Cycles: improve sample stratification on area lights for path tracing.

Previously we used a 1D sequence to select a light, and another 2D sequence
to sample a point on the light. For multiple lights this meant each light
would get a random subset of a 2D stratified sequence, which is not
guaranteed to be stratified anymore.

Now we use only a 2D sequence, split into segments along the X axis, one for
each light. The samples that fall within a segment then each are a stratified
sequence, at least in the limit. So for example for two lights, we split up
the unit square into two segments [0,0.5[ x [0,1[ and [0.5,1[ x [0,1[.

This doesn't make much difference in most scenes, mainly helps if you have a
few large area lights or some types of HDR backgrounds.

5 files changed, 29 insertions, 28 deletions

diff --git a/intern/cycles/kernel/kernel_light.h b/intern/cycles/kernel/kernel_light.h
index 59db6cbd430..9f1a48f9bba 100644
--- a/intern/cycles/kernel/kernel_light.h
+++ b/intern/cycles/kernel/kernel_light.h
@@ -1013,20 +1013,21 @@ ccl_device_forceinline void triangle_light_sample(KernelGlobals *kg, int prim, i
 
 /* Light Distribution */
 
-ccl_device int light_distribution_sample(KernelGlobals *kg, float randt)
+ccl_device int light_distribution_sample(KernelGlobals *kg, float *randu)
 {
-	/* this is basically std::upper_bound as used by pbrt, to find a point light or
+	/* This is basically std::upper_bound as used by pbrt, to find a point light or
 	 * triangle to emit from, proportional to area. a good improvement would be to
 	 * also sample proportional to power, though it's not so well defined with
-	 * OSL shaders. */
+	 * arbitrary shaders. */
 	int first = 0;
 	int len = kernel_data.integrator.num_distribution + 1;
+	float r = *randu;
 
 	while(len > 0) {
 		int half_len = len >> 1;
 		int middle = first + half_len;
 
-		if(randt < kernel_tex_fetch(__light_distribution, middle).x) {
+		if(r < kernel_tex_fetch(__light_distribution, middle).x) {
 			len = half_len;
 		}
 		else {
@@ -1035,9 +1036,17 @@ ccl_device int light_distribution_sample(KernelGlobals *kg, float randt)
 		}
 	}
 
-	/* clamping should not be needed but float rounding errors seem to
-	 * make this fail on rare occasions */
-	return clamp(first-1, 0, kernel_data.integrator.num_distribution-1);
+	/* Clamping should not be needed but float rounding errors seem to
+	 * make this fail on rare occasions. */
+	int index = clamp(first-1, 0, kernel_data.integrator.num_distribution-1);
+
+	/* Rescale to reuse random number. this helps the 2D samples within
+	 * each area light be stratified as well. */
+	float distr_min = kernel_tex_fetch(__light_distribution, index).x;
+	float distr_max = kernel_tex_fetch(__light_distribution, index+1).x;
+	*randu = (r - distr_min)/(distr_max - distr_min);
+
+	return index;
 }
 
 /* Generic Light */
@@ -1049,7 +1058,6 @@ ccl_device bool light_select_reached_max_bounces(KernelGlobals *kg, int index, i
 }
 
 ccl_device_noinline bool light_sample(KernelGlobals *kg,
-                                      float randt,
                                       float randu,
                                       float randv,
                                       float time,
@@ -1058,7 +1066,7 @@ ccl_device_noinline bool light_sample(KernelGlobals *kg,
                                       LightSample *ls)
 {
 	/* sample index */
-	int index = light_distribution_sample(kg, randt);
+	int index = light_distribution_sample(kg, &randu);
 
