Code refactor: use KernelLight instead of float4 arrays.

Original patch by Stefan with modifications by Brecht.

7 files changed, 233 insertions, 137 deletions

diff --git a/intern/cycles/kernel/geom/geom_object.h b/intern/cycles/kernel/geom/geom_object.h
index 6bb4d134b99..9d833b77bdb 100644
--- a/intern/cycles/kernel/geom/geom_object.h
+++ b/intern/cycles/kernel/geom/geom_object.h
@@ -60,12 +60,17 @@ ccl_device_inline Transform object_fetch_transform(KernelGlobals *kg, int object
 
 ccl_device_inline Transform lamp_fetch_transform(KernelGlobals *kg, int lamp, bool inverse)
 {
-	int offset = lamp*LIGHT_SIZE + (inverse? 8 : 5);
-
 	Transform tfm;
-	tfm.x = kernel_tex_fetch(__light_data, offset + 0);
-	tfm.y = kernel_tex_fetch(__light_data, offset + 1);
-	tfm.z = kernel_tex_fetch(__light_data, offset + 2);
+	if(inverse) {
+		tfm.x = kernel_tex_fetch(__lights, lamp).itfm[0];
+		tfm.y = kernel_tex_fetch(__lights, lamp).itfm[1];
+		tfm.z = kernel_tex_fetch(__lights, lamp).itfm[2];
+	}
+	else {
+		tfm.x = kernel_tex_fetch(__lights, lamp).tfm[0];
+		tfm.y = kernel_tex_fetch(__lights, lamp).tfm[1];
+		tfm.z = kernel_tex_fetch(__lights, lamp).tfm[2];
+	}
 	tfm.w = make_float4(0.0f, 0.0f, 0.0f, 1.0f);
 
 	return tfm;
@@ -240,8 +245,7 @@ ccl_device_inline float lamp_random_number(KernelGlobals *kg, int lamp)
 	if(lamp == LAMP_NONE)
 		return 0.0f;
 
-	float4 f = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 4);
-	return f.y;
+	return kernel_tex_fetch(__lights, lamp).random;
 }
 
 /* Per object random number for shader variation */
diff --git a/intern/cycles/kernel/kernel_light.h b/intern/cycles/kernel/kernel_light.h
index dfa3150dc92..aaf7a7abdd4 100644
--- a/intern/cycles/kernel/kernel_light.h
+++ b/intern/cycles/kernel/kernel_light.h
@@ -255,11 +255,11 @@ ccl_device_inline bool background_portal_data_fetch_and_check_side(KernelGlobals
                                                                    float3 *lightpos,
                                                                    float3 *dir)
 {
-	float4 data0 = kernel_tex_fetch(__light_data, (index + kernel_data.integrator.portal_offset)*LIGHT_SIZE + 0);
-	float4 data3 = kernel_tex_fetch(__light_data, (index + kernel_data.integrator.portal_offset)*LIGHT_SIZE + 3);
+	int portal = kernel_data.integrator.portal_offset + index;
+	const ccl_global KernelLight *klight = &kernel_tex_fetch(__lights, portal);
 
-	*lightpos = make_float3(data0.y, data0.z, data0.w);
-	*dir = make_float3(data3.y, data3.z, data3.w);
+	*lightpos = make_float3(klight->co[0], klight->co[1], klight->co[2]);
+	*dir = make_float3(klight->area.dir[0], klight->area.dir[1], klight->area.dir[2]);
 
 	/* Check whether portal is on the right side. */
 	if(dot(*dir, P - *lightpos) > 1e-4f)
@@ -291,11 +291,10 @@ ccl_device_inline float background_portal_pdf(KernelGlobals *kg,
 		}
 		num_possible++;
 
-		float4 data1 = kernel_tex_fetch(__light_data, (p + kernel_data.integrator.portal_offset)*LIGHT_SIZE + 1);
-		float4 data2 = kernel_tex_fetch(__light_data, (p + kernel_data.integrator.portal_offset)*LIGHT_SIZE + 2);
-
-		float3 axisu = make_float3(data1.y, data1.z, data1.w);
-		float3 axisv = make_float3(data2.y, data2.z, data2.w);
+		int portal = kernel_data.integrator.portal_offset + p;
+		const ccl_global KernelLight *klight = &kernel_tex_fetch(__lights, portal);
+		float3 axisu = make_float3(klight->area.axisu[0], klight->area.axisu[1], klight->area.axisu[2]);
+		float3 axisv = make_float3(klight->area.axisv[0], klight->area.axisv[1], klight->area.axisv[2]);
 
