Cycles: Implement denoising option for reducing noise in the rendered image

This commit contains the first part of the new Cycles denoising option,
which filters the resulting image using information gathered during rendering
to get rid of noise while preserving visual features as well as possible.

To use the option, enable it in the render layer options. The default settings
fit a wide range of scenes, but the user can tweak individual settings to
control the tradeoff between a noise-free image, image details, and calculation
time.

Note that the denoiser may still change in the future and that some features
are not implemented yet. The most important missing feature is animation
denoising, which uses information from multiple frames at once to produce a
flicker-free and smoother result. These features will be added in the future.

Finally, thanks to all the people who supported this project:

- Google (through the GSoC) and Theory Studios for sponsoring the development
- The authors of the papers I used for implementing the denoiser (more details
  on them will be included in the technical docs)
- The other Cycles devs for feedback on the code, especially Sergey for
  mentoring the GSoC project and Brecht for the code review!
- And of course the users who helped with testing, reported bugs and things
  that could and/or should work better!

1 files changed, 67 insertions, 50 deletions

diff --git a/intern/cycles/kernel/kernel_path_branched.h b/intern/cycles/kernel/kernel_path_branched.h
index ddcb57161ea..10816d3e5d1 100644
--- a/intern/cycles/kernel/kernel_path_branched.h
+++ b/intern/cycles/kernel/kernel_path_branched.h
@@ -56,10 +56,10 @@ ccl_device_inline void kernel_branched_path_ao(KernelGlobals *kg,
 			light_ray.dD = differential3_zero();
 
 			if(!shadow_blocked(kg, emission_sd, state, &light_ray, &ao_shadow)) {
-				path_radiance_accum_ao(L, throughput*num_samples_inv, ao_alpha, ao_bsdf, ao_shadow, state->bounce);
+				path_radiance_accum_ao(L, state, throughput*num_samples_inv, ao_alpha, ao_bsdf, ao_shadow);
 			}
 			else {
-				path_radiance_accum_total_ao(L, throughput*num_samples_inv, ao_bsdf);
+				path_radiance_accum_total_ao(L, state, throughput*num_samples_inv, ao_bsdf);
 			}
 		}
 	}
@@ -72,14 +72,32 @@ ccl_device_noinline void kernel_branched_path_surface_indirect_light(KernelGloba
 	RNG *rng, ShaderData *sd, ShaderData *indirect_sd, ShaderData *emission_sd,
 	float3 throughput, float num_samples_adjust, PathState *state, PathRadiance *L)
 {
+	float sum_sample_weight = 0.0f;
+#ifdef __DENOISING_FEATURES__
+	if(state->denoising_feature_weight > 0.0f) {
+		for(int i = 0; i < sd->num_closure; i++) {
+			const ShaderClosure *sc = &sd->closure[i];
+
+			/* transparency is not handled here, but in outer loop */
+			if(!CLOSURE_IS_BSDF(sc->type) || CLOSURE_IS_BSDF_TRANSPARENT(sc->type)) {
+				continue;
+			}
+
+			sum_sample_weight += sc->sample_weight;
+		}
+	}
+	else {
+		sum_sample_weight = 1.0f;
+	}
+#endif  /* __DENOISING_FEATURES__ */
+
 	for(int i = 0; i < sd->num_closure; i++) {
 		const ShaderClosure *sc = &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)
+		if(!CLOSURE_IS_BSDF(sc->type) || CLOSURE_IS_BSDF_TRANSPARENT(sc->type)) {
 			continue;
+		}
 
 		int num_samples;
 
@@ -111,7 +129,8 @@ ccl_device_noinline void kernel_branched_path_surface_indirect_light(KernelGloba
 			                                        &tp,
 			                                        &ps,
 			                                        L,
-			                                        &bsdf_ray))
+			                                        &bsdf_ray,
+			                                        sum_sample_weight))
 			{
 				continue;
 			}
@@ -243,14 +262,19 @@ ccl_device void kernel_branched_path_subsurface_scatter(KernelGlobals *kg,
 }
 #endif  /* __SUBSURFACE__ */
 
-ccl_device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, int sample, Ray ray, ccl_global float *buffer)
+ccl_device float kernel_branched_path_integrate(KernelGlobals *kg,
+                                                RNG *rng,
+                                                int sample,
+                                                Ray ray,
+                                                ccl_global float *buffer,
+                                                PathRadiance *L,
+                                                bool *is_shadow_catcher)
 {
 	/* 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);
+	path_radiance_init(L, kernel_data.film.use_light_pass);
 
