Cycles microdisplacement: ngons and attributes for subdivision meshes

This adds support for ngons and attributes on subdivision meshes. Ngons are
needed for proper attribute interpolation as well as correct Catmull-Clark
subdivision. Several changes are made to achieve this:

- new primitive `SubdFace` added to `Mesh`
- 3 more textures are used to store info on patches from subd meshes
- Blender export uses loop interface instead of tessface for subd meshes
- `Attribute` class is updated with a simplified way to pass primitive counts
  around and to support ngons.
- extra points for ngons are generated for O(1) attribute interpolation
- curves are temporally disabled on subd meshes to avoid various bugs with
  implementation
- old unneeded code is removed from `subd/`
- various fixes and improvements

Reviewed By: brecht

Differential Revision: https://developer.blender.org/D2108

30 files changed, 1229 insertions, 1193 deletions

diff --git a/intern/cycles/app/cycles_xml.cpp b/intern/cycles/app/cycles_xml.cpp
index 3aca46e2dc7..3d3aca33881 100644
--- a/intern/cycles/app/cycles_xml.cpp
+++ b/intern/cycles/app/cycles_xml.cpp
@@ -35,7 +35,6 @@
 #include "shader.h"
 #include "scene.h"
 
-#include "subd_mesh.h"
 #include "subd_patch.h"
 #include "subd_split.h"
 
@@ -417,6 +416,7 @@ static void xml_read_mesh(const XMLReadState& state, pugi::xml_node node)
 	xml_read_int_array(verts, node, "verts");
 	xml_read_int_array(nverts, node, "nverts");
 
+#if 0
 	if(xml_equal_string(node, "subdivision", "catmull-clark")) {
 		/* create subd mesh */
 		SubdMesh sdmesh;
@@ -460,7 +460,9 @@ static void xml_read_mesh(const XMLReadState& state, pugi::xml_node node)
 		DiagSplit dsplit(sdparams);
 		sdmesh.tessellate(&dsplit);
 	}
-	else {
+	else
+#endif
+	{
 		/* create vertices */
 		mesh->verts = P;
 
@@ -568,7 +570,7 @@ static void xml_read_patch(const XMLReadState& state, pugi::xml_node node)
 		mesh->used_shaders.push_back(state.shader);
 
 		/* split */
-		SubdParams sdparams(mesh, 0, state.smooth);
+		SubdParams sdparams(mesh);
 		xml_read_float(&sdparams.dicing_rate, node, "dicing_rate");
 
 		DiagSplit dsplit(sdparams);
diff --git a/intern/cycles/blender/blender_mesh.cpp b/intern/cycles/blender/blender_mesh.cpp
index ec11a893b5a..0744bbf1a0a 100644
--- a/intern/cycles/blender/blender_mesh.cpp
+++ b/intern/cycles/blender/blender_mesh.cpp
@@ -24,7 +24,6 @@
 #include "blender_session.h"
 #include "blender_util.h"
 
-#include "subd_mesh.h"
 #include "subd_patch.h"
 #include "subd_split.h"
 
@@ -335,44 +334,71 @@ static void attr_create_vertex_color(Scene *scene,
                                      Mesh *mesh,
                                      BL::Mesh& b_mesh,
                                      const vector<int>& nverts,
-                                     const vector<int>& face_flags)
+                                     const vector<int>& face_flags,
+                                     bool subdivision)
 {
-	BL::Mesh::tessface_vertex_colors_iterator l;
-	for(b_mesh.tessface_vertex_colors.begin(l); l != b_mesh.tessface_vertex_colors.end(); ++l) {
-		if(!mesh->need_attribute(scene, ustring(l->name().c_str())))
-			continue;
+	if(subdivision) {
+		BL::Mesh::vertex_colors_iterator l;
 
-		Attribute *attr = mesh->attributes.add(
-			ustring(l->name().c_str()), TypeDesc::TypeColor, ATTR_ELEMENT_CORNER_BYTE);
+		for(b_mesh.vertex_colors.begin(l); l != b_mesh.vertex_colors.end(); ++l) {
+			if(!mesh->need_attribute(scene, ustring(l->name().c_str())))
+				continue;
 
-		BL::MeshColorLayer::data_iterator c;
-		uchar4 *cdata = attr->data_uchar4();
-		size_t i = 0;
+			Attribute *attr = mesh->subd_attributes.add(ustring(l->name().c_str()),
+			                                            TypeDesc::TypeColor,
+			                                            ATTR_ELEMENT_CORNER_BYTE);
 
-		for(l->data.begin(c); c != l->data.end(); ++c, ++i) {
-			int tri_a[3], tri_b[3];
-			face_split_tri_indices(nverts[i], face_flags[i], tri_a, tri_b);
+			BL::Mesh::polygons_iterator p;
+			uchar4 *cdata = attr->data_uchar4();
 
-			uchar4 colors[4];
-			colors[0] = color_float_to_byte(color_srgb_to_scene_linear(get_float3(c->color1())));
-			colors[1] = color_float_to_byte(color_srgb_to_scene_linear(get_float3(c->color2())));
-			colors[2] = color_float_to_byte(color_srgb_to_scene_linear(get_float3(c->color3())));
-			if(nverts[i] == 4) {
-				colors[3] = color_float_to_byte(color_srgb_to_scene_linear(get_float3(c->color4())));
+			for(b_mesh.polygons.begin(p); p != b_mesh.polygons.end(); ++p) {
+				int n = p->loop_total();
+				for(int i = 0; i < n; i++) {
+					float3 color = get_float3(l->data[p->loop_start() + i].color());
+					*(cdata++) = color_float_to_byte(color_srgb_to_scene_linear(color));
+				}
 			}
+		}
+	}
+	else {
+		BL::Mesh::tessface_vertex_colors_iterator l;
+		for(b_mesh.tessface_vertex_colors.begin(l); l != b_mesh.tessface_vertex_colors.end(); ++l) {
+			if(!mesh->need_attribute(scene, ustring(l->name().c_str())))
+				continue;
+
+			Attribute *attr = mesh->attributes.add(ustring(l->name().c_str()),
+			                                       TypeDesc::TypeColor,
+			                                       ATTR_ELEMENT_CORNER_BYTE);
+
+			BL::MeshColorLayer::data_iterator c;
+			uchar4 *cdata = attr->data_uchar4();
+			size_t i = 0;
+
+			for(l->data.begin(c); c != l->data.end(); ++c, ++i) {
+				int tri_a[3], tri_b[3];
+				face_split_tri_indices(nverts[i], face_flags[i], tri_a, tri_b);
+
+				uchar4 colors[4];
+				colors[0] = color_float_to_byte(color_srgb_to_scene_linear(get_float3(c->color1())));
+				colors[1] = color_float_to_byte(color_srgb_to_scene_linear(get_float3(c->color2())));
+				colors[2] = color_float_to_byte(color_srgb_to_scene_linear(get_float3(c->color3())));
+				if(nverts[i] == 4) {
+					colors[3] = color_float_to_byte(color_srgb_to_scene_linear(get_float3(c->color4())));
+				}
 
-			cdata[0] = colors[tri_a[0]];
-			cdata[1] = colors[tri_a[1]];
-			cdata[2] = colors[tri_a[2]];
+				cdata[0] = colors[tri_a[0]];
+				cdata[1] = colors[tri_a[1]];
+				cdata[2] = colors[tri_a[2]];
 
-			if(nverts[i] == 4) {
-				cdata[3] = colors[tri_b[0]];
-				cdata[4] = colors[tri_b[1]];
-				cdata[5] = colors[tri_b[2]];
-				cdata += 6;
+				if(nverts[i] == 4) {
+					cdata[3] = colors[tri_b[0]];
+					cdata[4] = colors[tri_b[1]];
+					cdata[5] = colors[tri_b[2]];
+					cdata += 6;
+				}
+				else
+					cdata += 3;
 			}
-			else
-				cdata += 3;
 		}
 	}
 }
@@ -382,9 +408,40 @@ static void attr_create_uv_map(Scene *scene,
                                Mesh *mesh,
                                BL::Mesh& b_mesh,
                                const vector<int>& nverts,
-                               const vector<int>& face_flags)
+                               const vector<int>& face_flags,
+                               bool subdivision)
 {
-	if(b_mesh.tessface_uv_textures.length() != 0) {
+	if(subdivision) {
+		BL::Mesh::uv_layers_iterator l;
+		int i = 0;
+
+		for(b_mesh.uv_layers.begin(l); l != b_mesh.uv_layers.end(); ++l, ++i) {
+			bool active_render = b_mesh.uv_textures[i].active_render();
+			AttributeStandard std = (active_render)? ATTR_STD_UV: ATTR_STD_NONE;
+			ustring name = ustring(l->name().c_str());
+
+			/* UV map */
+			if(mesh->need_attribute(scene, name) || mesh->need_attribute(scene, std)) {
+				Attribute *attr;
+
+				if(active_render)
+					attr = mesh->subd_attributes.add(std, name);
+				else
+					attr = mesh->subd_attributes.add(name, TypeDesc::TypePoint, ATTR_ELEMENT_CORNER);
+
+				BL::Mesh::polygons_iterator p;
+				float3 *fdata = attr->data_float3();
+
+				for(b_mesh.polygons.begin(p); p != b_mesh.polygons.end(); ++p) {
+					int n = p->loop_total();
+					for(int j = 0; j < n; j++) {
+						*(fdata++) = get_float3(l->data[p->loop_start() + j].uv());
+					}
+				}
+			}
+		}
+	}
+	else if(b_mesh.tessface_uv_textures.length() != 0) {
 		BL::Mesh::tessface_uv_textures_iterator l;
 
 		for(b_mesh.tessface_uv_textures.begin(l); l != b_mesh.tessface_uv_textures.end(); ++l) {
@@ -465,11 +522,13 @@ static void attr_create_uv_map(Scene *scene,
 /* Create vertex pointiness attributes. */
 static void attr_create_pointiness(Scene *scene,
                                    Mesh *mesh,
-                                   BL::Mesh& b_mesh)
+                                   BL::Mesh& b_mesh,
+                                   bool subdivision)
 {
 	if(mesh->need_attribute(scene, ATTR_STD_POINTINESS)) {
 		const int numverts = b_mesh.vertices.length();
-		Attribute *attr = mesh->attributes.add(ATTR_STD_POINTINESS);
+		AttributeSet& attributes = (subdivision)? mesh->subd_attributes: mesh->attributes;
+		Attribute *attr = attributes.add(ATTR_STD_POINTINESS);
 		float *data = attr->data_float();
 		int *counter = new int[numverts];
 		float *raw_data = new float[numverts];
@@ -532,30 +591,44 @@ static void attr_create_pointiness(Scene *scene,
 static void create_mesh(Scene *scene,
                         Mesh *mesh,
                         BL::Mesh& b_mesh,
-                        const vector<Shader*>& used_shaders)
+                        const vector<Shader*>& used_shaders,
+                        bool subdivision=false)
 {
 	/* count vertices and faces */
 	int numverts = b_mesh.vertices.length();
-	int numfaces = b_mesh.tessfaces.length();
+	int numfaces = (!subdivision) ? b_mesh.tessfaces.length() : b_mesh.polygons.length();
 	int numtris = 0;
+	int numcorners = 0;
+	int numngons = 0;
 	bool use_loop_normals = b_mesh.use_auto_smooth();
 
 	BL::Mesh::vertices_iterator v;
 	BL::Mesh::tessfaces_iterator f;
+	BL::Mesh::polygons_iterator p;
 
-	for(b_mesh.tessfaces.begin(f); f != b_mesh.tessfaces.end(); ++f) {
-		int4 vi = get_int4(f->vertices_raw());
-		numtris += (vi[3] == 0)? 1: 2;
+	if(!subdivision) {
+		for(b_mesh.tessfaces.begin(f); f != b_mesh.tessfaces.end(); ++f) {
+			int4 vi = get_int4(f->vertices_raw());
+			numtris += (vi[3] == 0)? 1: 2;
+		}
+	}
+	else {
+		for(b_mesh.polygons.begin(p); p != b_mesh.polygons.end(); ++p) {
+			numngons += (p->loop_total() == 4)? 0: 1;
+			numcorners += p->loop_total();
+		}
 	}
 
 	/* allocate memory */
 	mesh->reserve_mesh(numverts, numtris);
+	mesh->reserve_subd_faces(numfaces, numngons, numcorners);
 
 	/* create vertex coordinates and normals */
 	for(b_mesh.vertices.begin(v); v != b_mesh.vertices.end(); ++v)
 		mesh->add_vertex(get_float3(v->co()));
 
-	Attribute *attr_N = mesh->attributes.add(ATTR_STD_VERTEX_NORMAL);
+	AttributeSet& attributes = (subdivision)? mesh->subd_attributes: mesh->attributes;
+	Attribute *attr_N = attributes.add(ATTR_STD_VERTEX_NORMAL);
 	float3 *N = attr_N->data_float3();
 
 	for(b_mesh.vertices.begin(v); v != b_mesh.vertices.end(); ++v, ++N)
@@ -564,7 +637,7 @@ static void create_mesh(Scene *scene,
 
 	/* create generated coordinates from undeformed coordinates */
 	if(mesh->need_attribute(scene, ATTR_STD_GENERATED)) {
-		Attribute *attr = mesh->attributes.add(ATTR_STD_GENERATED);
+		Attribute *attr = attributes.add(ATTR_STD_GENERATED);
 
 		float3 loc, size;
 		mesh_texture_space(b_mesh, loc, size);
@@ -577,67 +650,103 @@ static void create_mesh(Scene *scene,
 	}
 
 	/* Create needed vertex attributes. */
-	attr_create_pointiness(scene, mesh, b_mesh);
+	attr_create_pointiness(scene, mesh, b_mesh, subdivision);
 
 	/* create faces */
 	vector<int> nverts(numfaces);
 	vector<int> face_flags(numfaces, FACE_FLAG_NONE);
 	int fi = 0;
 
-	for(b_mesh.tessfaces.begin(f); f != b_mesh.tessfaces.end(); ++f, ++fi) {
-		int4 vi = get_int4(f->vertices_raw());
-		int n = (vi[3] == 0)? 3: 4;
-		int shader = clamp(f->material_index(), 0, used_shaders.size()-1);
-		bool smooth = f->use_smooth() || use_loop_normals;
-
-		/* split vertices if normal is different
-		 *
-		 * note all vertex attributes must have been set here so we can split
-		 * and copy attributes in split_vertex without remapping later */
-		if(use_loop_normals) {
-			BL::Array<float, 12> loop_normals = f->split_normals();
-
-			for(int i = 0; i < n; i++) {
-				float3 loop_N = make_float3(loop_normals[i * 3], loop_normals[i * 3 + 1], loop_normals[i * 3 + 2]);
-
-				if(N[vi[i]] != loop_N) {
-					int new_vi = mesh->split_vertex(vi[i]);
-
-					/* set new normal and vertex index */
-					N = attr_N->data_float3();
-					N[new_vi] = loop_N;
-					vi[i] = new_vi;
+	if(!subdivision) {
+		for(b_mesh.tessfaces.begin(f); f != b_mesh.tessfaces.end(); ++f, ++fi) {
+			int4 vi = get_int4(f->vertices_raw());
+			int n = (vi[3] == 0)? 3: 4;
+			int shader = clamp(f->material_index(), 0, used_shaders.size()-1);
+			bool smooth = f->use_smooth() || use_loop_normals;
+
+			/* split vertices if normal is different
+			 *
+			 * note all vertex attributes must have been set here so we can split
+			 * and copy attributes in split_vertex without remapping later */
+			if(use_loop_normals) {
+				BL::Array<float, 12> loop_normals = f->split_normals();
+
+				for(int i = 0; i < n; i++) {
+					float3 loop_N = make_float3(loop_normals[i * 3], loop_normals[i * 3 + 1], loop_normals[i * 3 + 2]);
+
+					if(N[vi[i]] != loop_N) {
+						int new_vi = mesh->split_vertex(vi[i]);
+
+						/* set new normal and vertex index */
+						N = attr_N->data_float3();
+						N[new_vi] = loop_N;
+						vi[i] = new_vi;
+					}
 				}
 			}
-		}
 