 	/* fetch light data */
 	float4 l = kernel_tex_fetch(__light_distribution, index);
diff --git a/intern/cycles/kernel/kernel_path_surface.h b/intern/cycles/kernel/kernel_path_surface.h
index 6c3a444e48a..e798fcc6a2c 100644
--- a/intern/cycles/kernel/kernel_path_surface.h
+++ b/intern/cycles/kernel/kernel_path_surface.h
@@ -85,17 +85,16 @@ ccl_device_noinline void kernel_branched_path_surface_connect_light(
 			float num_samples_inv = num_samples_adjust/num_samples;
 
 			for(int j = 0; j < num_samples; j++) {
-				float light_t = path_branched_rng_1D(kg, state->rng_hash, state, j, num_samples, PRNG_LIGHT);
 				float light_u, light_v;
 				path_branched_rng_2D(kg, state->rng_hash, state, j, num_samples, PRNG_LIGHT_U, &light_u, &light_v);
 				float terminate = path_branched_rng_light_termination(kg, state->rng_hash, state, j, num_samples);
 
 				/* only sample triangle lights */
 				if(kernel_data.integrator.num_all_lights)
-					light_t = 0.5f*light_t;
+					light_u = 0.5f*light_u;
 
 				LightSample ls;
-				if(light_sample(kg, light_t, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
+				if(light_sample(kg, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
 					/* Same as above, probability needs to be corrected since the sampling was forced to select a mesh light. */
 					if(kernel_data.integrator.num_all_lights)
 						ls.pdf *= 2.0f;
@@ -118,13 +117,12 @@ ccl_device_noinline void kernel_branched_path_surface_connect_light(
 	}
 	else {
 		/* sample one light at random */
-		float light_t = path_state_rng_1D(kg, state, PRNG_LIGHT);
 		float light_u, light_v;
 		path_state_rng_2D(kg, state, PRNG_LIGHT_U, &light_u, &light_v);
 		float terminate = path_state_rng_light_termination(kg, state);
 
 		LightSample ls;
-		if(light_sample(kg, light_t, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
+		if(light_sample(kg, 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, terminate)) {
 				/* trace shadow ray */
@@ -238,7 +236,6 @@ ccl_device_inline void kernel_path_surface_connect_light(KernelGlobals *kg,
 #endif
 
 	/* sample illumination from lights to find path contribution */
-	float light_t = path_state_rng_1D(kg, state, PRNG_LIGHT);
 	float light_u, light_v;
 	path_state_rng_2D(kg, state, PRNG_LIGHT_U, &light_u, &light_v);
 
@@ -251,7 +248,7 @@ ccl_device_inline void kernel_path_surface_connect_light(KernelGlobals *kg,
 #endif
 
 	LightSample ls;
-	if(light_sample(kg, light_t, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
+	if(light_sample(kg, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
 		float terminate = path_state_rng_light_termination(kg, state);
 		if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate)) {
 			/* trace shadow ray */
diff --git a/intern/cycles/kernel/kernel_path_volume.h b/intern/cycles/kernel/kernel_path_volume.h
index c9c7f447c42..fb6b8d56a38 100644
--- a/intern/cycles/kernel/kernel_path_volume.h
+++ b/intern/cycles/kernel/kernel_path_volume.h
@@ -31,7 +31,6 @@ ccl_device_inline void kernel_path_volume_connect_light(
 		return;
 
 	/* sample illumination from lights to find path contribution */
-	float light_t = path_state_rng_1D(kg, state, PRNG_LIGHT);
 	float light_u, light_v;
 	path_state_rng_2D(kg, state, PRNG_LIGHT_U, &light_u, &light_v);
 
@@ -45,7 +44,7 @@ ccl_device_inline void kernel_path_volume_connect_light(
 	light_ray.time = sd->time;
 #  endif
 
