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, 43 insertions, 46 deletions

diff --git a/intern/cycles/kernel/kernel_path.h b/intern/cycles/kernel/kernel_path.h
index 58da141aed3..0d31ae32aa6 100644
--- a/intern/cycles/kernel/kernel_path.h
+++ b/intern/cycles/kernel/kernel_path.h
@@ -90,10 +90,10 @@ ccl_device_noinline void kernel_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, ao_alpha, ao_bsdf, ao_shadow, state->bounce);
+			path_radiance_accum_ao(L, state, throughput, ao_alpha, ao_bsdf, ao_shadow);
 		}
 		else {
-			path_radiance_accum_total_ao(L, throughput, ao_bsdf);
+			path_radiance_accum_total_ao(L, state, throughput, ao_bsdf);
 		}
 	}
 }
@@ -366,6 +366,8 @@ ccl_device void kernel_path_indirect(KernelGlobals *kg,
 			throughput /= probability;
 		}
 
+		kernel_update_denoising_features(kg, sd, state, L);
+
 #ifdef __AO__
 		/* ambient occlusion */
 		if(kernel_data.integrator.use_ambient_occlusion || (sd->flag & SD_AO)) {
@@ -427,18 +429,19 @@ ccl_device void kernel_path_indirect(KernelGlobals *kg,
 }
 
 
-ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
-                                               RNG *rng,
-                                               int sample,
-                                               Ray ray,
-                                               ccl_global float *buffer)
+ccl_device_inline float kernel_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;
@@ -517,7 +520,7 @@ ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
 			float3 emission;
 
 			if(indirect_lamp_emission(kg, &emission_sd, &state, &light_ray, &emission))
-				path_radiance_accum_emission(&L, throughput, emission, state.bounce);
+				path_radiance_accum_emission(L, throughput, emission, state.bounce);
 		}
 #endif  /* __LAMP_MIS__ */
 
@@ -549,7 +552,7 @@ ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
 
 				/* emission */
 				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);
 
 				/* scattering */
 				VolumeIntegrateResult result = VOLUME_PATH_ATTENUATED;
@@ -559,7 +562,7 @@ ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
 
 					/* direct light sampling */
 					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 sample. if we use distance sampling and take just
@@ -577,7 +580,7 @@ ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
 				kernel_volume_decoupled_free(kg, &volume_segment);
 
 				if(result == VOLUME_PATH_SCATTERED) {
-					if(kernel_path_volume_bounce(kg, rng, &sd, &throughput, &state, &L, &ray))
+					if(kernel_path_volume_bounce(kg, rng, &sd, &throughput, &state, L, &ray))
 						continue;
 					else
 						break;
@@ -591,15 +594,15 @@ ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
 			{
 				/* integrate along volume segment with distance sampling */
 				VolumeIntegrateResult result = kernel_volume_integrate(
-					kg, &state, &sd, &volume_ray, &L, &throughput, rng, heterogeneous);
+					kg, &state, &sd, &volume_ray, L, &throughput, rng, heterogeneous);
 
 #  ifdef __VOLUME_SCATTER__
 				if(result == VOLUME_PATH_SCATTERED) {
 					/* direct lighting */
-					kernel_path_volume_connect_light(kg, rng, &sd, &emission_sd, throughput, &state, &L);
+					kernel_path_volume_connect_light(kg, rng, &sd, &emission_sd, throughput, &state, L);
 
 					/* indirect light bounce */
-					if(kernel_path_volume_bounce(kg, rng, &sd, &throughput, &state, &L, &ray))
+					if(kernel_path_volume_bounce(kg, rng, &sd, &throughput, &state, L, &ray))
 						continue;
 					else
 						break;
@@ -623,7 +626,7 @@ ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
 #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;
@@ -640,10 +643,10 @@ ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
 #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);
 				}
 			}
 		}
@@ -677,7 +680,7 @@ ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
 #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);
 
 		/* blurring of bsdf after bounces, for rays that have a small likelihood
 		 * of following this particular path (diffuse, rough glossy) */
@@ -695,7 +698,7 @@ ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
 		if(sd.flag & SD_EMISSION) {
 			/* todo: is isect.t wrong here for transparent surfaces? */
 			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__ */
 
@@ -715,10 +718,12 @@ ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
 			throughput /= probability;
 		}
 
+		kernel_update_denoising_features(kg, &sd, &state, L);
+
 #ifdef __AO__
 		/* ambient occlusion */
 		if(kernel_data.integrator.use_ambient_occlusion || (sd.flag & SD_AO)) {
-			kernel_path_ao(kg, &sd, &emission_sd, &L, &state, rng, throughput, shader_bsdf_alpha(kg, &sd));
+			kernel_path_ao(kg, &sd, &emission_sd, L, &state, rng, throughput, shader_bsdf_alpha(kg, &sd));
 		}
 #endif  /* __AO__ */
 
@@ -729,7 +734,7 @@ ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
 			if(kernel_path_subsurface_scatter(kg,
 			                                  &sd,
 			                                  &emission_sd,
-			                                  &L,
+			                                  L,
 			                                  &state,
 			                                  rng,
 			                                  &ray,
@@ -742,15 +747,15 @@ ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
 #endif  /* __SUBSURFACE__ */
 
 		/* direct lighting */
-		kernel_path_surface_connect_light(kg, rng, &sd, &emission_sd, throughput, &state, &L);
+		kernel_path_surface_connect_light(kg, rng, &sd, &emission_sd, throughput, &state, L);
 
 		/* compute direct lighting and next bounce */
-		if(!kernel_path_surface_bounce(kg, rng, &sd, &throughput, &state, &L, &ray))
+		if(!kernel_path_surface_bounce(kg, rng, &sd, &throughput, &state, L, &ray))
 			break;
 	}
 
 #ifdef __SUBSURFACE__
-		kernel_path_subsurface_accum_indirect(&ss_indirect, &L);
+		kernel_path_subsurface_accum_indirect(&ss_indirect, L);
 
 		/* Trace indirect subsurface rays by restarting the loop. this uses less
 		 * stack memory than invoking kernel_path_indirect.
@@ -760,7 +765,7 @@ ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
 			                                      &ss_indirect,
 			                                      &state,
 			                                      &ray,
-			                                      &L,
+			                                      L,
 			                                      &throughput);
 		}
 		else {
@@ -769,24 +774,15 @@ ccl_device_inline float4 kernel_path_integrate(KernelGlobals *kg,
 	}
 #endif  /* __SUBSURFACE__ */
 
-	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_path_trace(KernelGlobals *kg,
@@ -807,15 +803,16 @@ ccl_device void kernel_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_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_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);
 }