-		/* create triangles */
-		if(n == 4) {
-			if(is_zero(cross(mesh->verts[vi[1]] - mesh->verts[vi[0]], mesh->verts[vi[2]] - mesh->verts[vi[0]])) ||
-			   is_zero(cross(mesh->verts[vi[2]] - mesh->verts[vi[0]], mesh->verts[vi[3]] - mesh->verts[vi[0]])))
-			{
-				// TODO(mai): order here is probably wrong
-				mesh->add_triangle(vi[0], vi[1], vi[3], shader, smooth, true);
-				mesh->add_triangle(vi[2], vi[3], vi[1], shader, smooth, true);
-				face_flags[fi] |= FACE_FLAG_DIVIDE_24;
+			/* create triangles */
+			if(n == 4) {
+				if(is_zero(cross(mesh->verts[vi[1]] - mesh->verts[vi[0]], mesh->verts[vi[2]] - mesh->verts[vi[0]])) ||
+				   is_zero(cross(mesh->verts[vi[2]] - mesh->verts[vi[0]], mesh->verts[vi[3]] - mesh->verts[vi[0]])))
+				{
+					mesh->add_triangle(vi[0], vi[1], vi[3], shader, smooth);
+					mesh->add_triangle(vi[2], vi[3], vi[1], shader, smooth);
+					face_flags[fi] |= FACE_FLAG_DIVIDE_24;
+				}
+				else {
+					mesh->add_triangle(vi[0], vi[1], vi[2], shader, smooth);
+					mesh->add_triangle(vi[0], vi[2], vi[3], shader, smooth);
+					face_flags[fi] |= FACE_FLAG_DIVIDE_13;
+				}
 			}
 			else {
-				mesh->add_triangle(vi[0], vi[1], vi[2], shader, smooth, true);
-				mesh->add_triangle(vi[0], vi[2], vi[3], shader, smooth, true);
-				face_flags[fi] |= FACE_FLAG_DIVIDE_13;
+				mesh->add_triangle(vi[0], vi[1], vi[2], shader, smooth);
 			}
+
+			nverts[fi] = n;
 		}
-		else
-			mesh->add_triangle(vi[0], vi[1], vi[2], shader, smooth, false);
+	}
+	else {
+		vector<int> vi;
+
+		for(b_mesh.polygons.begin(p); p != b_mesh.polygons.end(); ++p) {
+			int n = p->loop_total();
+			int shader = clamp(p->material_index(), 0, used_shaders.size()-1);
+			bool smooth = p->use_smooth() || use_loop_normals;
 
-		nverts[fi] = n;
+			vi.reserve(n);
+			for(int i = 0; i < n; i++) {
+				vi[i] = b_mesh.loops[p->loop_start() + i].vertex_index();
+
+				/* split vertices if normal is different
+				 *
+				 * note all vertex attributes must have been set here so we can split
+				 * and copy attributes in split_vertex without remapping later */
+				if(use_loop_normals) {
+					float3 loop_N = get_float3(b_mesh.loops[p->loop_start() + i].normal());
+
+					if(N[vi[i]] != loop_N) {
+						int new_vi = mesh->split_vertex(vi[i]);
+
+						/* set new normal and vertex index */
+						N = attr_N->data_float3();
+						N[new_vi] = loop_N;
+						vi[i] = new_vi;
+					}
+				}
+			}
+
+			/* create subd faces */
+			mesh->add_subd_face(&vi[0], n, shader, smooth);
+		}
 	}
 
 	/* Create all needed attributes.
 	 * The calculate functions will check whether they're needed or not.
 	 */
-	attr_create_vertex_color(scene, mesh, b_mesh, nverts, face_flags);
-	attr_create_uv_map(scene, mesh, b_mesh, nverts, face_flags);
+	attr_create_vertex_color(scene, mesh, b_mesh, nverts, face_flags, subdivision);
+	attr_create_uv_map(scene, mesh, b_mesh, nverts, face_flags, subdivision);
 
 	/* for volume objects, create a matrix to transform from object space to
 	 * mesh texture space. this does not work with deformations but that can
@@ -662,10 +771,9 @@ static void create_subd_mesh(Scene *scene,
                              float dicing_rate,
                              int max_subdivisions)
 {
-	Mesh basemesh;
-	create_mesh(scene, &basemesh, b_mesh, used_shaders);
+	create_mesh(scene, mesh, b_mesh, used_shaders, true);
 
-	SubdParams sdparams(mesh, 0, true, false);
+	SubdParams sdparams(mesh);
 	sdparams.dicing_rate = max(0.1f, RNA_float_get(cmesh, "dicing_rate") * dicing_rate);
 	sdparams.max_level = max_subdivisions;
 
@@ -675,7 +783,7 @@ static void create_subd_mesh(Scene *scene,
 
 	/* tesselate */
 	DiagSplit dsplit(sdparams);
-	basemesh.tessellate(&dsplit);
+	mesh->tessellate(&dsplit);
 }
 
 /* Sync */
@@ -817,20 +925,21 @@ Mesh *BlenderSync::sync_mesh(BL::Object& b_ob,
 			b_ob.update_from_editmode();
 
 		bool need_undeformed = mesh->need_attribute(scene, ATTR_STD_GENERATED);
-		BL::Mesh b_mesh = object_to_mesh(b_data, b_ob, b_scene, true, !preview, need_undeformed);
+		bool subdivision = experimental && cmesh.data && RNA_enum_get(&cmesh, "subdivision_type");
+		BL::Mesh b_mesh = object_to_mesh(b_data, b_ob, b_scene, true, !preview, need_undeformed, subdivision);
 
 		if(b_mesh) {
 			if(render_layer.use_surfaces && !hide_tris) {
-				if(cmesh.data && experimental && RNA_enum_get(&cmesh, "subdivision_type"))
+				if(subdivision)
 					create_subd_mesh(scene, mesh, b_ob, b_mesh, &cmesh, used_shaders,
 					                 dicing_rate, max_subdivisions);
 				else
-					create_mesh(scene, mesh, b_mesh, used_shaders);
+					create_mesh(scene, mesh, b_mesh, used_shaders, false);
 
 				create_mesh_volume_attributes(scene, b_ob, mesh, b_scene.frame_current());
 			}
 
-			if(render_layer.use_hair)
+			if(render_layer.use_hair && !subdivision)
 				sync_curves(mesh, b_mesh, b_ob, false);
 
 			if(can_free_caches) {
@@ -957,7 +1066,7 @@ void BlenderSync::sync_mesh_motion(BL::Object& b_ob,
 
 	if(ccl::BKE_object_is_deform_modified(b_ob, b_scene, preview)) {
 		/* get derived mesh */
-		b_mesh = object_to_mesh(b_data, b_ob, b_scene, true, !preview, false);
+		b_mesh = object_to_mesh(b_data, b_ob, b_scene, true, !preview, false, false);
 	}
 
 	if(!b_mesh) {
diff --git a/intern/cycles/blender/blender_util.h b/intern/cycles/blender/blender_util.h
index 188d23d0c59..34405c3ea1a 100644
--- a/intern/cycles/blender/blender_util.h
+++ b/intern/cycles/blender/blender_util.h
@@ -45,14 +45,17 @@ static inline BL::Mesh object_to_mesh(BL::BlendData& data,
                                       BL::Scene& scene,
                                       bool apply_modifiers,
                                       bool render,
-                                      bool calc_undeformed)
+                                      bool calc_undeformed,
+                                      bool subdivision)
 {
 	BL::Mesh me = data.meshes.new_from_object(scene, object, apply_modifiers, (render)? 2: 1, false, calc_undeformed);
 	if((bool)me) {
 		if(me.use_auto_smooth()) {
 			me.calc_normals_split();
 		}
-		me.calc_tessface(true);
+		if(!subdivision) {
+			me.calc_tessface(true);
+		}
 	}
 	return me;
 }
diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt
index bd3969b2889..685574eb394 100644
--- a/intern/cycles/kernel/CMakeLists.txt
+++ b/intern/cycles/kernel/CMakeLists.txt
@@ -161,6 +161,7 @@ set(SRC_GEOM_HEADERS
 	geom/geom_motion_triangle.h
 	geom/geom_object.h
 	geom/geom_primitive.h
+	geom/geom_subd_triangle.h
 	geom/geom_triangle.h
 	geom/geom_triangle_intersect.h
 	geom/geom_volume.h
diff --git a/intern/cycles/kernel/geom/geom.h b/intern/cycles/kernel/geom/geom.h
index d2c7edb11ea..493afdc4f62 100644
--- a/intern/cycles/kernel/geom/geom.h
+++ b/intern/cycles/kernel/geom/geom.h
@@ -18,6 +18,7 @@
 #include "geom_attribute.h"
 #include "geom_object.h"
 #include "geom_triangle.h"
+#include "geom_subd_triangle.h"
 #include "geom_triangle_intersect.h"
 #include "geom_motion_triangle.h"
 #include "geom_motion_curve.h"
diff --git a/intern/cycles/kernel/geom/geom_attribute.h b/intern/cycles/kernel/geom/geom_attribute.h
index c7364e9edac..5d78cf8f9fc 100644
--- a/intern/cycles/kernel/geom/geom_attribute.h
+++ b/intern/cycles/kernel/geom/geom_attribute.h
@@ -25,6 +25,24 @@ CCL_NAMESPACE_BEGIN
  * Lookup of attributes is different between OSL and SVM, as OSL is ustring
  * based while for SVM we use integer ids. */
 
+ccl_device_inline uint subd_triangle_patch(KernelGlobals *kg, const ShaderData *sd);
+
+ccl_device_inline uint attribute_primitive_type(KernelGlobals *kg, const ShaderData *sd)
+{
+#ifdef __HAIR__
+	if(ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE) {
+		return ATTR_PRIM_CURVE;
+	}
+	else
+#endif
+	if(subd_triangle_patch(kg, sd) != ~0) {
+		return ATTR_PRIM_SUBD;
+	}
+	else {
+		return ATTR_PRIM_TRIANGLE;
+	}
+}
+
 /* Find attribute based on ID */
 
 ccl_device_inline int find_attribute(KernelGlobals *kg, const ShaderData *sd, uint id, AttributeElement *elem)
@@ -34,9 +52,7 @@ ccl_device_inline int find_attribute(KernelGlobals *kg, const ShaderData *sd, ui
 
 	/* for SVM, find attribute by unique id */
 	uint attr_offset = ccl_fetch(sd, object)*kernel_data.bvh.attributes_map_stride;
-#ifdef __HAIR__
-	attr_offset = (ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE)? attr_offset + ATTR_PRIM_CURVE: attr_offset;
-#endif
+	attr_offset += attribute_primitive_type(kg, sd);
 	uint4 attr_map = kernel_tex_fetch(__attributes_map, attr_offset);
 	