 		if(!ray_quad_intersect(P, direction, 1e-4f, FLT_MAX, lightpos, axisu, axisv, dir, NULL, NULL, NULL, NULL))
 			continue;
@@ -346,10 +345,10 @@ ccl_device float3 background_portal_sample(KernelGlobals *kg,
 
 		if(portal == 0) {
 			/* p is the portal to be sampled. */
-			float4 data1 = kernel_tex_fetch(__light_data, (p + kernel_data.integrator.portal_offset)*LIGHT_SIZE + 1);
-			float4 data2 = kernel_tex_fetch(__light_data, (p + kernel_data.integrator.portal_offset)*LIGHT_SIZE + 2);
-			float3 axisu = make_float3(data1.y, data1.z, data1.w);
-			float3 axisv = make_float3(data2.y, data2.z, data2.w);
+			int portal = kernel_data.integrator.portal_offset + p;
+			const ccl_global KernelLight *klight = &kernel_tex_fetch(__lights, portal);
+			float3 axisu = make_float3(klight->area.axisu[0], klight->area.axisu[1], klight->area.axisu[2]);
+			float3 axisv = make_float3(klight->area.axisv[0], klight->area.axisv[1], klight->area.axisv[2]);
 
 			*pdf = area_light_sample(P, &lightpos,
 			                         axisu, axisv,
@@ -479,14 +478,10 @@ ccl_device float3 sphere_light_sample(float3 P, float3 center, float radius, flo
 	return disk_light_sample(normalize(P - center), randu, randv)*radius;
 }
 
-ccl_device float spot_light_attenuation(float4 data1, float4 data2, LightSample *ls)
+ccl_device float spot_light_attenuation(float3 dir, float spot_angle, float spot_smooth, LightSample *ls)
 {
-	float3 dir = make_float3(data2.y, data2.z, data2.w);
 	float3 I = ls->Ng;
 
-	float spot_angle = data1.w;
-	float spot_smooth = data2.x;
-
 	float attenuation = dot(dir, I);
 
 	if(attenuation <= spot_angle) {
@@ -518,12 +513,10 @@ ccl_device_inline bool lamp_light_sample(KernelGlobals *kg,
                                          float3 P,
                                          LightSample *ls)
 {
-	float4 data0 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 0);
-	float4 data1 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 1);
-
-	LightType type = (LightType)__float_as_int(data0.x);
+	const ccl_global KernelLight *klight = &kernel_tex_fetch(__lights, lamp);
+	LightType type = (LightType)klight->type;
 	ls->type = type;
-	ls->shader = __float_as_int(data1.x);
+	ls->shader = klight->shader_id;
 	ls->object = PRIM_NONE;
 	ls->prim = PRIM_NONE;
 	ls->lamp = lamp;
@@ -532,10 +525,10 @@ ccl_device_inline bool lamp_light_sample(KernelGlobals *kg,
 
 	if(type == LIGHT_DISTANT) {
 		/* distant light */
-		float3 lightD = make_float3(data0.y, data0.z, data0.w);
+		float3 lightD = make_float3(klight->co[0], klight->co[1], klight->co[2]);
 		float3 D = lightD;
-		float radius = data1.y;
-		float invarea = data1.w;
+		float radius = klight->distant.radius;
+		float invarea = klight->distant.invarea;
 
 		if(radius > 0.0f)
 			D = distant_light_sample(D, radius, randu, randv);
@@ -562,10 +555,10 @@ ccl_device_inline bool lamp_light_sample(KernelGlobals *kg,
 	}
 #endif
 	else {
-		ls->P = make_float3(data0.y, data0.z, data0.w);
+		ls->P = make_float3(klight->co[0], klight->co[1], klight->co[2]);
 
 		if(type == LIGHT_POINT || type == LIGHT_SPOT) {
-			float radius = data1.y;
+			float radius = klight->spot.radius;
 
 			if(radius > 0.0f)
 				/* sphere light */
@@ -574,14 +567,19 @@ ccl_device_inline bool lamp_light_sample(KernelGlobals *kg,
 			ls->D = normalize_len(ls->P - P, &ls->t);
 			ls->Ng = -ls->D;
 