-	if(light_sample(kg, light_t, light_u, light_v, sd->time, sd->P, state->bounce, &ls))
+	if(light_sample(kg, light_u, light_v, sd->time, sd->P, state->bounce, &ls))
 	{
 		float terminate = path_state_rng_light_termination(kg, state);
 		if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate)) {
@@ -195,16 +194,15 @@ ccl_device void kernel_branched_path_volume_connect_light(
 
 			for(int j = 0; j < num_samples; j++) {
 				/* sample random position on random triangle */
-				float light_t = path_branched_rng_1D_for_decision(kg, state->rng_hash, state, j, num_samples, PRNG_LIGHT);
 				float light_u, light_v;
 				path_branched_rng_2D(kg, state->rng_hash, state, j, num_samples, PRNG_LIGHT_U, &light_u, &light_v);
 
 				/* only sample triangle lights */
 				if(kernel_data.integrator.num_all_lights)
-					light_t = 0.5f*light_t;
+					light_u = 0.5f*light_u;
 
 				LightSample ls;
-				light_sample(kg, light_t, light_u, light_v, sd->time, ray->P, state->bounce, &ls);
+				light_sample(kg, light_u, light_v, sd->time, ray->P, state->bounce, &ls);
 
 				float3 tp = throughput;
 
@@ -219,7 +217,7 @@ ccl_device void kernel_branched_path_volume_connect_light(
 				kernel_assert(result == VOLUME_PATH_SCATTERED);
 
 				/* todo: split up light_sample so we don't have to call it again with new position */
-				if(light_sample(kg, light_t, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
+				if(light_sample(kg, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
 					if(kernel_data.integrator.num_all_lights)
 						ls.pdf *= 2.0f;
 
@@ -239,12 +237,11 @@ ccl_device void kernel_branched_path_volume_connect_light(
 	}
 	else {
 		/* sample random position on random light */
-		float light_t = path_state_rng_1D(kg, state, PRNG_LIGHT);
 		float light_u, light_v;
 		path_state_rng_2D(kg, state, PRNG_LIGHT_U, &light_u, &light_v);
 
 		LightSample ls;
-		light_sample(kg, light_t, light_u, light_v, sd->time, ray->P, state->bounce, &ls);
+		light_sample(kg, light_u, light_v, sd->time, ray->P, state->bounce, &ls);
 
 		float3 tp = throughput;
 
@@ -259,7 +256,7 @@ ccl_device void kernel_branched_path_volume_connect_light(
 		kernel_assert(result == VOLUME_PATH_SCATTERED);
 
 		/* todo: split up light_sample so we don't have to call it again with new position */
-		if(light_sample(kg, light_t, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
+		if(light_sample(kg, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
 			/* sample random light */
 			float terminate = path_state_rng_light_termination(kg, state);
 			if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate)) {
diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index 8f65c00491c..2c7e9b82feb 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -292,7 +292,7 @@ enum PathTraceDimension {
 	PRNG_BSDF_U = 0,
 	PRNG_BSDF_V = 1,
 	PRNG_BSDF = 2,
-	PRNG_LIGHT = 3,
+	PRNG_UNUSED3 = 3,
 	PRNG_LIGHT_U = 4,
 	PRNG_LIGHT_V = 5,
 	PRNG_LIGHT_TERMINATE = 6,
diff --git a/intern/cycles/kernel/split/kernel_direct_lighting.h b/intern/cycles/kernel/split/kernel_direct_lighting.h
index 8e3f7555550..5009828df92 100644
--- a/intern/cycles/kernel/split/kernel_direct_lighting.h
+++ b/intern/cycles/kernel/split/kernel_direct_lighting.h
@@ -81,14 +81,13 @@ ccl_device void kernel_direct_lighting(KernelGlobals *kg,
 
 		if(flag) {
 			/* Sample illumination from lights to find path contribution. */
-			float light_t = path_state_rng_1D(kg, state, PRNG_LIGHT);
 			float light_u, light_v;
 			path_state_rng_2D(kg, state, PRNG_LIGHT_U, &light_u, &light_v);
 			float terminate = path_state_rng_light_termination(kg, state);
 
 			LightSample ls;
 			if(light_sample(kg,
-			                light_t, light_u, light_v,
+			                light_u, light_v,
 			                sd->time,
 			                sd->P,
 			                state->bounce,