 	while(attr_map.x != id) {
diff --git a/intern/cycles/kernel/geom/geom_primitive.h b/intern/cycles/kernel/geom/geom_primitive.h
index b1b1e919e00..44734d1b70d 100644
--- a/intern/cycles/kernel/geom/geom_primitive.h
+++ b/intern/cycles/kernel/geom/geom_primitive.h
@@ -26,7 +26,10 @@ CCL_NAMESPACE_BEGIN
 ccl_device float primitive_attribute_float(KernelGlobals *kg, const ShaderData *sd, AttributeElement elem, int offset, float *dx, float *dy)
 {
 	if(ccl_fetch(sd, type) & PRIMITIVE_ALL_TRIANGLE) {
-		return triangle_attribute_float(kg, sd, elem, offset, dx, dy);
+		if(subd_triangle_patch(kg, sd) == ~0)
+			return triangle_attribute_float(kg, sd, elem, offset, dx, dy);
+		else
+			return subd_triangle_attribute_float(kg, sd, elem, offset, dx, dy);
 	}
 #ifdef __HAIR__
 	else if(ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE) {
@@ -48,7 +51,10 @@ ccl_device float primitive_attribute_float(KernelGlobals *kg, const ShaderData *
 ccl_device float3 primitive_attribute_float3(KernelGlobals *kg, const ShaderData *sd, AttributeElement elem, int offset, float3 *dx, float3 *dy)
 {
 	if(ccl_fetch(sd, type) & PRIMITIVE_ALL_TRIANGLE) {
-		return triangle_attribute_float3(kg, sd, elem, offset, dx, dy);
+		if(subd_triangle_patch(kg, sd) == ~0)
+			return triangle_attribute_float3(kg, sd, elem, offset, dx, dy);
+		else
+			return subd_triangle_attribute_float3(kg, sd, elem, offset, dx, dy);
 	}
 #ifdef __HAIR__
 	else if(ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE) {
diff --git a/intern/cycles/kernel/geom/geom_subd_triangle.h b/intern/cycles/kernel/geom/geom_subd_triangle.h
new file mode 100644
index 00000000000..e4597aba56e
--- /dev/null
+++ b/intern/cycles/kernel/geom/geom_subd_triangle.h
@@ -0,0 +1,262 @@
+/*
+ * Copyright 2011-2016 Blender Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/* Functions for retrieving attributes on triangles produced from subdivision meshes */
+
+CCL_NAMESPACE_BEGIN
+
+/* Patch index for triangle, -1 if not subdivision triangle */
+
+ccl_device_inline uint subd_triangle_patch(KernelGlobals *kg, const ShaderData *sd)
+{
+	return kernel_tex_fetch(__tri_patch, ccl_fetch(sd, prim));
+}
+
+/* UV coords of triangle within patch */
+
+ccl_device_inline void subd_triangle_patch_uv(KernelGlobals *kg, const ShaderData *sd, float2 uv[3])
+{
+	uint4 tri_vindex = kernel_tex_fetch(__tri_vindex, ccl_fetch(sd, prim));
+
+	uv[0] = kernel_tex_fetch(__tri_patch_uv, tri_vindex.x);
+	uv[1] = kernel_tex_fetch(__tri_patch_uv, tri_vindex.y);
+	uv[2] = kernel_tex_fetch(__tri_patch_uv, tri_vindex.z);
+}
+
+/* Vertex indices of patch */
+
+ccl_device_inline uint4 subd_triangle_patch_indices(KernelGlobals *kg, int patch)
+{
+	uint4 indices;
+
+	indices.x = kernel_tex_fetch(__patches, patch+0);
+	indices.y = kernel_tex_fetch(__patches, patch+1);
+	indices.z = kernel_tex_fetch(__patches, patch+2);
+	indices.w = kernel_tex_fetch(__patches, patch+3);
+
+	return indices;
+}
+
+/* Originating face for patch */
+
+ccl_device_inline uint subd_triangle_patch_face(KernelGlobals *kg, int patch)
+{
+	return kernel_tex_fetch(__patches, patch+4);
+}
+
+/* Number of corners on originating face */
+
+ccl_device_inline uint subd_triangle_patch_num_corners(KernelGlobals *kg, int patch)
+{
+	return kernel_tex_fetch(__patches, patch+5) & 0xffff;
+}
+
+/* Indices of the four corners that are used by the patch */
+
+ccl_device_inline void subd_triangle_patch_corners(KernelGlobals *kg, int patch, int corners[4])
+{
+	uint4 data;
+
+	data.x = kernel_tex_fetch(__patches, patch+4);
+	data.y = kernel_tex_fetch(__patches, patch+5);
+	data.z = kernel_tex_fetch(__patches, patch+6);
+	data.w = kernel_tex_fetch(__patches, patch+7);
+
+	int num_corners = data.y & 0xffff;
+
+	if(num_corners == 4) {
+		/* quad */
+		corners[0] = data.z;
+		corners[1] = data.z+1;
+		corners[2] = data.z+2;
+		corners[3] = data.z+3;
+	}
+	else {
+		/* ngon */
+		int c = data.y >> 16;
+
+		corners[0] = data.z + c;
+		corners[1] = data.z + mod(c+1, num_corners);
+		corners[2] = data.w;
+		corners[3] = data.z + mod(c-1, num_corners);
+	}
+}
+
+/* Reading attributes on various subdivision triangle elements */
+
+ccl_device float subd_triangle_attribute_float(KernelGlobals *kg, const ShaderData *sd, AttributeElement elem, int offset, float *dx, float *dy)
+{
+	int patch = subd_triangle_patch(kg, sd);
+
+	if(elem == ATTR_ELEMENT_FACE) {
+		if(dx) *dx = 0.0f;
+		if(dy) *dy = 0.0f;
+
+		return kernel_tex_fetch(__attributes_float, offset + subd_triangle_patch_face(kg, patch));
+	}
+	else if(elem == ATTR_ELEMENT_VERTEX || elem == ATTR_ELEMENT_VERTEX_MOTION) {
+		float2 uv[3];
+		subd_triangle_patch_uv(kg, sd, uv);
+		uint4 v = subd_triangle_patch_indices(kg, patch);
+
+		float a, b, c;
+
+		float f0 = kernel_tex_fetch(__attributes_float, offset + v.x);
+		float f1 = kernel_tex_fetch(__attributes_float, offset + v.y);
+		float f2 = kernel_tex_fetch(__attributes_float, offset + v.z);
+		float f3 = kernel_tex_fetch(__attributes_float, offset + v.w);
+
+		if(subd_triangle_patch_num_corners(kg, patch) != 4) {
+			f1 = (f1+f0)*0.5f;
+			f3 = (f3+f0)*0.5f;
+		}
+
+		a = mix(mix(f0, f1, uv[0].x), mix(f3, f2, uv[0].x), uv[0].y);
+		b = mix(mix(f0, f1, uv[1].x), mix(f3, f2, uv[1].x), uv[1].y);
+		c = mix(mix(f0, f1, uv[2].x), mix(f3, f2, uv[2].x), uv[2].y);
+
+#ifdef __RAY_DIFFERENTIALS__
+		if(dx) *dx = ccl_fetch(sd, du).dx*a + ccl_fetch(sd, dv).dx*b - (ccl_fetch(sd, du).dx + ccl_fetch(sd, dv).dx)*c;
+		if(dy) *dy = ccl_fetch(sd, du).dy*a + ccl_fetch(sd, dv).dy*b - (ccl_fetch(sd, du).dy + ccl_fetch(sd, dv).dy)*c;
+#endif
+
+		return ccl_fetch(sd, u)*a + ccl_fetch(sd, v)*b + (1.0f - ccl_fetch(sd, u) - ccl_fetch(sd, v))*c;
+	}
+	else if(elem == ATTR_ELEMENT_CORNER) {
+		int corners[4];
+		subd_triangle_patch_corners(kg, patch, corners);
+
+		float2 uv[3];
+		subd_triangle_patch_uv(kg, sd, uv);
+
+		float a, b, c;
+
+		float f0 = kernel_tex_fetch(__attributes_float, corners[0] + offset);
+		float f1 = kernel_tex_fetch(__attributes_float, corners[1] + offset);
+		float f2 = kernel_tex_fetch(__attributes_float, corners[2] + offset);
+		float f3 = kernel_tex_fetch(__attributes_float, corners[3] + offset);
+
+		if(subd_triangle_patch_num_corners(kg, patch) != 4) {
+			f1 = (f1+f0)*0.5f;
+			f3 = (f3+f0)*0.5f;
+		}
+
+		a = mix(mix(f0, f1, uv[0].x), mix(f3, f2, uv[0].x), uv[0].y);
+		b = mix(mix(f0, f1, uv[1].x), mix(f3, f2, uv[1].x), uv[1].y);
+		c = mix(mix(f0, f1, uv[2].x), mix(f3, f2, uv[2].x), uv[2].y);
+
+#ifdef __RAY_DIFFERENTIALS__
+		if(dx) *dx = ccl_fetch(sd, du).dx*a + ccl_fetch(sd, dv).dx*b - (ccl_fetch(sd, du).dx + ccl_fetch(sd, dv).dx)*c;
+		if(dy) *dy = ccl_fetch(sd, du).dy*a + ccl_fetch(sd, dv).dy*b - (ccl_fetch(sd, du).dy + ccl_fetch(sd, dv).dy)*c;
+#endif
+
+		return ccl_fetch(sd, u)*a + ccl_fetch(sd, v)*b + (1.0f - ccl_fetch(sd, u) - ccl_fetch(sd, v))*c;
+	}
+	else {
+		if(dx) *dx = 0.0f;
+		if(dy) *dy = 0.0f;
+
+		return 0.0f;
+	}
+}
+
+ccl_device float3 subd_triangle_attribute_float3(KernelGlobals *kg, const ShaderData *sd, AttributeElement elem, int offset, float3 *dx, float3 *dy)
+{
+	int patch = subd_triangle_patch(kg, sd);
+
+	if(elem == ATTR_ELEMENT_FACE) {
+		if(dx) *dx = make_float3(0.0f, 0.0f, 0.0f);
+		if(dy) *dy = make_float3(0.0f, 0.0f, 0.0f);
+
+		return float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + subd_triangle_patch_face(kg, patch)));
+	}
+	else if(elem == ATTR_ELEMENT_VERTEX || elem == ATTR_ELEMENT_VERTEX_MOTION) {
+		float2 uv[3];
+		subd_triangle_patch_uv(kg, sd, uv);
+		uint4 v = subd_triangle_patch_indices(kg, patch);
+
+		float3 a, b, c;
+
+		float3 f0 = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + v.x));
+		float3 f1 = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + v.y));
+		float3 f2 = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + v.z));
+		float3 f3 = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + v.w));
+
+		if(subd_triangle_patch_num_corners(kg, patch) != 4) {
+			f1 = (f1+f0)*0.5f;
+			f3 = (f3+f0)*0.5f;
+		}
+
+		a = mix(mix(f0, f1, uv[0].x), mix(f3, f2, uv[0].x), uv[0].y);
+		b = mix(mix(f0, f1, uv[1].x), mix(f3, f2, uv[1].x), uv[1].y);
+		c = mix(mix(f0, f1, uv[2].x), mix(f3, f2, uv[2].x), uv[2].y);
+
+#ifdef __RAY_DIFFERENTIALS__
+		if(dx) *dx = ccl_fetch(sd, du).dx*a + ccl_fetch(sd, dv).dx*b - (ccl_fetch(sd, du).dx + ccl_fetch(sd, dv).dx)*c;
+		if(dy) *dy = ccl_fetch(sd, du).dy*a + ccl_fetch(sd, dv).dy*b - (ccl_fetch(sd, du).dy + ccl_fetch(sd, dv).dy)*c;
+#endif
+
+		return ccl_fetch(sd, u)*a + ccl_fetch(sd, v)*b + (1.0f - ccl_fetch(sd, u) - ccl_fetch(sd, v))*c;
+	}
+	else if(elem == ATTR_ELEMENT_CORNER || elem == ATTR_ELEMENT_CORNER_BYTE) {
+		int corners[4];
+		subd_triangle_patch_corners(kg, patch, corners);
+
+		float2 uv[3];
+		subd_triangle_patch_uv(kg, sd, uv);
+
+		float3 a, b, c;
+		float3 f0, f1, f2, f3;
+
+		if(elem == ATTR_ELEMENT_CORNER) {
+			f0 = float4_to_float3(kernel_tex_fetch(__attributes_float3, corners[0] + offset));
+			f1 = float4_to_float3(kernel_tex_fetch(__attributes_float3, corners[1] + offset));
+			f2 = float4_to_float3(kernel_tex_fetch(__attributes_float3, corners[2] + offset));
+			f3 = float4_to_float3(kernel_tex_fetch(__attributes_float3, corners[3] + offset));
+		}
+		else {
+			f0 = color_byte_to_float(kernel_tex_fetch(__attributes_uchar4, corners[0] + offset));
+			f1 = color_byte_to_float(kernel_tex_fetch(__attributes_uchar4, corners[1] + offset));
+			f2 = color_byte_to_float(kernel_tex_fetch(__attributes_uchar4, corners[2] + offset));
+			f3 = color_byte_to_float(kernel_tex_fetch(__attributes_uchar4, corners[3] + offset));
+		}
+
+		if(subd_triangle_patch_num_corners(kg, patch) != 4) {
+			f1 = (f1+f0)*0.5f;
+			f3 = (f3+f0)*0.5f;
+		}
+
+		a = mix(mix(f0, f1, uv[0].x), mix(f3, f2, uv[0].x), uv[0].y);
+		b = mix(mix(f0, f1, uv[1].x), mix(f3, f2, uv[1].x), uv[1].y);
+		c = mix(mix(f0, f1, uv[2].x), mix(f3, f2, uv[2].x), uv[2].y);
+
+#ifdef __RAY_DIFFERENTIALS__
+		if(dx) *dx = ccl_fetch(sd, du).dx*a + ccl_fetch(sd, dv).dx*b - (ccl_fetch(sd, du).dx + ccl_fetch(sd, dv).dx)*c;
+		if(dy) *dy = ccl_fetch(sd, du).dy*a + ccl_fetch(sd, dv).dy*b - (ccl_fetch(sd, du).dy + ccl_fetch(sd, dv).dy)*c;
+#endif
+
+		return ccl_fetch(sd, u)*a + ccl_fetch(sd, v)*b + (1.0f - ccl_fetch(sd, u) - ccl_fetch(sd, v))*c;
+	}
+	else {
+		if(dx) *dx = make_float3(0.0f, 0.0f, 0.0f);
+		if(dy) *dy = make_float3(0.0f, 0.0f, 0.0f);
+
+		return make_float3(0.0f, 0.0f, 0.0f);
+	}
+}
+
+CCL_NAMESPACE_END
+
diff --git a/intern/cycles/kernel/kernel_textures.h b/intern/cycles/kernel/kernel_textures.h
index 5ba262c1044..7d6fec02331 100644
--- a/intern/cycles/kernel/kernel_textures.h
+++ b/intern/cycles/kernel/kernel_textures.h
@@ -41,11 +41,16 @@ KERNEL_TEX(float4, texture_float4, __objects_vector)
 KERNEL_TEX(uint, texture_uint, __tri_shader)
 KERNEL_TEX(float4, texture_float4, __tri_vnormal)
 KERNEL_TEX(uint4, texture_uint4, __tri_vindex)
+KERNEL_TEX(uint, texture_uint, __tri_patch)
+KERNEL_TEX(float2, texture_float2, __tri_patch_uv)
 
 /* curves */
 KERNEL_TEX(float4, texture_float4, __curves)
 KERNEL_TEX(float4, texture_float4, __curve_keys)
 
+/* patches */
+KERNEL_TEX(uint, texture_uint, __patches)
+
 /* attributes */
 KERNEL_TEX(uint4, texture_uint4, __attributes_map)
 KERNEL_TEX(float, texture_float, __attributes_float)
@@ -173,9 +178,6 @@ KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_086)
 KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_087)
 KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_088)
 KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_089)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_090)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_091)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_092)
 
 #  else
 /* bindless textures */
diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index a9be2ae717a..b7021dbf7f1 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -573,8 +573,13 @@ typedef enum PrimitiveType {
 
 /* Attributes */
 
-#define ATTR_PRIM_TYPES		2
-#define ATTR_PRIM_CURVE		1
+typedef enum AttributePrimitive {
+	ATTR_PRIM_TRIANGLE = 0,
+	ATTR_PRIM_CURVE,
+	ATTR_PRIM_SUBD,
+
+	ATTR_PRIM_TYPES
+} AttributePrimitive;
 
 typedef enum AttributeElement {
 	ATTR_ELEMENT_NONE,
diff --git a/intern/cycles/kernel/osl/osl_services.cpp b/intern/cycles/kernel/osl/osl_services.cpp
index 2bb2be5e6b3..caae24405f1 100644
--- a/intern/cycles/kernel/osl/osl_services.cpp
+++ b/intern/cycles/kernel/osl/osl_services.cpp
@@ -787,7 +787,7 @@ bool OSLRenderServices::get_attribute(ShaderData *sd, bool derivatives, ustring
                                       TypeDesc type, ustring name, void *val)
 {
 	KernelGlobals *kg = sd->osl_globals;
-	bool is_curve;
+	int prim_type = 0;
 	int object;
 
 	/* lookup of attribute on another object */
@@ -798,18 +798,17 @@ bool OSLRenderServices::get_attribute(ShaderData *sd, bool derivatives, ustring
 			return false;
 
 		object = it->second;
-		is_curve = false;
 	}
 	else {
 		object = sd->object;
-		is_curve = (sd->type & PRIMITIVE_ALL_CURVE) != 0;
+		prim_type = attribute_primitive_type(kg, sd);
 
 		if(object == OBJECT_NONE)
 			return get_background_attribute(kg, sd, name, type, derivatives, val);
 	}
 
 	/* find attribute on object */
-	object = object*ATTR_PRIM_TYPES + (is_curve == true);
+	object = object*ATTR_PRIM_TYPES + prim_type;
 	OSLGlobals::AttributeMap& attribute_map = kg->osl->attribute_map[object];
 	OSLGlobals::AttributeMap::iterator it = attribute_map.find(name);
 
diff --git a/intern/cycles/kernel/svm/svm_attribute.h b/intern/cycles/kernel/svm/svm_attribute.h
index 63bbb27d873..6c557684099 100644
--- a/intern/cycles/kernel/svm/svm_attribute.h
+++ b/intern/cycles/kernel/svm/svm_attribute.h
@@ -28,9 +28,7 @@ ccl_device void svm_node_attr_init(KernelGlobals *kg, ShaderData *sd,
 		/* find attribute by unique id */
 		uint id = node.y;
 		uint attr_offset = ccl_fetch(sd, object)*kernel_data.bvh.attributes_map_stride;
-#ifdef __HAIR__
-		attr_offset = (ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE)? attr_offset + ATTR_PRIM_CURVE: attr_offset;
-#endif
+		attr_offset += attribute_primitive_type(kg, sd);
 		uint4 attr_map = kernel_tex_fetch(__attributes_map, attr_offset);
 		
 		while(attr_map.x != id) {
diff --git a/intern/cycles/kernel/svm/svm_image.h b/intern/cycles/kernel/svm/svm_image.h
index 44732734c31..b6b90dfff81 100644
--- a/intern/cycles/kernel/svm/svm_image.h
+++ b/intern/cycles/kernel/svm/svm_image.h
@@ -271,9 +271,6 @@ ccl_device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y,
 		case 87: r = kernel_tex_image_interp(__tex_image_byte4_087, x, y); break;
 		case 88: r = kernel_tex_image_interp(__tex_image_byte4_088, x, y); break;
 		case 89: r = kernel_tex_image_interp(__tex_image_byte4_089, x, y); break;
-		case 90: r = kernel_tex_image_interp(__tex_image_byte4_090, x, y); break;
-		case 91: r = kernel_tex_image_interp(__tex_image_byte4_091, x, y); break;
-		case 92: r = kernel_tex_image_interp(__tex_image_byte4_092, x, y); break;
 		default:
 			kernel_assert(0);
 			return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
diff --git a/intern/cycles/render/CMakeLists.txt b/intern/cycles/render/CMakeLists.txt
index a632ddc0598..8eaa9de3874 100644
--- a/intern/cycles/render/CMakeLists.txt
+++ b/intern/cycles/render/CMakeLists.txt
@@ -30,6 +30,7 @@ set(SRC
 	light.cpp
 	mesh.cpp
 	mesh_displace.cpp
+	mesh_subdivision.cpp
 	nodes.cpp
 	object.cpp
 	osl.cpp
diff --git a/intern/cycles/render/attribute.cpp b/intern/cycles/render/attribute.cpp
index 71a3cba6811..e8ff81fe08e 100644
--- a/intern/cycles/render/attribute.cpp
+++ b/intern/cycles/render/attribute.cpp
@@ -51,13 +51,13 @@ void Attribute::set(ustring name_, TypeDesc type_, AttributeElement element_)
 		type == TypeDesc::TypeNormal || type == TypeDesc::TypeMatrix);
 }
 