-			float invarea = data1.z;
+			float invarea = klight->spot.invarea;
 			ls->eval_fac = (0.25f*M_1_PI_F)*invarea;
 			ls->pdf = invarea;
 
 			if(type == LIGHT_SPOT) {
 				/* spot light attenuation */
-				float4 data2 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 2);
-				ls->eval_fac *= spot_light_attenuation(data1, data2, ls);
+				float3 dir = make_float3(klight->spot.dir[0],
+                                         klight->spot.dir[1],
+				                         klight->spot.dir[2]);
+				ls->eval_fac *= spot_light_attenuation(dir,
+				                                       klight->spot.spot_angle,
+				                                       klight->spot.spot_smooth,
+				                                       ls);
 				if(ls->eval_fac == 0.0f) {
 					return false;
 				}
@@ -594,12 +592,15 @@ ccl_device_inline bool lamp_light_sample(KernelGlobals *kg,
 		}
 		else {
 			/* area light */
-			float4 data2 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 2);
-			float4 data3 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 3);
-
-			float3 axisu = make_float3(data1.y, data1.z, data1.w);
-			float3 axisv = make_float3(data2.y, data2.z, data2.w);
-			float3 D = make_float3(data3.y, data3.z, data3.w);
+			float3 axisu = make_float3(klight->area.axisu[0],
+			                           klight->area.axisu[1],
+			                           klight->area.axisu[2]);
+			float3 axisv = make_float3(klight->area.axisv[0],
+			                           klight->area.axisv[1],
+			                           klight->area.axisv[2]);
+			float3 D = make_float3(klight->area.dir[0],
+			                       klight->area.dir[1],
+			                       klight->area.dir[2]);
 
 			if(dot(ls->P - P, D) > 0.0f) {
 				return false;
@@ -618,7 +619,7 @@ ccl_device_inline bool lamp_light_sample(KernelGlobals *kg,
 			ls->Ng = D;
 			ls->D = normalize_len(ls->P - P, &ls->t);
 
-			float invarea = data2.x;
+			float invarea = klight->area.invarea;
 			ls->eval_fac = 0.25f*invarea;
 		}
 	}
@@ -630,12 +631,10 @@ ccl_device_inline bool lamp_light_sample(KernelGlobals *kg,
 
 ccl_device bool lamp_light_eval(KernelGlobals *kg, int lamp, float3 P, float3 D, float t, LightSample *ls)
 {
-	float4 data0 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 0);
-	float4 data1 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 1);
-
-	LightType type = (LightType)__float_as_int(data0.x);
+	const ccl_global KernelLight *klight = &kernel_tex_fetch(__lights, lamp);
+	LightType type = (LightType)klight->type;
 	ls->type = type;
-	ls->shader = __float_as_int(data1.x);
+	ls->shader = klight->shader_id;
 	ls->object = PRIM_NONE;
 	ls->prim = PRIM_NONE;
 	ls->lamp = lamp;
@@ -648,7 +647,7 @@ ccl_device bool lamp_light_eval(KernelGlobals *kg, int lamp, float3 P, float3 D,
 
 	if(type == LIGHT_DISTANT) {
 		/* distant light */
-		float radius = data1.y;
+		float radius = klight->distant.radius;
 
 		if(radius == 0.0f)
 			return false;
@@ -670,9 +669,9 @@ ccl_device bool lamp_light_eval(KernelGlobals *kg, int lamp, float3 P, float3 D,
 		 *             P
 		 */
 
-		float3 lightD = make_float3(data0.y, data0.z, data0.w);
+		float3 lightD = make_float3(klight->co[0], klight->co[1], klight->co[2]);
 		float costheta = dot(-lightD, D);
-		float cosangle = data1.z;
+		float cosangle = klight->distant.cosangle;
 
 		if(costheta < cosangle)
 			return false;
@@ -683,13 +682,14 @@ ccl_device bool lamp_light_eval(KernelGlobals *kg, int lamp, float3 P, float3 D,
 		ls->t = FLT_MAX;
 
 		/* compute pdf */
-		float invarea = data1.w;
+		float invarea = klight->distant.invarea;
 		ls->pdf = invarea/(costheta*costheta*costheta);
 		ls->eval_fac = ls->pdf;
 	}
 	else if(type == LIGHT_POINT || type == LIGHT_SPOT) {
-		float3 lightP = make_float3(data0.y, data0.z, data0.w);
-		float radius = data1.y;
+		float3 lightP = make_float3(klight->co[0], klight->co[1], klight->co[2]);
+
+		float radius = klight->spot.radius;
 
 		/* sphere light */
 		if(radius == 0.0f)
@@ -704,14 +704,19 @@ ccl_device bool lamp_light_eval(KernelGlobals *kg, int lamp, float3 P, float3 D,
 		ls->Ng = -D;
 		ls->D = D;
 