 	/* shader data memory used for both volumes and surfaces, saves stack space */
 	ShaderData sd;
@@ -330,7 +354,7 @@ ccl_device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, in
 				int all = kernel_data.integrator.sample_all_lights_direct;
 
 				kernel_branched_path_volume_connect_light(kg, rng, &sd,
-					&emission_sd, throughput, &state, &L, all,
+					&emission_sd, throughput, &state, L, all,
 					&volume_ray, &volume_segment);
 
 				/* indirect light sampling */
@@ -362,7 +386,7 @@ ccl_device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, in
 					                             &sd,
 					                             &tp,
 					                             &ps,
-					                             &L,
+					                             L,
 					                             &pray))
 					{
 						kernel_path_indirect(kg,
@@ -373,19 +397,19 @@ ccl_device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, in
 						                     tp*num_samples_inv,
 						                     num_samples,
 						                     &ps,
-						                     &L);
+						                     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);
+						path_radiance_sum_indirect(L);
+						path_radiance_reset_indirect(L);
 					}
 				}
 			}
 
 			/* emission and transmittance */
 			if(volume_segment.closure_flag & SD_EMISSION)
-				path_radiance_accum_emission(&L, throughput, volume_segment.accum_emission, state.bounce);
+				path_radiance_accum_emission(L, throughput, volume_segment.accum_emission, state.bounce);
 			throughput *= volume_segment.accum_transmittance;
 
 			/* free cached steps */
@@ -407,20 +431,20 @@ ccl_device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, in
 				path_state_branch(&ps, j, num_samples);
 
 				VolumeIntegrateResult result = kernel_volume_integrate(
-					kg, &ps, &sd, &volume_ray, &L, &tp, rng, heterogeneous);
+					kg, &ps, &sd, &volume_ray, L, &tp, rng, heterogeneous);
 
 #ifdef __VOLUME_SCATTER__
 				if(result == VOLUME_PATH_SCATTERED) {
 					/* todo: support equiangular, MIS and all light sampling.
 					 * alternatively get decoupled ray marching working on the GPU */
-					kernel_path_volume_connect_light(kg, rng, &sd, &emission_sd, tp, &state, &L);
+					kernel_path_volume_connect_light(kg, rng, &sd, &emission_sd, tp, &state, L);
 
 					if(kernel_path_volume_bounce(kg,
 					                             rng,
 					                             &sd,
 					                             &tp,
 					                             &ps,
-					                             &L,
+					                             L,
 					                             &pray))
 					{
 						kernel_path_indirect(kg,
@@ -431,12 +455,12 @@ ccl_device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, in
 						                     tp,
 						                     num_samples,
 						                     &ps,
-						                     &L);
+						                     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);
+						path_radiance_sum_indirect(L);
+						path_radiance_reset_indirect(L);
 					}
 				}
 #endif  /* __VOLUME_SCATTER__ */
@@ -462,7 +486,7 @@ ccl_device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, in
 #ifdef __BACKGROUND__
 			/* sample background shader */
 			float3 L_background = indirect_background(kg, &emission_sd, &state, &ray);
-			path_radiance_accum_background(&L, &state, throughput, L_background);
+			path_radiance_accum_background(L, &state, throughput, L_background);
 #endif  /* __BACKGROUND__ */
 
 			break;
@@ -476,10 +500,10 @@ ccl_device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, in
 #ifdef __SHADOW_TRICKS__
 		if((sd.object_flag & SD_OBJECT_SHADOW_CATCHER)) {
 			if(state.flag & PATH_RAY_CAMERA) {
-				state.flag |= (PATH_RAY_SHADOW_CATCHER | PATH_RAY_SHADOW_CATCHER_ONLY);
+				state.flag |= (PATH_RAY_SHADOW_CATCHER | PATH_RAY_SHADOW_CATCHER_ONLY | PATH_RAY_STORE_SHADOW_INFO);
 				state.catcher_object = sd.object;
 				if(!kernel_data.background.transparent) {
-					L.shadow_color = indirect_background(kg, &emission_sd, &state, &ray);
+					L->shadow_color = indirect_background(kg, &emission_sd, &state, &ray);
 				}
 			}
 		}
@@ -509,13 +533,13 @@ ccl_device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, in
 #endif  /* __HOLDOUT__ */
 
 		/* holdout mask objects do not write data passes */
-		kernel_write_data_passes(kg, buffer, &L, &sd, sample, &state, throughput);
+		kernel_write_data_passes(kg, buffer, L, &sd, sample, &state, throughput);
 
 #ifdef __EMISSION__
 		/* emission */
 		if(sd.flag & SD_EMISSION) {
 			float3 emission = indirect_primitive_emission(kg, &sd, isect.t, state.flag, state.ray_pdf);
-			path_radiance_accum_emission(&L, throughput, emission, state.bounce);
+			path_radiance_accum_emission(L, throughput, emission, state.bounce);
 		}
 #endif  /* __EMISSION__ */
 