-void Attribute::resize(int numverts, int numtris, int numsteps, int numcurves, int numkeys, bool reserve_only)
+void Attribute::resize(Mesh *mesh, AttributePrimitive prim, bool reserve_only)
 {
 	if(reserve_only) {
-		buffer.reserve(buffer_size(numverts, numtris, numsteps, numcurves, numkeys));
+		buffer.reserve(buffer_size(mesh, prim));
 	}
 	else {
-		buffer.resize(buffer_size(numverts, numtris, numsteps, numcurves, numkeys), 0);
+		buffer.resize(buffer_size(mesh, prim), 0);
 	}
 }
 
@@ -118,6 +118,8 @@ size_t Attribute::data_sizeof() const
 {
 	if(element == ATTR_ELEMENT_VOXEL)
 		return sizeof(VoxelAttribute);
+	else if(element == ATTR_ELEMENT_CORNER_BYTE)
+		return sizeof(uchar4);
 	else if(type == TypeDesc::TypeFloat)
 		return sizeof(float);
 	else if(type == TypeDesc::TypeMatrix)
@@ -126,10 +128,10 @@ size_t Attribute::data_sizeof() const
 		return sizeof(float3);
 }
 
-size_t Attribute::element_size(int numverts, int numtris, int numsteps, int numcurves, int numkeys) const
+size_t Attribute::element_size(Mesh *mesh, AttributePrimitive prim) const
 {
 	size_t size;
-	
+
 	switch(element) {
 		case ATTR_ELEMENT_OBJECT:
 		case ATTR_ELEMENT_MESH:
@@ -137,38 +139,54 @@ size_t Attribute::element_size(int numverts, int numtris, int numsteps, int numc
 			size = 1;
 			break;
 		case ATTR_ELEMENT_VERTEX:
-			size = numverts;
+			size = mesh->verts.size() + mesh->num_ngons;
+			if(prim == ATTR_PRIM_SUBD) {
+				size -= mesh->num_subd_verts;
+			}
 			break;
 		case ATTR_ELEMENT_VERTEX_MOTION:
-			size = numverts * (numsteps - 1);
+			size = (mesh->verts.size() + mesh->num_ngons) * (mesh->motion_steps - 1);
+			if(prim == ATTR_PRIM_SUBD) {
+				size -= mesh->num_subd_verts * (mesh->motion_steps - 1);
+			}
 			break;
 		case ATTR_ELEMENT_FACE:
-			size = numtris;
+			if(prim == ATTR_PRIM_TRIANGLE) {
+				size = mesh->num_triangles();
+			}
+			else {
+				size = mesh->subd_faces.size() + mesh->num_ngons;
+			}
 			break;
 		case ATTR_ELEMENT_CORNER:
 		case ATTR_ELEMENT_CORNER_BYTE:
-			size = numtris*3;
+			if(prim == ATTR_PRIM_TRIANGLE) {
+				size = mesh->num_triangles()*3;
+			}
+			else {
+				size = mesh->subd_face_corners.size() + mesh->num_ngons;
+			}
 			break;
 		case ATTR_ELEMENT_CURVE:
-			size = numcurves;
+			size = mesh->num_curves();
 			break;
 		case ATTR_ELEMENT_CURVE_KEY:
-			size = numkeys;
+			size = mesh->curve_keys.size();
 			break;
 		case ATTR_ELEMENT_CURVE_KEY_MOTION:
-			size = numkeys * (numsteps - 1);
+			size = mesh->curve_keys.size() * (mesh->motion_steps - 1);
 			break;
 		default:
 			size = 0;
 			break;
 	}
-	
+
 	return size;
 }
 
-size_t Attribute::buffer_size(int numverts, int numtris, int numsteps, int numcurves, int numkeys) const
+size_t Attribute::buffer_size(Mesh *mesh, AttributePrimitive prim) const
 {
-	return element_size(numverts, numtris, numsteps, numcurves, numkeys)*data_sizeof();
+	return element_size(mesh, prim)*data_sizeof();
 }
 
 bool Attribute::same_storage(TypeDesc a, TypeDesc b)
@@ -188,6 +206,29 @@ bool Attribute::same_storage(TypeDesc a, TypeDesc b)
 	return false;
 }
 
+void Attribute::zero_data(void* dst)
+{
+	memset(dst, 0, data_sizeof());
+}
+
+void Attribute::add_with_weight(void* dst, void* src, float weight)
+{
+	if(element == ATTR_ELEMENT_CORNER_BYTE) {
+		for(int i = 0; i < 4; i++) {
+			((uchar*)dst)[i] += uchar(((uchar*)src)[i] * weight);
+		}
+	}
+	else if(same_storage(type, TypeDesc::TypeFloat)) {
+		*((float*)dst) += *((float*)src) * weight;
+	}
+	else if(same_storage(type, TypeDesc::TypeVector)) {
+		*((float4*)dst) += *((float4*)src) * weight;
+	}
+	else {
+		assert(!"not implemented for this type");
+	}
+}
+
 const char *Attribute::standard_name(AttributeStandard std)
 {
 	switch(std) {
@@ -257,6 +298,7 @@ AttributeSet::AttributeSet()
 {
 	triangle_mesh = NULL;
 	curve_mesh = NULL;
+	subd_mesh = NULL;
 }
 
 AttributeSet::~AttributeSet()
@@ -291,10 +333,12 @@ Attribute *AttributeSet::add(ustring name, TypeDesc type, AttributeElement eleme
 
 	/* this is weak .. */
 	if(triangle_mesh)
-		attr->resize(triangle_mesh->verts.size(), triangle_mesh->num_triangles(), triangle_mesh->motion_steps, 0, 0, false);
+		attr->resize(triangle_mesh, ATTR_PRIM_TRIANGLE, false);
 	if(curve_mesh)
-		attr->resize(0, 0, curve_mesh->motion_steps, curve_mesh->num_curves(), curve_mesh->curve_keys.size(), false);
-	
+		attr->resize(curve_mesh, ATTR_PRIM_CURVE, false);
+	if(subd_mesh)
+		attr->resize(subd_mesh, ATTR_PRIM_SUBD, false);
+
 	return attr;
 }
 
@@ -330,7 +374,7 @@ Attribute *AttributeSet::add(AttributeStandard std, ustring name)
 	if(name == ustring())
 		name = Attribute::standard_name(std);
 
-	if(triangle_mesh) {
+	if(triangle_mesh || subd_mesh) {
 		switch(std) {
 			case ATTR_STD_VERTEX_NORMAL:
 				attr = add(name, TypeDesc::TypeNormal, ATTR_ELEMENT_VERTEX);
@@ -452,9 +496,11 @@ void AttributeSet::resize(bool reserve_only)
 {
 	foreach(Attribute& attr, attributes) {
 		if(triangle_mesh)
-			attr.resize(triangle_mesh->verts.size(), triangle_mesh->num_triangles(), triangle_mesh->motion_steps, 0, 0, reserve_only);
+			attr.resize(triangle_mesh, ATTR_PRIM_TRIANGLE, reserve_only);
 		if(curve_mesh)
-			attr.resize(0, 0, 0, curve_mesh->num_curves(), curve_mesh->curve_keys.size(), reserve_only);
+			attr.resize(curve_mesh, ATTR_PRIM_CURVE, reserve_only);
+		if(subd_mesh)
+			attr.resize(subd_mesh, ATTR_PRIM_SUBD, reserve_only);
 	}
 }
 
@@ -477,6 +523,10 @@ AttributeRequest::AttributeRequest(ustring name_)
 	curve_type = TypeDesc::TypeFloat;
 	curve_element = ATTR_ELEMENT_NONE;
 	curve_offset = 0;
+
+	subd_type = TypeDesc::TypeFloat;
+	subd_element = ATTR_ELEMENT_NONE;
+	subd_offset = 0;
 }
 
 AttributeRequest::AttributeRequest(AttributeStandard std_)
@@ -491,6 +541,10 @@ AttributeRequest::AttributeRequest(AttributeStandard std_)
 	curve_type = TypeDesc::TypeFloat;
 	curve_element = ATTR_ELEMENT_NONE;
 	curve_offset = 0;
+
+	subd_type = TypeDesc::TypeFloat;
+	subd_element = ATTR_ELEMENT_NONE;
+	subd_offset = 0;
 }
 
 /* AttributeRequestSet */
diff --git a/intern/cycles/render/attribute.h b/intern/cycles/render/attribute.h
index 41b3626afd3..e51bdf28d66 100644
--- a/intern/cycles/render/attribute.h
+++ b/intern/cycles/render/attribute.h
@@ -58,11 +58,11 @@ public:
 	Attribute() {}
 	~Attribute();
 	void set(ustring name, TypeDesc type, AttributeElement element);
-	void resize(int numverts, int numfaces, int numsteps, int numcurves, int numkeys, bool reserve_only);
+	void resize(Mesh *mesh, AttributePrimitive prim, bool reserve_only);
 
 	size_t data_sizeof() const;
-	size_t element_size(int numverts, int numfaces, int numsteps, int numcurves, int numkeys) const;
-	size_t buffer_size(int numverts, int numfaces, int numsteps, int numcurves, int numkeys) const;
+	size_t element_size(Mesh *mesh, AttributePrimitive prim) const;
+	size_t buffer_size(Mesh *mesh, AttributePrimitive prim) const;
 
 	char *data() { return (buffer.size())? &buffer[0]: NULL; };
 	float3 *data_float3() { return (float3*)data(); }
@@ -79,6 +79,9 @@ public:
 	const Transform *data_transform() const { return (const Transform*)data(); }
 	const VoxelAttribute *data_voxel() const { return (const VoxelAttribute*)data(); }
 
+	void zero_data(void* dst);
+	void add_with_weight(void* dst, void* src, float weight);
+
 	void add(const float& f);
 	void add(const float3& f);
 	void add(const uchar4& f);
@@ -99,6 +102,7 @@ class AttributeSet {
 public:
 	Mesh *triangle_mesh;
 	Mesh *curve_mesh;
+	Mesh *subd_mesh;
 	list<Attribute> attributes;
 
 	AttributeSet();
@@ -130,9 +134,9 @@ public:
 	AttributeStandard std;
 
 	/* temporary variables used by MeshManager */
-	TypeDesc triangle_type, curve_type;
-	AttributeElement triangle_element, curve_element;
-	int triangle_offset, curve_offset;
+	TypeDesc triangle_type, curve_type, subd_type;
+	AttributeElement triangle_element, curve_element, subd_element;
+	int triangle_offset, curve_offset, subd_offset;
 
 	explicit AttributeRequest(ustring name_);
 	explicit AttributeRequest(AttributeStandard std);
diff --git a/intern/cycles/render/mesh.cpp b/intern/cycles/render/mesh.cpp
index 8b0ed9f77b2..8d7c8fa9adb 100644
--- a/intern/cycles/render/mesh.cpp
+++ b/intern/cycles/render/mesh.cpp
@@ -35,9 +35,6 @@
 #include "util_progress.h"
 #include "util_set.h"
 
-#include "subd_split.h"
-#include "subd_patch.h"
-
 CCL_NAMESPACE_BEGIN
 
 /* Triangle */
@@ -104,6 +101,18 @@ void Mesh::Curve::bounds_grow(const int k,
 	bounds.grow(upper, mr);
 }
 
+/* SubdFace */
+
+float3 Mesh::SubdFace::normal(const Mesh *mesh) const
+{
+	float3 v0 = mesh->verts[mesh->subd_face_corners[start_corner+0]];
+	float3 v1 = mesh->verts[mesh->subd_face_corners[start_corner+1]];
+	float3 v2 = mesh->verts[mesh->subd_face_corners[start_corner+2]];
+
+	return safe_normalize(cross(v1 - v0, v2 - v0));
+}
+
+
 /* Mesh */
 
 NODE_DEFINE(Mesh)
@@ -150,13 +159,22 @@ Mesh::Mesh()
 	curve_offset = 0;
 	curvekey_offset = 0;
 
+	patch_offset = 0;
+	face_offset = 0;
+	corner_offset = 0;
+
+	num_subd_verts = 0;
+
 	attributes.triangle_mesh = this;
 	curve_attributes.curve_mesh = this;
+	subd_attributes.subd_mesh = this;
 
 	geometry_flags = GEOMETRY_NONE;
 
 	has_volume = false;
 	has_surface_bssrdf = false;
+
+	num_ngons = 0;
 }
 
 Mesh::~Mesh()
@@ -171,7 +189,10 @@ void Mesh::resize_mesh(int numverts, int numtris)
 	shader.resize(numtris);
 	smooth.resize(numtris);
 
-	forms_quad.resize(numtris);
+	if(subd_faces.size()) {
+		triangle_patch.resize(numtris);
+		vert_patch_uv.resize(numverts);
+	}
 
 	attributes.resize();
 }
@@ -184,7 +205,10 @@ void Mesh::reserve_mesh(int numverts, int numtris)
 	shader.reserve(numtris);
 	smooth.reserve(numtris);
 
-	forms_quad.reserve(numtris);
+	if(subd_faces.size()) {
+		triangle_patch.reserve(numtris);
+		vert_patch_uv.reserve(numverts);
+	}
 
 	attributes.resize(true);
 }
@@ -209,6 +233,24 @@ void Mesh::reserve_curves(int numcurves, int numkeys)
 	curve_attributes.resize(true);
 }
 
+void Mesh::resize_subd_faces(int numfaces, int num_ngons_, int numcorners)
+{
+	subd_faces.resize(numfaces);
+	subd_face_corners.resize(numcorners);
+	num_ngons = num_ngons_;
+
+	subd_attributes.resize();
+}
+
+void Mesh::reserve_subd_faces(int numfaces, int num_ngons_, int numcorners)
+{
+	subd_faces.reserve(numfaces);
+	subd_face_corners.reserve(numcorners);
+	num_ngons = num_ngons_;
+
+	subd_attributes.resize(true);
+}
+
 void Mesh::clear()
 {
 	/* clear all verts and triangles */
@@ -217,15 +259,22 @@ void Mesh::clear()
 	shader.clear();
 	smooth.clear();
 