-		float invarea = data1.z;
+		float invarea = klight->spot.invarea;
 		ls->eval_fac = (0.25f*M_1_PI_F)*invarea;
 		ls->pdf = invarea;
 
 		if(type == LIGHT_SPOT) {
 			/* spot light attenuation */
-			float4 data2 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 2);
-			ls->eval_fac *= spot_light_attenuation(data1, data2, ls);
+			float3 dir = make_float3(klight->spot.dir[0],
+			                         klight->spot.dir[1],
+			                         klight->spot.dir[2]);
+			ls->eval_fac *= spot_light_attenuation(dir,
+			                                       klight->spot.spot_angle,
+			                                       klight->spot.spot_smooth,
+			                                       ls);
 
 			if(ls->eval_fac == 0.0f)
 				return false;
@@ -726,22 +731,25 @@ ccl_device bool lamp_light_eval(KernelGlobals *kg, int lamp, float3 P, float3 D,
 	}
 	else if(type == LIGHT_AREA) {
 		/* area light */
-		float4 data2 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 2);
-		float4 data3 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 3);
-
-		float invarea = data2.x;
+		float invarea = klight->area.invarea;
 		if(invarea == 0.0f)
 			return false;
 
-		float3 axisu = make_float3(data1.y, data1.z, data1.w);
-		float3 axisv = make_float3(data2.y, data2.z, data2.w);
-		float3 Ng = make_float3(data3.y, data3.z, data3.w);
+		float3 axisu = make_float3(klight->area.axisu[0],
+		                           klight->area.axisu[1],
+		                           klight->area.axisu[2]);
+		float3 axisv = make_float3(klight->area.axisv[0],
+		                           klight->area.axisv[1],
+		                           klight->area.axisv[2]);
+		float3 Ng = make_float3(klight->area.dir[0],
+		                        klight->area.dir[1],
+		                        klight->area.dir[2]);
 
 		/* one sided */
 		if(dot(D, Ng) >= 0.0f)
 			return false;
 
-		float3 light_P = make_float3(data0.y, data0.z, data0.w);
+		float3 light_P = make_float3(klight->co[0], klight->co[1], klight->co[2]);
 
 		if(!ray_quad_intersect(P, D, 0.0f, t, light_P,
 		                       axisu, axisv, Ng,
@@ -1040,7 +1048,7 @@ ccl_device int light_distribution_sample(KernelGlobals *kg, float *randu)
 		int half_len = len >> 1;
 		int middle = first + half_len;
 
-		if(r < kernel_tex_fetch(__light_distribution, middle).x) {
+		if(r < kernel_tex_fetch(__light_distribution, middle).totarea) {
 			len = half_len;
 		}
 		else {
@@ -1055,8 +1063,8 @@ ccl_device int light_distribution_sample(KernelGlobals *kg, float *randu)
 
 	/* 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;
+	float distr_min = kernel_tex_fetch(__light_distribution, index).totarea;
+	float distr_max = kernel_tex_fetch(__light_distribution, index+1).totarea;
 	*randu = (r - distr_min)/(distr_max - distr_min);
 
 	return index;
@@ -1066,8 +1074,7 @@ ccl_device int light_distribution_sample(KernelGlobals *kg, float *randu)
 
 ccl_device bool light_select_reached_max_bounces(KernelGlobals *kg, int index, int bounce)
 {
-	float4 data4 = kernel_tex_fetch(__light_data, index*LIGHT_SIZE + 4);
-	return (bounce > __float_as_int(data4.x));
+	return (bounce > kernel_tex_fetch(__lights, index).max_bounces);
 }
 
 ccl_device_noinline bool light_sample(KernelGlobals *kg,
@@ -1082,12 +1089,12 @@ ccl_device_noinline bool light_sample(KernelGlobals *kg,
 	int index = light_distribution_sample(kg, &randu);
 
 	/* fetch light data */
-	float4 l = kernel_tex_fetch(__light_distribution, index);
-	int prim = __float_as_int(l.y);
+	const ccl_global KernelLightDistribution *kdistribution = &kernel_tex_fetch(__light_distribution, index);
+	int prim = kdistribution->prim;
 
 	if(prim >= 0) {
-		int object = __float_as_int(l.w);
-		int shader_flag = __float_as_int(l.z);
+		int object = kdistribution->mesh_light.object_id;
+		int shader_flag = kdistribution->mesh_light.shader_flag;
 
 		triangle_light_sample(kg, prim, object, randu, randv, time, ls, P);
 		ls->shader |= shader_flag;
@@ -1106,8 +1113,7 @@ ccl_device_noinline bool light_sample(KernelGlobals *kg,
 
 ccl_device int light_select_num_samples(KernelGlobals *kg, int index)
 {
-	float4 data3 = kernel_tex_fetch(__light_data, index*LIGHT_SIZE + 3);
-	return __float_as_int(data3.x);
+	return kernel_tex_fetch(__lights, index).samples;
 }
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/kernel/kernel_textures.h b/intern/cycles/kernel/kernel_textures.h
index 427ed142e7f..601165a3848 100644
--- a/intern/cycles/kernel/kernel_textures.h
+++ b/intern/cycles/kernel/kernel_textures.h
@@ -55,8 +55,8 @@ KERNEL_TEX(float4, __attributes_float3)
 KERNEL_TEX(uchar4, __attributes_uchar4)
 
