust, state->bounce, is_lamp);
+		if(light_sample(kg, light_t, light_u, light_v, ccl_fetch(sd, time), ccl_fetch(sd, P), state->bounce, &ls)) {
+			/* sample random light */
+			if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
+				/* trace shadow ray */
+				float3 shadow;
+
+				if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
+					/* accumulate */
+					path_radiance_accum_light(L, throughput*num_samples_adjust, &L_light, shadow, num_samples_adjust, state->bounce, is_lamp);
+				}
 			}
 		}
 	}
@@ -206,15 +206,15 @@ ccl_device_inline void kernel_path_surface_connect_light(KernelGlobals *kg, ccl_
 #endif
 
 	LightSample ls;
-	light_sample(kg, light_t, light_u, light_v, ccl_fetch(sd, time), ccl_fetch(sd, P), state->bounce, &ls);
-
-	if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
-		/* trace shadow ray */
-		float3 shadow;
+	if(light_sample(kg, light_t, light_u, light_v, ccl_fetch(sd, time), ccl_fetch(sd, P), state->bounce, &ls)) {
+		if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
+			/* trace shadow ray */
+			float3 shadow;
 
-		if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
-			/* accumulate */
-			path_radiance_accum_light(L, throughput, &L_light, shadow, 1.0f, state->bounce, is_lamp);
+			if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
+				/* accumulate */
+				path_radiance_accum_light(L, throughput, &L_light, shadow, 1.0f, state->bounce, is_lamp);
+			}
 		}
 	}
 #endif
diff --git a/intern/cycles/kernel/kernel_path_volume.h b/intern/cycles/kernel/kernel_path_volume.h
index 5fd4f2fad4c..0bb8f1c517f 100644
--- a/intern/cycles/kernel/kernel_path_volume.h
+++ b/intern/cycles/kernel/kernel_path_volume.h
@@ -46,17 +46,16 @@ ccl_device_inline void kernel_path_volume_connect_light(
 	light_ray.time = sd->time;
 #  endif
 
-	light_sample(kg, light_t, light_u, light_v, sd->time, sd->P, state->bounce, &ls);
-	if(ls.pdf == 0.0f)
-		return;
-	
-	if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
-		/* trace shadow ray */
-		float3 shadow;
+	if(light_sample(kg, light_t, light_u, light_v, sd->time, sd->P, state->bounce, &ls))
+	{
+		if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
+			/* trace shadow ray */
+			float3 shadow;
 
-		if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
-			/* accumulate */
-			path_radiance_accum_light(L, throughput, &L_light, shadow, 1.0f, state->bounce, is_lamp);
+			if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
+				/* accumulate */
+				path_radiance_accum_light(L, throughput, &L_light, shadow, 1.0f, state->bounce, is_lamp);
+			}
 		}
 	}
 #endif
@@ -146,7 +145,7 @@ ccl_device void kernel_branched_path_volume_connect_light(KernelGlobals *kg, RNG
 				path_branched_rng_2D(kg, &lamp_rng, state, j, num_samples, PRNG_LIGHT_U, &light_u, &light_v);
 
 				LightSample ls;
-				lamp_light_sample(kg, i, light_u, light_v, ray->P, &ls);		
+				lamp_light_sample(kg, i, light_u, light_v, ray->P, &ls);
 
 				float3 tp = throughput;
 
@@ -156,23 +155,20 @@ ccl_device void kernel_branched_path_volume_connect_light(KernelGlobals *kg, RNG
 
 				VolumeIntegrateResult result = kernel_volume_decoupled_scatter(kg,
 					state, ray, sd, &tp, rphase, rscatter, segment, (ls.t != FLT_MAX)? &ls.P: NULL, false);
-					
+
 				(void)result;
 				kernel_assert(result == VOLUME_PATH_SCATTERED);
 
 				/* todo: split up light_sample so we don't have to call it again with new position */
-				lamp_light_sample(kg, i, light_u, light_v, sd->P, &ls);
-
-				if(ls.pdf == 0.0f)
-					continue;
-
-				if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
-					/* trace shadow ray */
-					float3 shadow;
-
-					if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
-						/* accumulate */
-						path_radiance_accum_light(L, tp*num_samples_inv, &L_light, shadow, num_samples_inv, state->bounce, is_lamp);
+				if(lamp_light_sample(kg, i, light_u, light_v, sd->P, &ls)) {
+					if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
+						/* trace shadow ray */
+						float3 shadow;
+
+						if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
+							/* accumulate */
+							path_radiance_accum_light(L, tp*num_samples_inv, &L_light, shadow, num_samples_inv, state->bounce, is_lamp);
+						}
 					}
 				}
 			}
@@ -212,18 +208,15 @@ ccl_device void kernel_branched_path_volume_connect_light(KernelGlobals *kg, RNG
 				kernel_assert(result == VOLUME_PATH_SCATTERED);
 