-	forms_quad.clear();
+	triangle_patch.clear();
+	vert_patch_uv.clear();
 
 	curve_keys.clear();
 	curve_radius.clear();
 	curve_first_key.clear();
 	curve_shader.clear();
 
+	subd_faces.clear();
+	subd_face_corners.clear();
+
+	num_subd_verts = 0;
+
 	attributes.clear();
 	curve_attributes.clear();
+	subd_attributes.clear();
 	used_shaders.clear();
 
 	transform_applied = false;
@@ -247,27 +296,46 @@ int Mesh::split_vertex(int vertex)
 		}
 	}
 
+	foreach(Attribute& attr, subd_attributes.attributes) {
+		if(attr.element == ATTR_ELEMENT_VERTEX) {
+			vector<char> tmp(attr.data_sizeof());
+			memcpy(&tmp[0], attr.data() + tmp.size()*vertex, tmp.size());
+			attr.add(&tmp[0]);
+		}
+	}
+
 	return verts.size() - 1;
 }
 
 void Mesh::add_vertex(float3 P)
 {
 	verts.push_back_reserved(P);
+
+	if(subd_faces.size()) {
+		vert_patch_uv.push_back_reserved(make_float2(0.0f, 0.0f));
+	}
 }
 
 void Mesh::add_vertex_slow(float3 P)
 {
 	verts.push_back_slow(P);
+
+	if(subd_faces.size()) {
+		vert_patch_uv.push_back_slow(make_float2(0.0f, 0.0f));
+	}
 }
 
-void Mesh::add_triangle(int v0, int v1, int v2, int shader_, bool smooth_, bool forms_quad_)
+void Mesh::add_triangle(int v0, int v1, int v2, int shader_, bool smooth_)
 {
 	triangles.push_back_reserved(v0);
 	triangles.push_back_reserved(v1);
 	triangles.push_back_reserved(v2);
 	shader.push_back_reserved(shader_);
 	smooth.push_back_reserved(smooth_);
-	forms_quad.push_back_reserved(forms_quad_);
+
+	if(subd_faces.size()) {
+		triangle_patch.push_back_reserved(-1);
+	}
 }
 
 void Mesh::add_curve_key(float3 co, float radius)
@@ -282,6 +350,25 @@ void Mesh::add_curve(int first_key, int shader)
 	curve_shader.push_back_reserved(shader);
 }
 
+void Mesh::add_subd_face(int* corners, int num_corners, int shader_, bool smooth_)
+{
+	size_t start_corner = subd_face_corners.size();
+
+	for(int i = 0; i < num_corners; i++) {
+		subd_face_corners.push_back_reserved(corners[i]);
+	}
+
+	int ptex_offset = 0;
+
+	if(subd_faces.size()) {
+		SubdFace& s = subd_faces[subd_faces.size()-1];
+		ptex_offset = s.ptex_offset + s.num_ptex_faces();
+	}
+
+	SubdFace face = {start_corner, num_corners, shader_, smooth_, ptex_offset};
+	subd_faces.push_back_reserved(face);
+}
+
 void Mesh::compute_bounds()
 {
 	BoundBox bnds = BoundBox::empty;
@@ -505,10 +592,23 @@ void Mesh::pack_normals(Scene *scene, uint *tri_shader, float4 *vnormal)
 
 void Mesh::pack_verts(const vector<uint>& tri_prim_index,
                       uint4 *tri_vindex,
+                      uint *tri_patch,
+                      float2 *tri_patch_uv,
                       size_t vert_offset,
                       size_t tri_offset)
 {
-	const size_t triangles_size = num_triangles();
+	size_t verts_size = verts.size();
+
+	if(verts_size && subd_faces.size()) {
+		float2 *vert_patch_uv_ptr = &vert_patch_uv[0];
+
+		for(size_t i = 0; i < verts_size; i++) {
+			tri_patch_uv[i] = vert_patch_uv_ptr[i];
+		}
+	}
+
+	size_t triangles_size = num_triangles();
+
 	if(triangles_size) {
 		for(size_t i = 0; i < triangles_size; i++) {
 			Triangle t = get_triangle(i);
@@ -516,6 +616,8 @@ void Mesh::pack_verts(const vector<uint>& tri_prim_index,
 			                           t.v[1] + vert_offset,
 			                           t.v[2] + vert_offset,
 			                           tri_prim_index[i + tri_offset]);
+
+			tri_patch[i] = (!subd_faces.size()) ? -1 : (triangle_patch[i]*8 + patch_offset);
 		}
 	}
 }
@@ -553,6 +655,55 @@ void Mesh::pack_curves(Scene *scene, float4 *curve_key_co, float4 *curve_data, s
 	}
 }
 
+void Mesh::pack_patches(uint *patch_data, uint vert_offset, uint face_offset, uint corner_offset)
+{
+	size_t num_faces = subd_faces.size();
+	int ngons = 0;
+
+	if(num_faces) {
+		for(size_t f = 0; f < num_faces; f++) {
+			SubdFace face = subd_faces[f];
+
+			if(face.is_quad()) {
+				int c[4];
+				memcpy(c, &subd_face_corners[face.start_corner], sizeof(int)*4);
+
+				*(patch_data++) = c[0] + vert_offset;
+				*(patch_data++) = c[1] + vert_offset;
+				*(patch_data++) = c[2] + vert_offset;
+				*(patch_data++) = c[3] + vert_offset;
+
+				*(patch_data++) = f+face_offset;
+				*(patch_data++) = face.num_corners;
+				*(patch_data++) = face.start_corner + corner_offset;
+				*(patch_data++) = 0;
+			}
+			else {
+				for(int i = 0; i < face.num_corners; i++) {
+					int c[4];
+					c[0] = subd_face_corners[face.start_corner + mod(i + 0, face.num_corners)];
+					c[1] = subd_face_corners[face.start_corner + mod(i + 1, face.num_corners)];
+					c[2] = verts.size() - num_subd_verts + ngons;
+					c[3] = subd_face_corners[face.start_corner + mod(i - 1, face.num_corners)];
+
+					*(patch_data++) = c[0] + vert_offset;
+					*(patch_data++) = c[1] + vert_offset;
+					*(patch_data++) = c[2] + vert_offset;
+					*(patch_data++) = c[3] + vert_offset;
+
+					*(patch_data++) = f+face_offset;
+					*(patch_data++) = face.num_corners | (i << 16);
+					*(patch_data++) = face.start_corner + corner_offset;
+					*(patch_data++) = subd_face_corners.size() + ngons + corner_offset;
+				}
+
+				ngons++;
+			}
+		}
+	}
+}
+
+
 void Mesh::compute_bvh(DeviceScene *dscene,
                        SceneParams *params,
                        Progress *progress,
@@ -682,8 +833,9 @@ void MeshManager::update_osl_attributes(Device *device, Scene *scene, vector<Att
 			osl_attr.value = attr;
 			osl_attr.offset = 0;
 
-			og->attribute_map[i*ATTR_PRIM_TYPES][attr.name()] = osl_attr;
+			og->attribute_map[i*ATTR_PRIM_TYPES + ATTR_PRIM_TRIANGLE][attr.name()] = osl_attr;
 			og->attribute_map[i*ATTR_PRIM_TYPES + ATTR_PRIM_CURVE][attr.name()] = osl_attr;
+			og->attribute_map[i*ATTR_PRIM_TYPES + ATTR_PRIM_SUBD][attr.name()] = osl_attr;
 		}
 
 		/* find mesh attributes */
@@ -713,11 +865,11 @@ void MeshManager::update_osl_attributes(Device *device, Scene *scene, vector<Att
 				if(req.std != ATTR_STD_NONE) {
 					/* if standard attribute, add lookup by geom: name convention */
 					ustring stdname(string("geom:") + string(Attribute::standard_name(req.std)));
-					og->attribute_map[i*ATTR_PRIM_TYPES][stdname] = osl_attr;
+					og->attribute_map[i*ATTR_PRIM_TYPES + ATTR_PRIM_TRIANGLE][stdname] = osl_attr;
 				}
 				else if(req.name != ustring()) {
 					/* add lookup by mesh attribute name */
-					og->attribute_map[i*ATTR_PRIM_TYPES][req.name] = osl_attr;
+					og->attribute_map[i*ATTR_PRIM_TYPES + ATTR_PRIM_TRIANGLE][req.name] = osl_attr;
 				}
 			}
 
@@ -742,6 +894,28 @@ void MeshManager::update_osl_attributes(Device *device, Scene *scene, vector<Att
 					og->attribute_map[i*ATTR_PRIM_TYPES + ATTR_PRIM_CURVE][req.name] = osl_attr;
 				}
 			}
+
+			if(req.subd_element != ATTR_ELEMENT_NONE) {
+				osl_attr.elem = req.subd_element;
+				osl_attr.offset = req.subd_offset;
+
+				if(req.subd_type == TypeDesc::TypeFloat)
+					osl_attr.type = TypeDesc::TypeFloat;
+				else if(req.subd_type == TypeDesc::TypeMatrix)
+					osl_attr.type = TypeDesc::TypeMatrix;
+				else
+					osl_attr.type = TypeDesc::TypeColor;
+
+				if(req.std != ATTR_STD_NONE) {
+					/* if standard attribute, add lookup by geom: name convention */
+					ustring stdname(string("geom:") + string(Attribute::standard_name(req.std)));
+					og->attribute_map[i*ATTR_PRIM_TYPES + ATTR_PRIM_SUBD][stdname] = osl_attr;
+				}
+				else if(req.name != ustring()) {
+					/* add lookup by mesh attribute name */
+					og->attribute_map[i*ATTR_PRIM_TYPES + ATTR_PRIM_SUBD][req.name] = osl_attr;
+				}
+			}
 		}
 	}
 #else
@@ -822,22 +996,32 @@ void MeshManager::update_svm_attributes(Device *device, DeviceScene *dscene, Sce
 			}
 
 			index++;
-		}
 
-		/* terminator */
-		attr_map[index].x = ATTR_STD_NONE;
-		attr_map[index].y = 0;
-		attr_map[index].z = 0;
-		attr_map[index].w = 0;
+			if(mesh->subd_faces.size()) {
+				attr_map[index].x = id;
+				attr_map[index].y = req.subd_element;
+				attr_map[index].z = as_uint(req.subd_offset);
 
-		index++;
+				if(req.subd_type == TypeDesc::TypeFloat)
+					attr_map[index].w = NODE_ATTR_FLOAT;
+				else if(req.subd_type == TypeDesc::TypeMatrix)
+					attr_map[index].w = NODE_ATTR_MATRIX;
+				else
+					attr_map[index].w = NODE_ATTR_FLOAT3;
+			}
 
-		attr_map[index].x = ATTR_STD_NONE;
-		attr_map[index].y = 0;
-		attr_map[index].z = 0;
-		attr_map[index].w = 0;
+			index++;
+		}
 
-		index++;
+		/* terminator */
+		for(int i = 0; i < ATTR_PRIM_TYPES; i++) {
+			attr_map[index].x = ATTR_STD_NONE;
+			attr_map[index].y = 0;
+			attr_map[index].z = 0;
+			attr_map[index].w = 0;
+
+			index++;
+		}
 	}
 
 	/* copy to device */
@@ -847,17 +1031,13 @@ void MeshManager::update_svm_attributes(Device *device, DeviceScene *dscene, Sce
 
 static void update_attribute_element_size(Mesh *mesh,
                                           Attribute *mattr,
+                                          AttributePrimitive prim,
                                           size_t *attr_float_size,
                                           size_t *attr_float3_size,
                                           size_t *attr_uchar4_size)
 {
 	if(mattr) {
-		size_t size = mattr->element_size(
-			mesh->verts.size(),
-			mesh->num_triangles(),
-			mesh->motion_steps,
-			mesh->num_curves(),
-			mesh->curve_keys.size());
+		size_t size = mattr->element_size(mesh, prim);
 
 		if(mattr->element == ATTR_ELEMENT_VOXEL) {
 			/* pass */
@@ -885,6 +1065,7 @@ static void update_attribute_element_offset(Mesh *mesh,
                                             vector<uchar4>& attr_uchar4,
                                             size_t& attr_uchar4_offset,
                                             Attribute *mattr,
+                                            AttributePrimitive prim,
                                             TypeDesc& type,
                                             int& offset,
                                             AttributeElement& element)
@@ -895,12 +1076,7 @@ static void update_attribute_element_offset(Mesh *mesh,
 		type = mattr->type;
 
 		/* store attribute data in arrays */
-		size_t size = mattr->element_size(
-			mesh->verts.size(),
-			mesh->num_triangles(),
-			mesh->motion_steps,
-			mesh->num_curves(),
-			mesh->curve_keys.size());
+		size_t size = mattr->element_size(mesh, prim);
 
 		if(mattr->element == ATTR_ELEMENT_VOXEL) {
 			/* store slot in offset value */
@@ -954,10 +1130,18 @@ static void update_attribute_element_offset(Mesh *mesh,
 			offset -= mesh->vert_offset;
 		else if(element == ATTR_ELEMENT_VERTEX_MOTION)
 			offset -= mesh->vert_offset;
-		else if(element == ATTR_ELEMENT_FACE)
-			offset -= mesh->tri_offset;
-		else if(element == ATTR_ELEMENT_CORNER || element == ATTR_ELEMENT_CORNER_BYTE)
-			offset -= 3*mesh->tri_offset;
+		else if(element == ATTR_ELEMENT_FACE) {
+			if(prim == ATTR_PRIM_TRIANGLE)
+				offset -= mesh->tri_offset;
+			else
+				offset -= mesh->face_offset;
+		}
+		else if(element == ATTR_ELEMENT_CORNER || element == ATTR_ELEMENT_CORNER_BYTE) {
+			if(prim == ATTR_PRIM_TRIANGLE)
+				offset -= 3*mesh->tri_offset;
+			else
+				offset -= mesh->corner_offset;
+		}
 		else if(element == ATTR_ELEMENT_CURVE)
 			offset -= mesh->curve_offset;
 		else if(element == ATTR_ELEMENT_CURVE_KEY)
@@ -1007,23 +1191,23 @@ void MeshManager::device_update_attributes(Device *device, DeviceScene *dscene,
 		foreach(AttributeRequest& req, attributes.requests) {
 			Attribute *triangle_mattr = mesh->attributes.find(req);
 			Attribute *curve_mattr = mesh->curve_attributes.find(req);
-
-			/* todo: get rid of this exception, it's only here for giving some
-			 * working texture coordinate for subdivision as we can't preserve
-			 * any attributes yet */
-			if(!triangle_mattr && req.std == ATTR_STD_GENERATED) {
-				triangle_mattr = mesh->attributes.add(ATTR_STD_GENERATED);
-				if(mesh->verts.size())
-					memcpy(triangle_mattr->data_float3(), &mesh->verts[0], sizeof(float3)*mesh->verts.size());
-			}
+			Attribute *subd_mattr = mesh->subd_attributes.find(req);
 
 			update_attribute_element_size(mesh,
 			                              triangle_mattr,
+			                              ATTR_PRIM_TRIANGLE,
 			                              &attr_float_size,
 			                              &attr_float3_size,
 			                              &attr_uchar4_size);
 			update_attribute_element_size(mesh,
 			                              curve_mattr,
+			                              ATTR_PRIM_CURVE,
+			                              &attr_float_size,
+			                              &attr_float3_size,
+			                              &attr_uchar4_size);
+			update_attribute_element_size(mesh,
+			                              subd_mattr,
+			                              ATTR_PRIM_SUBD,
 			                              &attr_float_size,
 			                              &attr_float3_size,
 			                              &attr_uchar4_size);
@@ -1048,12 +1232,14 @@ void MeshManager::device_update_attributes(Device *device, DeviceScene *dscene,
 		foreach(AttributeRequest& req, attributes.requests) {
 			Attribute *triangle_mattr = mesh->attributes.find(req);
 			Attribute *curve_mattr = mesh->curve_attributes.find(req);
+			Attribute *subd_mattr = mesh->subd_attributes.find(req);
 
 			update_attribute_element_offset(mesh,
 			                                attr_float, attr_float_offset,
 			                                attr_float3, attr_float3_offset,
 			                                attr_uchar4, attr_uchar4_offset,
 			                                triangle_mattr,
+			                                ATTR_PRIM_TRIANGLE,
 			                                req.triangle_type,
 			                                req.triangle_offset,
 			                                req.triangle_element);
@@ -1063,10 +1249,21 @@ void MeshManager::device_update_attributes(Device *device, DeviceScene *dscene,
 			                                attr_float3, attr_float3_offset,
 			                                attr_uchar4, attr_uchar4_offset,
 			                                curve_mattr,
+			                                ATTR_PRIM_CURVE,
 			                                req.curve_type,
 			                                req.curve_offset,
 			                                req.curve_element);
 