 /* lights */
-KERNEL_TEX(float4, __light_distribution)
-KERNEL_TEX(float4, __light_data)
+KERNEL_TEX(KernelLightDistribution, __light_distribution)
+KERNEL_TEX(KernelLight, __lights)
 KERNEL_TEX(float2, __light_background_marginal_cdf)
 KERNEL_TEX(float2, __light_background_conditional_cdf)
 
diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index b6f1a3a6d98..198ce39c63f 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -36,7 +36,6 @@ CCL_NAMESPACE_BEGIN
 
 /* Constants */
 #define OBJECT_VECTOR_SIZE	6
-#define LIGHT_SIZE		11
 #define FILTER_TABLE_SIZE	1024
 #define RAMP_TABLE_SIZE		256
 #define SHUTTER_TABLE_SIZE		256
@@ -1456,6 +1455,61 @@ typedef struct KernelObject {
 } KernelObject;;
 static_assert_align(KernelObject, 16);
 
+typedef struct KernelSpotLight {
+	float radius;
+	float invarea;
+	float spot_angle;
+	float spot_smooth;
+	float dir[3];
+} KernelSpotLight;
+
+/* PointLight is SpotLight with only radius and invarea being used. */
+
+typedef struct KernelAreaLight {
+	float axisu[3];
+	float invarea;
+	float axisv[3];
+	float dir[3];
+} KernelAreaLight;
+
+typedef struct KernelDistantLight {
+	float radius;
+	float cosangle;
+	float invarea;
+} KernelDistantLight;
+
+typedef struct KernelLight {
+	int type;
+	float co[3];
+	int shader_id;
+	int samples;
+	float max_bounces;
+	float random;
+	float4 tfm[3];
+	float4 itfm[3];
+	union {
+		KernelSpotLight spot;
+		KernelAreaLight area;
+		KernelDistantLight distant;
+	};
+} KernelLight;
+static_assert_align(KernelLight, 16);
+
+typedef struct KernelLightDistribution {
+	float totarea;
+	int prim;
+	union {
+		struct {
+			int shader_flag;
+			int object_id;
+		} mesh_light;
+		struct {
+			float pad;
+			float size;
+		} lamp;
+	};
+} KernelLightDistribution;
+static_assert_align(KernelLightDistribution, 16);
 
 /* Declarations required for split kernel */
 
diff --git a/intern/cycles/render/light.cpp b/intern/cycles/render/light.cpp
index b62453cf5fc..a3f67361f6e 100644
--- a/intern/cycles/render/light.cpp
+++ b/intern/cycles/render/light.cpp
@@ -288,7 +288,7 @@ void LightManager::device_update_distribution(Device *, DeviceScene *dscene, Sce
 	VLOG(1) << "Total " << num_distribution << " of light distribution primitives.";
 
 	/* emission area */
-	float4 *distribution = dscene->light_distribution.alloc(num_distribution + 1);
+	KernelLightDistribution *distribution = dscene->light_distribution.alloc(num_distribution + 1);
 	float totarea = 0.0f;
 
 	/* triangles */
@@ -334,10 +334,10 @@ void LightManager::device_update_distribution(Device *, DeviceScene *dscene, Sce
 			                         : scene->default_surface;
 
 			if(shader->use_mis && shader->has_surface_emission) {
-				distribution[offset].x = totarea;
-				distribution[offset].y = __int_as_float(i + mesh->tri_offset);
-				distribution[offset].z = __int_as_float(shader_flag);
-				distribution[offset].w = __int_as_float(object_id);
+				distribution[offset].totarea = totarea;
+				distribution[offset].prim = i + mesh->tri_offset;
+				distribution[offset].mesh_light.shader_flag = shader_flag;
+				distribution[offset].mesh_light.object_id = object_id;
 				offset++;
 
 				Mesh::Triangle t = mesh->get_triangle(i);
@@ -372,10 +372,10 @@ void LightManager::device_update_distribution(Device *, DeviceScene *dscene, Sce
 		if(!light->is_enabled)
 			continue;
 
-		distribution[offset].x = totarea;
-		distribution[offset].y = __int_as_float(~light_index);
-		distribution[offset].z = 1.0f;
-		distribution[offset].w = light->size;
+		distribution[offset].totarea = totarea;
+		distribution[offset].prim = ~light_index;
+		distribution[offset].lamp.pad = 1.0f;
+		distribution[offset].lamp.size = light->size;
 		totarea += lightarea;
 
 		if(light->size > 0.0f && light->use_mis)
@@ -390,15 +390,15 @@ void LightManager::device_update_distribution(Device *, DeviceScene *dscene, Sce
 	}
 