 				/* todo: split up light_sample so we don't have to call it again with new position */
-				light_sample(kg, light_t, light_u, light_v, sd->time, sd->P, state->bounce, &ls);
-
-				if(ls.pdf == 0.0f)
-					continue;
-
-				if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
-					/* trace shadow ray */
-					float3 shadow;
-
-					if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
-						/* accumulate */
-						path_radiance_accum_light(L, tp*num_samples_inv, &L_light, shadow, num_samples_inv, state->bounce, is_lamp);
+				if(light_sample(kg, light_t, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
+					if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
+						/* trace shadow ray */
+						float3 shadow;
+
+						if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
+							/* accumulate */
+							path_radiance_accum_light(L, tp*num_samples_inv, &L_light, shadow, num_samples_inv, state->bounce, is_lamp);
+						}
 					}
 				}
 			}
@@ -251,19 +244,16 @@ ccl_device void kernel_branched_path_volume_connect_light(KernelGlobals *kg, RNG
 		kernel_assert(result == VOLUME_PATH_SCATTERED);
 
 		/* todo: split up light_sample so we don't have to call it again with new position */
-		light_sample(kg, light_t, light_u, light_v, sd->time, sd->P, state->bounce, &ls);
-
-		if(ls.pdf == 0.0f)
-			return;
-
-		/* sample random light */
-		if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
-			/* trace shadow ray */
-			float3 shadow;
-
-			if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
-				/* accumulate */
-				path_radiance_accum_light(L, tp, &L_light, shadow, 1.0f, state->bounce, is_lamp);
+		if(light_sample(kg, light_t, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
+			/* sample random light */
+			if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp)) {
+				/* trace shadow ray */
+				float3 shadow;
+
+				if(!shadow_blocked(kg, emission_sd, state, &light_ray, &shadow)) {
+					/* accumulate */
+					path_radiance_accum_light(L, tp, &L_light, shadow, 1.0f, state->bounce, is_lamp);
+				}
 			}
 		}
 	}
diff --git a/intern/cycles/kernel/split/kernel_direct_lighting.h b/intern/cycles/kernel/split/kernel_direct_lighting.h
index ebe91097496..6ad736fc2c1 100644
--- a/intern/cycles/kernel/split/kernel_direct_lighting.h
+++ b/intern/cycles/kernel/split/kernel_direct_lighting.h
@@ -74,30 +74,31 @@ ccl_device char kernel_direct_lighting(
 			path_state_rng_2D(kg, rng, state, PRNG_LIGHT_U, &light_u, &light_v);
 
 			LightSample ls;
-			light_sample(kg,
-			             light_t, light_u, light_v,
-			             ccl_fetch(sd, time),
-			             ccl_fetch(sd, P),
-			             state->bounce,
-			             &ls);
+			if(light_sample(kg,
+			                light_t, light_u, light_v,
+			                ccl_fetch(sd, time),
+			                ccl_fetch(sd, P),
+			                state->bounce,
+			                &ls)) {
 
-			Ray light_ray;
+				Ray light_ray;
 #ifdef __OBJECT_MOTION__
-			light_ray.time = ccl_fetch(sd, time);
+				light_ray.time = ccl_fetch(sd, time);
 #endif
 
-			BsdfEval L_light;
-			bool 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.
-				 */
-				LightRay_coop[ray_index] = light_ray;
-				BSDFEval_coop[ray_index] = L_light;
-				ISLamp_coop[ray_index] = is_lamp;
-				/* Mark ray state for next shadow kernel. */
-				ADD_RAY_FLAG(ray_state, ray_index, RAY_SHADOW_RAY_CAST_DL);
-				enqueue_flag = 1;
+				BsdfEval L_light;
+				bool 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.
+					 */
+					LightRay_coop[ray_index] = light_ray;
+					BSDFEval_coop[ray_index] = L_light;
+					ISLamp_coop[ray_index] = is_lamp;
+					/* Mark ray state for next shadow kernel. */
+					ADD_RAY_FLAG(ray_state, ray_index, RAY_SHADOW_RAY_CAST_DL);
+					enqueue_flag = 1;
+				}
 			}
 		}
 #endif  /* __EMISSION__ */