+			update_attribute_element_offset(mesh,
+			                                attr_float, attr_float_offset,
+			                                attr_float3, attr_float3_offset,
+			                                attr_uchar4, attr_uchar4_offset,
+			                                subd_mattr,
+			                                ATTR_PRIM_SUBD,
+			                                req.subd_type,
+			                                req.subd_offset,
+			                                req.subd_element);
+
 			if(progress.get_cancel()) return;
 		}
 	}
@@ -1100,19 +1297,37 @@ void MeshManager::mesh_calc_offset(Scene *scene)
 {
 	size_t vert_size = 0;
 	size_t tri_size = 0;
+
 	size_t curve_key_size = 0;
 	size_t curve_size = 0;
 
+	size_t patch_size = 0;
+	size_t face_size = 0;
+	size_t corner_size = 0;
+
 	foreach(Mesh *mesh, scene->meshes) {
 		mesh->vert_offset = vert_size;
 		mesh->tri_offset = tri_size;
+
 		mesh->curvekey_offset = curve_key_size;
 		mesh->curve_offset = curve_size;
 
+		mesh->patch_offset = patch_size;
+		mesh->face_offset = face_size;
+		mesh->corner_offset = corner_size;
+
 		vert_size += mesh->verts.size();
 		tri_size += mesh->num_triangles();
+
 		curve_key_size += mesh->curve_keys.size();
 		curve_size += mesh->num_curves();
+
+		if(mesh->subd_faces.size()) {
+			Mesh::SubdFace& last = mesh->subd_faces[mesh->subd_faces.size()-1];
+			patch_size += (last.ptex_offset + last.num_ptex_faces()) * 8;
+		}
+		face_size += mesh->subd_faces.size();
+		corner_size += mesh->subd_face_corners.size();
 	}
 }
 
@@ -1125,14 +1340,25 @@ void MeshManager::device_update_mesh(Device *device,
 	/* Count. */
 	size_t vert_size = 0;
 	size_t tri_size = 0;
+
 	size_t curve_key_size = 0;
 	size_t curve_size = 0;
+
+	size_t patch_size = 0;
+
 	foreach(Mesh *mesh, scene->meshes) {
 		vert_size += mesh->verts.size();
 		tri_size += mesh->num_triangles();
+
 		curve_key_size += mesh->curve_keys.size();
 		curve_size += mesh->num_curves();
+
+		if(mesh->subd_faces.size()) {
+			Mesh::SubdFace& last = mesh->subd_faces[mesh->subd_faces.size()-1];
+			patch_size += (last.ptex_offset + last.num_ptex_faces()) * 8;
+		}
 	}
+
 	/* Create mapping from triangle to primitive triangle array. */
 	vector<uint> tri_prim_index(tri_size);
 	if(for_displacement) {
@@ -1155,6 +1381,7 @@ void MeshManager::device_update_mesh(Device *device,
 			}
 		}
 	}
+
 	/* Fill in all the arrays. */
 	if(tri_size != 0) {
 		/* normals */
@@ -1163,6 +1390,8 @@ void MeshManager::device_update_mesh(Device *device,
 		uint *tri_shader = dscene->tri_shader.resize(tri_size);
 		float4 *vnormal = dscene->tri_vnormal.resize(vert_size);
 		uint4 *tri_vindex = dscene->tri_vindex.resize(tri_size);
+		uint *tri_patch = dscene->tri_patch.resize(tri_size);
+		float2 *tri_patch_uv = dscene->tri_patch_uv.resize(vert_size);
 
 		foreach(Mesh *mesh, scene->meshes) {
 			mesh->pack_normals(scene,
@@ -1170,6 +1399,8 @@ void MeshManager::device_update_mesh(Device *device,
 			                   &vnormal[mesh->vert_offset]);
 			mesh->pack_verts(tri_prim_index,
 			                 &tri_vindex[mesh->tri_offset],
+			                 &tri_patch[mesh->tri_offset],
+			                 &tri_patch_uv[mesh->vert_offset],
 			                 mesh->vert_offset,
 			                 mesh->tri_offset);
 			if(progress.get_cancel()) return;
@@ -1181,7 +1412,10 @@ void MeshManager::device_update_mesh(Device *device,
 		device->tex_alloc("__tri_shader", dscene->tri_shader);
 		device->tex_alloc("__tri_vnormal", dscene->tri_vnormal);
 		device->tex_alloc("__tri_vindex", dscene->tri_vindex);
+		device->tex_alloc("__tri_patch", dscene->tri_patch);
+		device->tex_alloc("__tri_patch_uv", dscene->tri_patch_uv);
 	}
+
 	if(curve_size != 0) {
 		progress.set_status("Updating Mesh", "Copying Strands to device");
 
@@ -1196,6 +1430,20 @@ void MeshManager::device_update_mesh(Device *device,
 		device->tex_alloc("__curve_keys", dscene->curve_keys);
 		device->tex_alloc("__curves", dscene->curves);
 	}
+
+	if(patch_size != 0) {
+		progress.set_status("Updating Mesh", "Copying Patches to device");
+
+		uint *patch_data = dscene->patches.resize(patch_size);
+
+		foreach(Mesh *mesh, scene->meshes) {
+			mesh->pack_patches(&patch_data[mesh->patch_offset], mesh->vert_offset, mesh->face_offset, mesh->corner_offset);
+			if(progress.get_cancel()) return;
+		}
+
+		device->tex_alloc("__patches", dscene->patches);
+	}
+
 	if(for_displacement) {
 		float4 *prim_tri_verts = dscene->prim_tri_verts.resize(tri_size * 3);
 		foreach(Mesh *mesh, scene->meshes) {
@@ -1433,7 +1681,9 @@ void MeshManager::device_update(Device *device, DeviceScene *dscene, Scene *scen
 			num_bvh++;
 		}
 	}
+
 	TaskPool pool;
+
 	foreach(Mesh *mesh, scene->meshes) {
 		if(mesh->need_update) {
 			pool.push(function_bind(&Mesh::compute_bvh,
@@ -1448,6 +1698,7 @@ void MeshManager::device_update(Device *device, DeviceScene *dscene, Scene *scen
 			}
 		}
 	}
+
 	TaskPool::Summary summary;
 	pool.wait_work(&summary);
 	VLOG(2) << "Objects BVH build pool statistics:\n"
@@ -1505,8 +1756,11 @@ void MeshManager::device_free(Device *device, DeviceScene *dscene)
 	device->tex_free(dscene->tri_shader);
 	device->tex_free(dscene->tri_vnormal);
 	device->tex_free(dscene->tri_vindex);
+	device->tex_free(dscene->tri_patch);
+	device->tex_free(dscene->tri_patch_uv);
 	device->tex_free(dscene->curves);
 	device->tex_free(dscene->curve_keys);
+	device->tex_free(dscene->patches);
 	device->tex_free(dscene->attributes_map);
 	device->tex_free(dscene->attributes_float);
 	device->tex_free(dscene->attributes_float3);
@@ -1523,8 +1777,11 @@ void MeshManager::device_free(Device *device, DeviceScene *dscene)
 	dscene->tri_shader.clear();
 	dscene->tri_vnormal.clear();
 	dscene->tri_vindex.clear();
+	dscene->tri_patch.clear();
+	dscene->tri_patch_uv.clear();
 	dscene->curves.clear();
 	dscene->curve_keys.clear();
+	dscene->patches.clear();
 	dscene->attributes_map.clear();
 	dscene->attributes_float.clear();
 	dscene->attributes_float3.clear();
@@ -1574,77 +1831,5 @@ bool Mesh::need_attribute(Scene * /*scene*/, ustring name)
 	return false;
 }
 
-void Mesh::tessellate(DiagSplit *split)
-{
-	int num_faces = num_triangles();
-
-	add_face_normals();
-	add_vertex_normals();
-
-	Attribute *attr_fN = attributes.find(ATTR_STD_FACE_NORMAL);
-	float3 *fN = attr_fN->data_float3();
-
-	Attribute *attr_vN = attributes.find(ATTR_STD_VERTEX_NORMAL);
-	float3 *vN = attr_vN->data_float3();
-
-	for(int f = 0; f < num_faces; f++) {
-		if(!forms_quad[f]) {
-			/* triangle */
-			LinearTrianglePatch patch;
-			Triangle triangle = get_triangle(f);
-			float3 *hull = patch.hull;
-			float3 *normals = patch.normals;
-
-			for(int i = 0; i < 3; i++) {
-				hull[i] = verts[triangle.v[i]];
-			}
-
-			if(smooth[f]) {
-				for(int i = 0; i < 3; i++) {
-					normals[i] = vN[triangle.v[i]];
-				}
-			}
-			else {
-				for(int i = 0; i < 3; i++) {
-					normals[i] = fN[f];
-				}
-			}
-
-			split->split_triangle(&patch);
-		}
-		else {
-			/* quad */
-			LinearQuadPatch patch;
-			Triangle triangle0 = get_triangle(f);
-			Triangle triangle1 = get_triangle(f+1);
-			float3 *hull = patch.hull;
-			float3 *normals = patch.normals;
-
-			hull[0] = verts[triangle0.v[0]];
-			hull[1] = verts[triangle0.v[1]];
-			hull[3] = verts[triangle0.v[2]];
-			hull[2] = verts[triangle1.v[2]];
-
-			if(smooth[f]) {
-				normals[0] = vN[triangle0.v[0]];
-				normals[1] = vN[triangle0.v[1]];
-				normals[3] = vN[triangle0.v[2]];
-				normals[2] = vN[triangle1.v[2]];
-			}
-			else {
-				for(int i = 0; i < 4; i++) {
-					normals[i] = fN[f];
-				}
-			}
-
-			split->split_quad(&patch);
-
-			// consume second triangle in quad
-			f++;
-		}
-
-	}
-}
-
 CCL_NAMESPACE_END
 
diff --git a/intern/cycles/render/mesh.h b/intern/cycles/render/mesh.h
index 0aea55544f2..adb639201ce 100644
--- a/intern/cycles/render/mesh.h
+++ b/intern/cycles/render/mesh.h
@@ -97,6 +97,19 @@ public:
 		return curve_first_key.size();
 	}
 
+	/* Mesh SubdFace */
+	struct SubdFace {
+		int start_corner;
+		int num_corners;
+		int shader;
+		bool smooth;
+		int ptex_offset;
+
+		bool is_quad() { return num_corners == 4; }
+		float3 normal(const Mesh *mesh) const;
+		int num_ptex_faces() const { return num_corners == 4 ? 1 : num_corners; }
+	};
+
 	/* Displacement */
 	enum DisplacementMethod {
 		DISPLACE_BUMP = 0,
@@ -119,7 +132,10 @@ public:
 	array<float3> verts;
 	array<int> shader;
 	array<bool> smooth;
-	array<bool> forms_quad; /* used to tell if triangle is part of a quad patch */
+
+	/* used for storing patch info for subd triangles, only allocated if there are patches */
+	array<int> triangle_patch; /* must be < 0 for non subd triangles */
+	array<float2> vert_patch_uv;
 
 	bool has_volume;  /* Set in the device_update_flags(). */
 	bool has_surface_bssrdf;  /* Set in the device_update_flags(). */
@@ -129,9 +145,14 @@ public:
 	array<int> curve_first_key;
 	array<int> curve_shader;
 
+	array<SubdFace> subd_faces;
+	array<int> subd_face_corners;
+	int num_ngons;
+
 	vector<Shader*> used_shaders;
 	AttributeSet attributes;
 	AttributeSet curve_attributes;
+	AttributeSet subd_attributes;
 
 	BoundBox bounds;
 	bool transform_applied;
@@ -154,6 +175,12 @@ public:
 	size_t curve_offset;
 	size_t curvekey_offset;
 
+	size_t patch_offset;
+	size_t face_offset;
+	size_t corner_offset;
+
+	size_t num_subd_verts;
+
 	/* Functions */
 	Mesh();
 	~Mesh();
@@ -162,12 +189,15 @@ public:
 	void reserve_mesh(int numverts, int numfaces);
 	void resize_curves(int numcurves, int numkeys);
 	void reserve_curves(int numcurves, int numkeys);
+	void resize_subd_faces(int numfaces, int num_ngons, int numcorners);
+	void reserve_subd_faces(int numfaces, int num_ngons, int numcorners);
 	void clear();
 	void add_vertex(float3 P);
 	void add_vertex_slow(float3 P);
-	void add_triangle(int v0, int v1, int v2, int shader, bool smooth, bool forms_quad = false);
+	void add_triangle(int v0, int v1, int v2, int shader, bool smooth);
 	void add_curve_key(float3 loc, float radius);
 	void add_curve(int first_key, int shader);
+	void add_subd_face(int* corners, int num_corners, int shader_, bool smooth_);
 	int split_vertex(int vertex);
 