 	/* normalize cumulative distribution functions */
-	distribution[num_distribution].x = totarea;
-	distribution[num_distribution].y = 0.0f;
-	distribution[num_distribution].z = 0.0f;
-	distribution[num_distribution].w = 0.0f;
+	distribution[num_distribution].totarea = totarea;
+	distribution[num_distribution].prim = 0.0f;
+	distribution[num_distribution].lamp.pad = 0.0f;
+	distribution[num_distribution].lamp.size = 0.0f;
 
 	if(totarea > 0.0f) {
 		for(size_t i = 0; i < num_distribution; i++)
-			distribution[i].x /= totarea;
-		distribution[num_distribution].x = 1.0f;
+			distribution[i].totarea /= totarea;
+		distribution[num_distribution].totarea = 1.0f;
 	}
 
 	if(progress.get_cancel()) return;
@@ -620,7 +620,7 @@ void LightManager::device_update_points(Device *,
 		}
 	}
 
-	float4 *light_data = dscene->light_data.alloc(num_lights*LIGHT_SIZE);
+	KernelLight *klights = dscene->lights.alloc(num_lights);
 
 	if(num_lights == 0) {
 		VLOG(1) << "No effective light, ignoring points update.";
@@ -637,8 +637,8 @@ void LightManager::device_update_points(Device *,
 		float3 co = light->co;
 		Shader *shader = (light->shader) ? light->shader : scene->default_light;
 		int shader_id = scene->shader_manager->get_shader_id(shader);
-		float samples = __int_as_float(light->samples);
-		float max_bounces = __int_as_float(light->max_bounces);
+		int samples = light->samples;
+		int max_bounces = light->max_bounces;
 		float random = (float)light->random_id * (1.0f/(float)0xFFFFFFFF);
 
 		if(!light->cast_shadow)
@@ -661,6 +661,9 @@ void LightManager::device_update_points(Device *,
 			use_light_visibility = true;
 		}
 
+		klights[light_index].type = light->type;
+		klights[light_index].samples = samples;
+
 		if(light->type == LIGHT_POINT) {
 			shader_id &= ~SHADER_AREA_LIGHT;
 
@@ -670,10 +673,12 @@ void LightManager::device_update_points(Device *,
 			if(light->use_mis && radius > 0.0f)
 				shader_id |= SHADER_USE_MIS;
 
-			light_data[light_index*LIGHT_SIZE + 0] = make_float4(__int_as_float(light->type), co.x, co.y, co.z);
-			light_data[light_index*LIGHT_SIZE + 1] = make_float4(__int_as_float(shader_id), radius, invarea, 0.0f);
-			light_data[light_index*LIGHT_SIZE + 2] = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
-			light_data[light_index*LIGHT_SIZE + 3] = make_float4(samples, 0.0f, 0.0f, 0.0f);
+			klights[light_index].co[0] = co.x;
+			klights[light_index].co[1] = co.y;
+			klights[light_index].co[2] = co.z;
+
+			klights[light_index].spot.radius = radius;
+			klights[light_index].spot.invarea = invarea;
 		}
 		else if(light->type == LIGHT_DISTANT) {
 			shader_id &= ~SHADER_AREA_LIGHT;
@@ -690,10 +695,13 @@ void LightManager::device_update_points(Device *,
 			if(light->use_mis && area > 0.0f)
 				shader_id |= SHADER_USE_MIS;
 
-			light_data[light_index*LIGHT_SIZE + 0] = make_float4(__int_as_float(light->type), dir.x, dir.y, dir.z);
-			light_data[light_index*LIGHT_SIZE + 1] = make_float4(__int_as_float(shader_id), radius, cosangle, invarea);
-			light_data[light_index*LIGHT_SIZE + 2] = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
-			light_data[light_index*LIGHT_SIZE + 3] = make_float4(samples, 0.0f, 0.0f, 0.0f);
+			klights[light_index].co[0] = dir.x;
+			klights[light_index].co[1] = dir.y;
+			klights[light_index].co[2] = dir.z;
+
+			klights[light_index].distant.invarea = invarea;
+			klights[light_index].distant.radius = radius;
+			klights[light_index].distant.cosangle = cosangle;
 		}
 		else if(light->type == LIGHT_BACKGROUND) {
 			uint visibility = scene->background->visibility;
@@ -717,11 +725,6 @@ void LightManager::device_update_points(Device *,
 				shader_id |= SHADER_EXCLUDE_SCATTER;
 				use_light_visibility = true;
 			}
-
-			light_data[light_index*LIGHT_SIZE + 0] = make_float4(__int_as_float(light->type), 0.0f, 0.0f, 0.0f);
-			light_data[light_index*LIGHT_SIZE + 1] = make_float4(__int_as_float(shader_id), 0.0f, 0.0f, 0.0f);
-			light_data[light_index*LIGHT_SIZE + 2] = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
-			light_data[light_index*LIGHT_SIZE + 3] = make_float4(samples, 0.0f, 0.0f, 0.0f);
 		}
 		else if(light->type == LIGHT_AREA) {
 			float3 axisu = light->axisu*(light->sizeu*light->size);
@@ -735,10 +738,20 @@ void LightManager::device_update_points(Device *,
 			if(light->use_mis && area > 0.0f)
 				shader_id |= SHADER_USE_MIS;
 