@@ -539,10 +563,12 @@ ccl_device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, in
 			}
 		}
 
+		kernel_update_denoising_features(kg, &sd, &state, L);
+
 #ifdef __AO__
 		/* ambient occlusion */
 		if(kernel_data.integrator.use_ambient_occlusion || (sd.flag & SD_AO)) {
-			kernel_branched_path_ao(kg, &sd, &emission_sd, &L, &state, rng, throughput);
+			kernel_branched_path_ao(kg, &sd, &emission_sd, L, &state, rng, throughput);
 		}
 #endif  /* __AO__ */
 
@@ -550,7 +576,7 @@ ccl_device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, in
 		/* bssrdf scatter to a different location on the same object */
 		if(sd.flag & SD_BSSRDF) {
 			kernel_branched_path_subsurface_scatter(kg, &sd, &indirect_sd, &emission_sd,
-			                                        &L, &state, rng, &ray, throughput);
+			                                        L, &state, rng, &ray, throughput);
 		}
 #endif  /* __SUBSURFACE__ */
 
@@ -563,13 +589,13 @@ ccl_device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, in
 				int all = (kernel_data.integrator.sample_all_lights_direct) ||
 				          (state.flag & PATH_RAY_SHADOW_CATCHER);
 				kernel_branched_path_surface_connect_light(kg, rng,
-					&sd, &emission_sd, &hit_state, throughput, 1.0f, &L, all);
+					&sd, &emission_sd, &hit_state, throughput, 1.0f, L, all);
 			}
 #endif  /* __EMISSION__ */
 
 			/* indirect light */
 			kernel_branched_path_surface_indirect_light(kg, rng,
-				&sd, &indirect_sd, &emission_sd, throughput, 1.0f, &hit_state, &L);
+				&sd, &indirect_sd, &emission_sd, throughput, 1.0f, &hit_state, L);
 
 			/* continue in case of transparency */
 			throughput *= shader_bsdf_transparency(kg, &sd);
@@ -598,24 +624,15 @@ ccl_device float4 kernel_branched_path_integrate(KernelGlobals *kg, RNG *rng, in
 #endif  /* __VOLUME__ */
 	}
 
-	float3 L_sum;
 #ifdef __SHADOW_TRICKS__
-	if(state.flag & PATH_RAY_SHADOW_CATCHER) {
-		L_sum = path_radiance_sum_shadowcatcher(kg, &L, &L_transparent);
-	}
-	else
+	*is_shadow_catcher = (state.flag & PATH_RAY_SHADOW_CATCHER);
 #endif  /* __SHADOW_TRICKS__ */
-	{
-		L_sum = path_radiance_clamp_and_sum(kg, &L);
-	}
-
-	kernel_write_light_passes(kg, buffer, &L, sample);
 
 #ifdef __KERNEL_DEBUG__
 	kernel_write_debug_passes(kg, buffer, &state, &debug_data, sample);
 #endif  /* __KERNEL_DEBUG__ */
 
-	return make_float4(L_sum.x, L_sum.y, L_sum.z, 1.0f - L_transparent);
+	return 1.0f - L_transparent;
 }
 
 ccl_device void kernel_branched_path_trace(KernelGlobals *kg,
@@ -636,15 +653,16 @@ ccl_device void kernel_branched_path_trace(KernelGlobals *kg,
 	kernel_path_trace_setup(kg, rng_state, sample, x, y, &rng, &ray);
 
 	/* integrate */
-	float4 L;
-
-	if(ray.t != 0.0f)
-		L = kernel_branched_path_integrate(kg, &rng, sample, ray, buffer);
-	else
-		L = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
+	PathRadiance L;
+	bool is_shadow_catcher;
 
-	/* accumulate result in output buffer */
-	kernel_write_pass_float4(buffer, sample, L);
+	if(ray.t != 0.0f) {
+		float alpha = kernel_branched_path_integrate(kg, &rng, sample, ray, buffer, &L, &is_shadow_catcher);
+		kernel_write_result(kg, buffer, sample, &L, alpha, is_shadow_catcher);
+	}
+	else {
+		kernel_write_result(kg, buffer, sample, NULL, 0.0f, false);
+	}
 
 	path_rng_end(kg, rng_state, rng);
 }
@@ -654,4 +672,3 @@ ccl_device void kernel_branched_path_trace(KernelGlobals *kg,
 #endif  /* __BRANCHED_PATH__ */
 
 CCL_NAMESPACE_END
-