 	void compute_bounds();
@@ -177,9 +207,13 @@ public:
 	void pack_normals(Scene *scene, uint *shader, float4 *vnormal);
 	void pack_verts(const vector<uint>& tri_prim_index,
 	                uint4 *tri_vindex,
+	                uint *tri_patch,
+	                float2 *tri_patch_uv,
 	                size_t vert_offset,
 	                size_t tri_offset);
 	void pack_curves(Scene *scene, float4 *curve_key_co, float4 *curve_data, size_t curvekey_offset);
+	void pack_patches(uint *patch_data, uint vert_offset, uint face_offset, uint corner_offset);
+
 	void compute_bvh(DeviceScene *dscene,
 	                 SceneParams *params,
 	                 Progress *progress,
diff --git a/intern/cycles/render/mesh_subdivision.cpp b/intern/cycles/render/mesh_subdivision.cpp
new file mode 100644
index 00000000000..fe8e41e8d35
--- /dev/null
+++ b/intern/cycles/render/mesh_subdivision.cpp
@@ -0,0 +1,224 @@
+/*
+ * Copyright 2011-2016 Blender Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "mesh.h"
+#include "attribute.h"
+
+#include "subd_split.h"
+#include "subd_patch.h"
+
+#include "util_foreach.h"
+
+CCL_NAMESPACE_BEGIN
+
+void Mesh::tessellate(DiagSplit *split)
+{
+	int num_faces = subd_faces.size();
+
+	Attribute *attr_vN = subd_attributes.find(ATTR_STD_VERTEX_NORMAL);
+	float3* vN = attr_vN->data_float3();
+
+	for(int f = 0; f < num_faces; f++) {
+		SubdFace& face = subd_faces[f];
+
+		if(face.is_quad()) {
+			/* quad */
+			LinearQuadPatch patch;
+			float3 *hull = patch.hull;
+			float3 *normals = patch.normals;
+
+			patch.patch_index = face.ptex_offset;
+			patch.shader = face.shader;
+
+			for(int i = 0; i < 4; i++) {
+				hull[i] = verts[subd_face_corners[face.start_corner+i]];
+			}
+
+			if(face.smooth) {
+				for(int i = 0; i < 4; i++) {
+					normals[i] = vN[subd_face_corners[face.start_corner+i]];
+				}
+			}
+			else {
+				float3 N = face.normal(this);
+				for(int i = 0; i < 4; i++) {
+					normals[i] = N;
+				}
+			}
+
+			swap(hull[2], hull[3]);
+			swap(normals[2], normals[3]);
+
+			/* Quad faces need to be split at least once to line up with split ngons, we do this
+			 * here in this manner because if we do it later edge factors may end up slightly off.
+			 */
+			QuadDice::SubPatch subpatch;
+			subpatch.patch = &patch;
+
+			subpatch.P00 = make_float2(0.0f, 0.0f);
+			subpatch.P10 = make_float2(0.5f, 0.0f);
+			subpatch.P01 = make_float2(0.0f, 0.5f);
+			subpatch.P11 = make_float2(0.5f, 0.5f);
+			split->split_quad(&patch, &subpatch);
+
+			subpatch.P00 = make_float2(0.5f, 0.0f);
+			subpatch.P10 = make_float2(1.0f, 0.0f);
+			subpatch.P01 = make_float2(0.5f, 0.5f);
+			subpatch.P11 = make_float2(1.0f, 0.5f);
+			split->split_quad(&patch, &subpatch);
+
+			subpatch.P00 = make_float2(0.0f, 0.5f);
+			subpatch.P10 = make_float2(0.5f, 0.5f);
+			subpatch.P01 = make_float2(0.0f, 1.0f);
+			subpatch.P11 = make_float2(0.5f, 1.0f);
+			split->split_quad(&patch, &subpatch);
+
+			subpatch.P00 = make_float2(0.5f, 0.5f);
+			subpatch.P10 = make_float2(1.0f, 0.5f);
+			subpatch.P01 = make_float2(0.5f, 1.0f);
+			subpatch.P11 = make_float2(1.0f, 1.0f);
+			split->split_quad(&patch, &subpatch);
+		}
+		else {
+			/* ngon */
+			float3 center_vert = make_float3(0.0f, 0.0f, 0.0f);
+			float3 center_normal = make_float3(0.0f, 0.0f, 0.0f);
+
+			float inv_num_corners = 1.0f/float(face.num_corners);
+			for(int corner = 0; corner < face.num_corners; corner++) {
+				center_vert += verts[subd_face_corners[face.start_corner + corner]] * inv_num_corners;
+				center_normal += vN[subd_face_corners[face.start_corner + corner]] * inv_num_corners;
+			}
+
+			for(int corner = 0; corner < face.num_corners; corner++) {
+				LinearQuadPatch patch;
+				float3 *hull = patch.hull;
+				float3 *normals = patch.normals;
+
+				patch.patch_index = face.ptex_offset + corner;
+
+				patch.shader = face.shader;
+
+				hull[0] = verts[subd_face_corners[face.start_corner + mod(corner + 0, face.num_corners)]];
+				hull[1] = verts[subd_face_corners[face.start_corner + mod(corner + 1, face.num_corners)]];
+				hull[2] = verts[subd_face_corners[face.start_corner + mod(corner - 1, face.num_corners)]];
+				hull[3] = center_vert;
+
+				hull[1] = (hull[1] + hull[0]) * 0.5;
+				hull[2] = (hull[2] + hull[0]) * 0.5;
+
+				if(face.smooth) {
+					normals[0] = vN[subd_face_corners[face.start_corner + mod(corner + 0, face.num_corners)]];
+					normals[1] = vN[subd_face_corners[face.start_corner + mod(corner + 1, face.num_corners)]];
+					normals[2] = vN[subd_face_corners[face.start_corner + mod(corner - 1, face.num_corners)]];
+					normals[3] = center_normal;
+
+					normals[1] = (normals[1] + normals[0]) * 0.5;
+					normals[2] = (normals[2] + normals[0]) * 0.5;
+				}
+				else {
+					float3 N = face.normal(this);
+					for(int i = 0; i < 4; i++) {
+						normals[i] = N;
+					}
+				}
+
+				split->split_quad(&patch);
+			}
+		}
+	}
+
+	/* interpolate center points for attributes */
+	foreach(Attribute& attr, subd_attributes.attributes) {
+		char* data = attr.data();
+		size_t stride = attr.data_sizeof();
+		int ngons = 0;
+
+		switch(attr.element) {
+			case ATTR_ELEMENT_VERTEX: {
+				for(int f = 0; f < num_faces; f++) {
+					SubdFace& face = subd_faces[f];
+
+					if(!face.is_quad()) {
+						char* center = data + (verts.size() - num_subd_verts + ngons) * stride;
+						attr.zero_data(center);
+
+						float inv_num_corners = 1.0f / float(face.num_corners);
+
+						for(int corner = 0; corner < face.num_corners; corner++) {
+							attr.add_with_weight(center,
+							                     data + subd_face_corners[face.start_corner + corner] * stride,
+							                     inv_num_corners);
+						}
+
+						ngons++;
+					}
+				}
+			} break;
+			case ATTR_ELEMENT_VERTEX_MOTION: {
+				// TODO(mai): implement
+			} break;
+			case ATTR_ELEMENT_CORNER: {
+				for(int f = 0; f < num_faces; f++) {
+					SubdFace& face = subd_faces[f];
+
+					if(!face.is_quad()) {
+						char* center = data + (subd_face_corners.size() + ngons) * stride;
+						attr.zero_data(center);
+
+						float inv_num_corners = 1.0f / float(face.num_corners);
+
+						for(int corner = 0; corner < face.num_corners; corner++) {
+							attr.add_with_weight(center,
+							                     data + (face.start_corner + corner) * stride,
+							                     inv_num_corners);
+						}
+
+						ngons++;
+					}
+				}
+			} break;
+			case ATTR_ELEMENT_CORNER_BYTE: {
+				for(int f = 0; f < num_faces; f++) {
+					SubdFace& face = subd_faces[f];
+
+					if(!face.is_quad()) {
+						uchar* center = (uchar*)data + (subd_face_corners.size() + ngons) * stride;
+
+						float inv_num_corners = 1.0f / float(face.num_corners);
+						float4 val = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
+
+						for(int corner = 0; corner < face.num_corners; corner++) {
+							for(int i = 0; i < 4; i++) {
+								val[i] += float(*(data + (face.start_corner + corner) * stride + i)) * inv_num_corners;
+							}
+						}
+
+						for(int i = 0; i < 4; i++) {
+							center[i] = uchar(min(max(val[i], 0.0f), 255.0f));
+						}
+
+						ngons++;
+					}
+				}
+			} break;
+			default: break;
+		}
+	}
+}
+
+CCL_NAMESPACE_END
+
diff --git a/intern/cycles/render/scene.h b/intern/cycles/render/scene.h
index 05e807ff60c..9e72f197cce 100644
--- a/intern/cycles/render/scene.h
+++ b/intern/cycles/render/scene.h
@@ -74,10 +74,14 @@ public:
 	device_vector<uint> tri_shader;
 	device_vector<float4> tri_vnormal;
 	device_vector<uint4> tri_vindex;
+	device_vector<uint> tri_patch;
+	device_vector<float2> tri_patch_uv;
 
 	device_vector<float4> curves;
 	device_vector<float4> curve_keys;
 
+	device_vector<uint> patches;
+
 	/* objects */
 	device_vector<float4> objects;
 	device_vector<float4> objects_vector;
diff --git a/intern/cycles/subd/CMakeLists.txt b/intern/cycles/subd/CMakeLists.txt
index d1708868fd0..db497013693 100644
--- a/intern/cycles/subd/CMakeLists.txt
+++ b/intern/cycles/subd/CMakeLists.txt
@@ -14,14 +14,12 @@ set(INC_SYS
 
 set(SRC
 	subd_dice.cpp
-	subd_mesh.cpp
 	subd_patch.cpp
 	subd_split.cpp
 )
 
 set(SRC_HEADERS
 	subd_dice.h
-	subd_mesh.h
 	subd_patch.h
 	subd_split.h
 )
diff --git a/intern/cycles/subd/subd_dice.cpp b/intern/cycles/subd/subd_dice.cpp
index 7c74f21950e..36981a20f3c 100644
--- a/intern/cycles/subd/subd_dice.cpp
+++ b/intern/cycles/subd/subd_dice.cpp
@@ -48,6 +48,11 @@ void EdgeDice::reserve(int num_verts)
 	vert_offset = mesh->verts.size();
 	tri_offset = mesh->num_triangles();
 
+	/* todo: optimize so we can reserve in advance, this is like push_back_slow() */
+	if(vert_offset + num_verts > mesh->verts.capacity()) {
+		mesh->reserve_mesh(size_t((vert_offset + num_verts) * 1.2), mesh->num_triangles());
+	}
+
 	mesh->resize_mesh(vert_offset + num_verts, tri_offset);
 
 	Attribute *attr_vN = mesh->attributes.add(ATTR_STD_VERTEX_NORMAL);
@@ -66,6 +71,7 @@ int EdgeDice::add_vert(Patch *patch, float2 uv)
 
 	mesh_P[vert_offset] = P;
 	mesh_N[vert_offset] = N;
+	params.mesh->vert_patch_uv[vert_offset] = make_float2(uv.x, uv.y);
 
 	if(params.ptex) {
 		Attribute *attr_ptex_uv = params.mesh->attributes.add(ATTR_STD_PTEX_UV);
@@ -75,6 +81,8 @@ int EdgeDice::add_vert(Patch *patch, float2 uv)
 		ptex_uv[vert_offset] = make_float3(uv.x, uv.y, 0.0f);
 	}
 
+	params.mesh->num_subd_verts++;
+
 	return vert_offset++;
 }
 
@@ -86,7 +94,8 @@ void EdgeDice::add_triangle(Patch *patch, int v0, int v1, int v2)
 	if(mesh->triangles.size() == mesh->triangles.capacity())
 		mesh->reserve_mesh(mesh->verts.size(), size_t(max(mesh->num_triangles() + 1, 1) * 1.2));
 
-	mesh->add_triangle(v0, v1, v2, params.shader, params.smooth, false);
+	mesh->add_triangle(v0, v1, v2, patch->shader, true);
+	params.mesh->triangle_patch[params.mesh->num_triangles()-1] = patch->patch_index;
 
 	if(params.ptex) {
 		Attribute *attr_ptex_face_id = params.mesh->attributes.add(ATTR_STD_PTEX_FACE_ID);
@@ -340,160 +349,5 @@ void QuadDice::dice(SubPatch& sub, EdgeFactors& ef)
 	assert(vert_offset == params.mesh->verts.size());
 }
 
-/* TriangleDice */
-
-TriangleDice::TriangleDice(const SubdParams& params_)
-: EdgeDice(params_)
-{
-}
-
-void TriangleDice::reserve(EdgeFactors& ef, int M)
-{
-	int num_verts = ef.tu + ef.tv + ef.tw;
-
-	for(int m = M-2; m > 0; m -= 2)
-		num_verts += 3 + (m-1)*3;
-	
-	if(!(M & 1))
-		num_verts++;
-	
-	EdgeDice::reserve(num_verts);
-}
-
-float2 TriangleDice::map_uv(SubPatch& sub, float2 uv)
-{
-	/* map UV from subpatch to patch parametric coordinates */
-	return uv.x*sub.Pu + uv.y*sub.Pv + (1.0f - uv.x - uv.y)*sub.Pw;
-}
-
-int TriangleDice::add_vert(SubPatch& sub, float2 uv)
-{
-	return EdgeDice::add_vert(sub.patch, map_uv(sub, uv));
-}
-
-void TriangleDice::add_grid(SubPatch& sub, EdgeFactors& ef, int M)
-{
-	// XXX normals are flipped, why?
-
-	/* grid is constructed starting from the outside edges, and adding
-	 * progressively smaller inner triangles that connected to the outer
-	 * one, until M = 1 or 2, the we fill up the last part. */
-	vector<int> outer_u, outer_v, outer_w;
-	int m;
-
-	/* add outer corners vertices */
-	{
-		float2 p_u = make_float2(1.0f, 0.0f);
-		float2 p_v = make_float2(0.0f, 1.0f);
-		float2 p_w = make_float2(0.0f, 0.0f);
-
-		int corner_u = add_vert(sub, p_u);
-		int corner_v = add_vert(sub, p_v);
-		int corner_w = add_vert(sub, p_w);
-
-		outer_u.push_back(corner_v);
-		outer_v.push_back(corner_w);
-		outer_w.push_back(corner_u);
-
-		for(int i = 1; i < ef.tu; i++)
-			outer_u.push_back(add_vert(sub, interp(p_v, p_w, i/(float)ef.tu)));
-		for(int i = 1; i < ef.tv; i++)
-			outer_v.push_back(add_vert(sub, interp(p_w, p_u, i/(float)ef.tv)));
-		for(int i = 1; i < ef.tw; i++)
-			outer_w.push_back(add_vert(sub, interp(p_u, p_v, i/(float)ef.tw)));
-
-		outer_u.push_back(corner_w);
-		outer_v.push_back(corner_u);
-		outer_w.push_back(corner_v);
-	}
-
-	for(m = M-2; m > 0; m -= 2) {
-		vector<int> inner_u, inner_v, inner_w;
-
-		const float t0 = m / (float)M;
-		float2 center = make_float2(1.0f/3.0f, 1.0f/3.0f);
-
-		/* 3 corner vertices */
-		float2 p_u = interp(center, make_float2(1.0f, 0.0f), t0);
-		float2 p_v = interp(center, make_float2(0.0f, 1.0f), t0);
-		float2 p_w = interp(center, make_float2(0.0f, 0.0f), t0);
-
-		int corner_u = add_vert(sub, p_u);
-		int corner_v = add_vert(sub, p_v);
-		int corner_w = add_vert(sub, p_w);
-
-		/* construct array of vertex indices for each side */
-		inner_u.push_back(corner_v);
-		inner_v.push_back(corner_w);
-		inner_w.push_back(corner_u);
-
-		for(int i = 1; i < m; i++) {
-			/* add vertices between corners */
-			const float t1 = i / (float)m;
-
-			inner_u.push_back(add_vert(sub, interp(p_v, p_w, t1)));
-			inner_v.push_back(add_vert(sub, interp(p_w, p_u, t1)));
-			inner_w.push_back(add_vert(sub, interp(p_u, p_v, t1)));
-		}
-
-		inner_u.push_back(corner_w);
-		inner_v.push_back(corner_u);
-		inner_w.push_back(corner_v);
-
-		/* stitch together inner/outer with triangles */
-		stitch_triangles(sub.patch, outer_u, inner_u);
-		stitch_triangles(sub.patch, outer_v, inner_v);
-		stitch_triangles(sub.patch, outer_w, inner_w);
-
-		outer_u = inner_u;
-		outer_v = inner_v;
-		outer_w = inner_w;
-	}
-
-	/* fill up last part */
-	if(m == -1) {
-		/* single triangle */
-		add_triangle(sub.patch, outer_w[0], outer_u[0], outer_v[0]);
-	}
-	else {
-		/* center vertex + up to 6 triangles */
-		int center = add_vert(sub, make_float2(1.0f/3.0f, 1.0f/3.0f));
-
-		add_triangle(sub.patch, outer_w[0], outer_w[1], center);
-		/* if this is false then there is only one triangle on this side */
-		if(outer_w.size() > 2)
-			add_triangle(sub.patch, outer_w[1], outer_w[2], center);
-
-		add_triangle(sub.patch, outer_u[0], outer_u[1], center);
-		if(outer_u.size() > 2)
-			add_triangle(sub.patch, outer_u[1], outer_u[2], center);
-
-		add_triangle(sub.patch, outer_v[0], outer_v[1], center);
-		if(outer_v.size() > 2)
-			add_triangle(sub.patch, outer_v[1], outer_v[2], center);
-	}
-}
-
-void TriangleDice::dice(SubPatch& sub, EdgeFactors& ef)
-{
-	/* todo: handle 2 1 1 resolution */
-	int M = max(ef.tu, max(ef.tv, ef.tw));
-
-	/* Due to the "slant" of the edges of a triangle compared to a quad, the internal
-	 * triangles end up smaller, causing over-tessellation. This is to correct for this
-	 * difference in area. Technically its only correct for equilateral triangles, but
-	 * its better than how it was.
-	 *
-	 * (2*cos(radians(30))/3)**0.5
-	 */
-	float S = 0.7598356856515927f;
-	M = max((int)ceil(S*M), 1);
-
-	reserve(ef, M);
-	add_grid(sub, ef, M);
-
-	assert(vert_offset == params.mesh->verts.size());
-}
-
 CCL_NAMESPACE_END
 