-			light_data[light_index*LIGHT_SIZE + 0] = make_float4(__int_as_float(light->type), co.x, co.y, co.z);
-			light_data[light_index*LIGHT_SIZE + 1] = make_float4(__int_as_float(shader_id), axisu.x, axisu.y, axisu.z);
-			light_data[light_index*LIGHT_SIZE + 2] = make_float4(invarea, axisv.x, axisv.y, axisv.z);
-			light_data[light_index*LIGHT_SIZE + 3] = make_float4(samples, dir.x, dir.y, dir.z);
+			klights[light_index].co[0] = co.x;
+			klights[light_index].co[1] = co.y;
+			klights[light_index].co[2] = co.z;
+
+			klights[light_index].area.axisu[0] = axisu.x;
+			klights[light_index].area.axisu[1] = axisu.y;
+			klights[light_index].area.axisu[2] = axisu.z;
+			klights[light_index].area.axisv[0] = axisv.x;
+			klights[light_index].area.axisv[1] = axisv.y;
+			klights[light_index].area.axisv[2] = axisv.z;
+			klights[light_index].area.invarea = invarea;
+			klights[light_index].area.dir[0] = dir.x;
+			klights[light_index].area.dir[1] = dir.y;
+			klights[light_index].area.dir[2] = dir.z;
 		}
 		else if(light->type == LIGHT_SPOT) {
 			shader_id &= ~SHADER_AREA_LIGHT;
@@ -754,18 +767,29 @@ void LightManager::device_update_points(Device *,
 			if(light->use_mis && radius > 0.0f)
 				shader_id |= SHADER_USE_MIS;
 
-			light_data[light_index*LIGHT_SIZE + 0] = make_float4(__int_as_float(light->type), co.x, co.y, co.z);
-			light_data[light_index*LIGHT_SIZE + 1] = make_float4(__int_as_float(shader_id), radius, invarea, spot_angle);
-			light_data[light_index*LIGHT_SIZE + 2] = make_float4(spot_smooth, dir.x, dir.y, dir.z);
-			light_data[light_index*LIGHT_SIZE + 3] = make_float4(samples, 0.0f, 0.0f, 0.0f);
+			klights[light_index].co[0] = co.x;
+			klights[light_index].co[1] = co.y;
+			klights[light_index].co[2] = co.z;
+
+			klights[light_index].spot.radius = radius;
+			klights[light_index].spot.invarea = invarea;
+			klights[light_index].spot.spot_angle = spot_angle;
+			klights[light_index].spot.spot_smooth = spot_smooth;
+			klights[light_index].spot.dir[0] = dir.x;
+			klights[light_index].spot.dir[1] = dir.y;
+			klights[light_index].spot.dir[2] = dir.z;
 		}
 
-		light_data[light_index*LIGHT_SIZE + 4] = make_float4(max_bounces, random, 0.0f, 0.0f);
+		klights[light_index].shader_id = shader_id;
+
+		klights[light_index].max_bounces = max_bounces;
+		klights[light_index].random = random;
+
 
 		Transform tfm = light->tfm;
 		Transform itfm = transform_inverse(tfm);
-		memcpy(&light_data[light_index*LIGHT_SIZE + 5], &tfm, sizeof(float4)*3);
-		memcpy(&light_data[light_index*LIGHT_SIZE + 8], &itfm, sizeof(float4)*3);
+		memcpy(&klights[light_index].tfm, &tfm, sizeof(float4)*3);
+		memcpy(&klights[light_index].itfm, &itfm, sizeof(float4)*3);
 
 		light_index++;
 	}
@@ -782,21 +806,29 @@ void LightManager::device_update_points(Device *,
 		float3 axisu = light->axisu*(light->sizeu*light->size);
 		float3 axisv = light->axisv*(light->sizev*light->size);
 		float area = len(axisu)*len(axisv);
-		float invarea = (area > 0.0f) ? 1.0f / area : 1.0f;
+		float invarea = (area > 0.0f)? 1.0f/area: 1.0f;
 		float3 dir = light->dir;
 
 		dir = safe_normalize(dir);
 
-		light_data[light_index*LIGHT_SIZE + 0] = make_float4(__int_as_float(light->type), co.x, co.y, co.z);
-		light_data[light_index*LIGHT_SIZE + 1] = make_float4(area, axisu.x, axisu.y, axisu.z);
-		light_data[light_index*LIGHT_SIZE + 2] = make_float4(invarea, axisv.x, axisv.y, axisv.z);
-		light_data[light_index*LIGHT_SIZE + 3] = make_float4(-1, dir.x, dir.y, dir.z);
-		light_data[light_index*LIGHT_SIZE + 4] = make_float4(-1, 0.0f, 0.0f, 0.0f);
-
+		klights[light_index].co[0] = co.x;
+		klights[light_index].co[1] = co.y;
+		klights[light_index].co[2] = co.z;
+
+		klights[light_index].area.axisu[0] = axisu.x;
+		klights[light_index].area.axisu[1] = axisu.y;
+		klights[light_index].area.axisu[2] = axisu.z;
+		klights[light_index].area.axisv[0] = axisv.x;
+		klights[light_index].area.axisv[1] = axisv.y;
+		klights[light_index].area.axisv[2] = axisv.z;
+		klights[light_index].area.invarea = invarea;
+		klights[light_index].area.dir[0] = dir.x;
+		klights[light_index].area.dir[1] = dir.y;
+		klights[light_index].area.dir[2] = dir.z;
 		Transform tfm = light->tfm;
 		Transform itfm = transform_inverse(tfm);
-		memcpy(&light_data[light_index*LIGHT_SIZE + 5], &tfm, sizeof(float4)*3);
-		memcpy(&light_data[light_index*LIGHT_SIZE + 8], &itfm, sizeof(float4)*3);
+		memcpy(&klights[light_index].tfm, &tfm, sizeof(float4)*3);
+		memcpy(&klights[light_index].itfm, &itfm, sizeof(float4)*3);
 
 		light_index++;
 	}
@@ -806,7 +838,7 @@ void LightManager::device_update_points(Device *,
 	VLOG(1) << "Number of lights without contribution: "
 	        << num_scene_lights - light_index;
 
-	dscene->light_data.copy_to_device();
+	dscene->lights.copy_to_device();
 }
 
 void LightManager::device_update(Device *device, DeviceScene *dscene, Scene *scene, Progress& progress)
@@ -842,7 +874,7 @@ void LightManager::device_update(Device *device, DeviceScene *dscene, Scene *sce
 void LightManager::device_free(Device *, DeviceScene *dscene)
 {
 	dscene->light_distribution.free();
-	dscene->light_data.free();
+	dscene->lights.free();
 	dscene->light_background_marginal_cdf.free();
 	dscene->light_background_conditional_cdf.free();
 }
diff --git a/intern/cycles/render/scene.cpp b/intern/cycles/render/scene.cpp
index 24923b650e3..aca28fc32fb 100644
--- a/intern/cycles/render/scene.cpp
+++ b/intern/cycles/render/scene.cpp
@@ -66,7 +66,7 @@ DeviceScene::DeviceScene(Device *device)
   attributes_float3(device, "__attributes_float3", MEM_TEXTURE),
   attributes_uchar4(device, "__attributes_uchar4", MEM_TEXTURE),
   light_distribution(device, "__light_distribution", MEM_TEXTURE),
-  light_data(device, "__light_data", MEM_TEXTURE),
+  lights(device, "__lights", MEM_TEXTURE),
   light_background_marginal_cdf(device, "__light_background_marginal_cdf", MEM_TEXTURE),
   light_background_conditional_cdf(device, "__light_background_conditional_cdf", MEM_TEXTURE),
   particles(device, "__particles", MEM_TEXTURE),
diff --git a/intern/cycles/render/scene.h b/intern/cycles/render/scene.h
index 48e9ae4117a..3f089b9138f 100644
--- a/intern/cycles/render/scene.h
+++ b/intern/cycles/render/scene.h
@@ -96,8 +96,8 @@ public:
 	device_vector<uchar4> attributes_uchar4;
 
 	/* lights */
-	device_vector<float4> light_distribution;
-	device_vector<float4> light_data;
+	device_vector<KernelLightDistribution> light_distribution;
+	device_vector<KernelLight> lights;
 	device_vector<float2> light_background_marginal_cdf;
 	device_vector<float2> light_background_conditional_cdf;