diff --git a/intern/cycles/subd/subd_dice.h b/intern/cycles/subd/subd_dice.h
index 85bd0ea28f0..3002ec780e8 100644
--- a/intern/cycles/subd/subd_dice.h
+++ b/intern/cycles/subd/subd_dice.h
@@ -33,8 +33,6 @@ class Patch;
 
 struct SubdParams {
 	Mesh *mesh;
-	int shader;
-	bool smooth;
 	bool ptex;
 
 	int test_steps;
@@ -44,11 +42,9 @@ struct SubdParams {
 	Camera *camera;
 	Transform objecttoworld;
 
-	SubdParams(Mesh *mesh_, int shader_, bool smooth_ = true, bool ptex_ = false)
+	SubdParams(Mesh *mesh_, bool ptex_ = false)
 	{
 		mesh = mesh_;
-		shader = shader_;
-		smooth = smooth_;
 		ptex = ptex_;
 
 		test_steps = 3;
@@ -136,46 +132,6 @@ public:
 	void dice(SubPatch& sub, EdgeFactors& ef);
 };
 
-/* Triangle EdgeDice
- *
- * Edge tessellation factors and subpatch coordinates are as follows:
- *
- *        Pw
- *        /\
- *    tv /  \ tu
- *      /    \
- *     /      \
- *  Pu -------- Pv
- *        tw     
- */
-
-class TriangleDice : public EdgeDice {
-public:
-	struct SubPatch {
-		Patch *patch;
-
-		float2 Pu;
-		float2 Pv;
-		float2 Pw;
-	};
-
-	struct EdgeFactors {
-		int tu;
-		int tv;
-		int tw;
-	};
-
-	explicit TriangleDice(const SubdParams& params);
-
-	void reserve(EdgeFactors& ef, int M);
-
-	float2 map_uv(SubPatch& sub, float2 uv);
-	int add_vert(SubPatch& sub, float2 uv);
-
-	void add_grid(SubPatch& sub, EdgeFactors& ef, int M);
-	void dice(SubPatch& sub, EdgeFactors& ef);
-};
-
 CCL_NAMESPACE_END
 
 #endif /* __SUBD_DICE_H__ */
diff --git a/intern/cycles/subd/subd_mesh.cpp b/intern/cycles/subd/subd_mesh.cpp
deleted file mode 100644
index 56d7d2b2303..00000000000
--- a/intern/cycles/subd/subd_mesh.cpp
+++ /dev/null
@@ -1,419 +0,0 @@
-/*
- * Original code in the public domain -- castanyo@yahoo.es
- * 
- * Modifications copyright (c) 2011, Blender Foundation.
- * All rights reserved.
- * 
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions are
- * met:
- * * Redistributions of source code must retain the above copyright
- *   notice, this list of conditions and the following disclaimer.
- * * Redistributions in binary form must reproduce the above copyright
- *   notice, this list of conditions and the following disclaimer in the
- *   documentation and/or other materials provided with the distribution.
- * 
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- */
-
-#include <stdio.h>
-
-#include "subd_mesh.h"
-#include "subd_patch.h"
-#include "subd_split.h"
-
-#include "util_debug.h"
-#include "util_foreach.h"
-
-#ifdef WITH_OPENSUBDIV
-
-#include <osd/vertex.h>
-#include <osd/mesh.h>
-#include <osd/cpuComputeController.h>
-#include <osd/cpuVertexBuffer.h>
-#include <osd/cpuEvalLimitController.h>
-#include <osd/evalLimitContext.h>
-
-CCL_NAMESPACE_BEGIN
-
-/* typedefs */
-typedef OpenSubdiv::OsdVertex OsdVertex;
-typedef OpenSubdiv::FarMesh<OsdVertex> OsdFarMesh;
-typedef OpenSubdiv::FarMeshFactory<OsdVertex> OsdFarMeshFactory;
-typedef OpenSubdiv::HbrCatmarkSubdivision<OsdVertex> OsdHbrCatmarkSubdivision;
-typedef OpenSubdiv::HbrFace<OsdVertex> OsdHbrFace;
-typedef OpenSubdiv::HbrHalfedge<OsdVertex> OsdHbrHalfEdge;
-typedef OpenSubdiv::HbrMesh<OsdVertex> OsdHbrMesh;
-typedef OpenSubdiv::HbrVertex<OsdVertex> OsdHbrVertex;
-typedef OpenSubdiv::OsdCpuComputeContext OsdCpuComputeContext;
-typedef OpenSubdiv::OsdCpuComputeController OsdCpuComputeController;
-typedef OpenSubdiv::OsdCpuEvalLimitContext OsdCpuEvalLimitContext;
-typedef OpenSubdiv::OsdCpuEvalLimitController OsdCpuEvalLimitController;
-typedef OpenSubdiv::OsdCpuVertexBuffer OsdCpuVertexBuffer;
-typedef OpenSubdiv::OsdEvalCoords OsdEvalCoords;
-typedef OpenSubdiv::OsdVertexBufferDescriptor OsdVertexBufferDescriptor;
-
-/* OpenSubdiv Patch */
-
-class OpenSubdPatch : public Patch {
-public:
-	int face_id;
-
-	OpenSubdPatch(OsdFarMesh *farmesh, OsdCpuVertexBuffer *vbuf_base)
-	{
-		face_id = 0;
-
-		/* create buffers for evaluation */
-		vbuf_P = OsdCpuVertexBuffer::Create(3, 1);
-		vbuf_dPdu = OsdCpuVertexBuffer::Create(3, 1);
-		vbuf_dPdv = OsdCpuVertexBuffer::Create(3, 1);
-
-		P = vbuf_P->BindCpuBuffer();
-		dPdu = vbuf_dPdu->BindCpuBuffer();
-		dPdv = vbuf_dPdv->BindCpuBuffer();
-
-		/* setup evaluation context */
-		OsdVertexBufferDescriptor in_desc(0, 3, 3), out_desc(0, 3, 3); /* offset, length, stride */
-
-		evalctx = OsdCpuEvalLimitContext::Create(farmesh, false);
-		evalctx->GetVertexData().Bind(in_desc, vbuf_base, out_desc, vbuf_P, vbuf_dPdu, vbuf_dPdv);
-	}
-
-	~OpenSubdPatch()
-	{
-		evalctx->GetVertexData().Unbind();
-
-		delete evalctx;
-		delete vbuf_P;
-		delete vbuf_dPdu;
-		delete vbuf_dPdv;
-	}
-
-	void eval(float3 *P_, float3 *dPdu_, float3 *dPdv_, float u, float v)
-	{
-		OsdEvalCoords coords;
-		coords.u = u;
-		coords.v = v;
-		coords.face = face_id;
-
-		evalctrl.EvalLimitSample<OsdCpuVertexBuffer,OsdCpuVertexBuffer>(coords, evalctx, 0);
-
-		*P_ = make_float3(P[0], P[1], P[2]);
-		if(dPdu_) *dPdu_ = make_float3(dPdv[0], dPdv[1], dPdv[2]);
-		if(dPdv_) *dPdv_ = make_float3(dPdu[0], dPdu[1], dPdu[2]);
-
-		/* optimize: skip evaluating derivatives when not needed */
-		/* todo: swapped derivatives, different winding convention? */
-	}
-
-	BoundBox bound()
-	{
-		/* not implemented */
-		BoundBox bbox = BoundBox::empty;
-		return bbox;
-	}
-
-	int ptex_face_id()
-	{
-		return face_id;
-	}
-
-protected:
-	OsdCpuEvalLimitController evalctrl;
-	OsdCpuEvalLimitContext *evalctx;
-	OsdCpuVertexBuffer *vbuf_P;
-	OsdCpuVertexBuffer *vbuf_dPdu;
-	OsdCpuVertexBuffer *vbuf_dPdv;
-	float *P;
-	float *dPdu;
-	float *dPdv;
-};
-
-/* OpenSubdiv Mesh */
-
-OpenSubdMesh::OpenSubdMesh()
-{
-	/* create osd mesh */
-	static OsdHbrCatmarkSubdivision	catmark;
-	OsdHbrMesh *hbrmesh = new OsdHbrMesh(&catmark);
-
-	/* initialize class */
-	num_verts = 0;
-	num_ptex_faces = 0;
-	_hbrmesh = (void*)hbrmesh;
-}
-
-OpenSubdMesh::~OpenSubdMesh()
-{
-	OsdHbrMesh *hbrmesh = (OsdHbrMesh*)_hbrmesh;
-
-	if(hbrmesh)
-		delete hbrmesh;
-}
-
-void OpenSubdMesh::add_vert(const float3& co)
-{
-	OsdHbrMesh *hbrmesh = (OsdHbrMesh*)_hbrmesh;
-
-	OsdVertex v;
-	positions.push_back(co.x);
-	positions.push_back(co.y);
-	positions.push_back(co.z);
-	hbrmesh->NewVertex(num_verts++, v);
-}
-
-void OpenSubdMesh::add_face(int v0, int v1, int v2)
-{
-	int index[3] = {v0, v1, v2};
-	return add_face(index, 3);
-}
-
-void OpenSubdMesh::add_face(int v0, int v1, int v2, int v3)
-{
-	int index[4] = {v0, v1, v2, v3};
-	add_face(index, 4);
-}
-
-void OpenSubdMesh::add_face(int *index, int num)
-{
-	OsdHbrMesh *hbrmesh = (OsdHbrMesh*)_hbrmesh;
-
-#ifndef NDEBUG
-	/* sanity checks */
-	for(int j = 0; j < num; j++) {
-		OsdHbrVertex *origin = hbrmesh->GetVertex(index[j]);
-		OsdHbrVertex *destination = hbrmesh->GetVertex(index[(j+1)%num]);
-		OsdHbrHalfEdge *opposite = destination->GetEdge(origin);
-
-		if(origin==NULL || destination==NULL)
-			assert(!"An edge was specified that connected a nonexistent vertex\n");
-
-		if(origin == destination)
-			assert(!"An edge was specified that connected a vertex to itself\n");
-
-		if(opposite && opposite->GetOpposite())
-			assert(!"A non-manifold edge incident to more than 2 faces was found\n");
-
-		if(origin->GetEdge(destination)) {
-			assert(!"An edge connecting two vertices was specified more than once."
-		                "It's likely that an incident face was flipped\n");
-		}
-	}
-#endif
-
-	OsdHbrFace *face = hbrmesh->NewFace(num, index, 0);
-
-	/* this is required for limit eval patch table? */
-	face->SetPtexIndex(num_ptex_faces);
-
-	if(num == 4)
-		num_ptex_faces++;
-	else
-		num_ptex_faces += num;
-}
-
-bool OpenSubdMesh::finish()
-{
-	OsdHbrMesh *hbrmesh = (OsdHbrMesh*)_hbrmesh;
-
-	/* finish hbr mesh construction */
-	hbrmesh->SetInterpolateBoundaryMethod(OsdHbrMesh::k_InterpolateBoundaryEdgeOnly);
-	hbrmesh->Finish();
-
-	return true;
-}
-
-void OpenSubdMesh::tessellate(DiagSplit *split)
-{
-	if(num_ptex_faces == 0)
-		return;
-
-	const int level = 3;
-	const bool requirefvar = false;
-
-	/* convert HRB to FAR mesh */
-	OsdHbrMesh *hbrmesh = (OsdHbrMesh*)_hbrmesh;
-
-	OsdFarMeshFactory meshFactory(hbrmesh, level, true);
-	OsdFarMesh *farmesh = meshFactory.Create(requirefvar);
-	int num_hbr_verts = hbrmesh->GetNumVertices();
-
-	delete hbrmesh;
-	hbrmesh = NULL;
-	_hbrmesh = NULL;
-
-	/* refine HBR mesh with vertex coordinates */
-	OsdCpuComputeController *compute_controller = new OsdCpuComputeController();
-	OsdCpuComputeContext *compute_context = OsdCpuComputeContext::Create(farmesh);
-
-	OsdCpuVertexBuffer *vbuf_base = OsdCpuVertexBuffer::Create(3, num_hbr_verts);
-	vbuf_base->UpdateData(&positions[0], 0, num_verts);
-
-	compute_controller->Refine(compute_context, farmesh->GetKernelBatches(), vbuf_base);
-	compute_controller->Synchronize();
-
-	/* split & dice patches */
-	OpenSubdPatch patch(farmesh, vbuf_base);
-
-	for(int f = 0; f < num_ptex_faces; f++) {
-		patch.face_id = f;
-		split->split_quad(&patch);
-	}
-
-	/* clean up */
-	delete farmesh;
-	delete compute_controller;
-	delete compute_context;
-	delete vbuf_base;
-}
-
-CCL_NAMESPACE_END
-
-#else /* WITH_OPENSUBDIV */
-
-CCL_NAMESPACE_BEGIN
-
-/* Subd Vertex */
-
-class SubdVert
-{
-public:
-	int id;
-	float3 co;
-	
-	explicit SubdVert(int id_)
-	{
-		id = id_;
-		co = make_float3(0.0f, 0.0f, 0.0f);
-	}
-};
-
-/* Subd Face */
-
-class SubdFace
-{
-public:
-	int id;
-	int numverts;
-	int verts[4];
-
-	explicit SubdFace(int id_)
-	{
-		id = id_;
-		numverts = 0;
-	}
-};
-
-/* Subd Mesh */
-
-SubdMesh::SubdMesh()
-{
-}
-
-SubdMesh::~SubdMesh()
-{
-	foreach(SubdVert *vertex, verts)
-		delete vertex;
-	foreach(SubdFace *face, faces)
-		delete face;
-
-	verts.clear();
-	faces.clear();
-}
-
-SubdVert *SubdMesh::add_vert(const float3& co)
-{
-	SubdVert *v = new SubdVert(verts.size());
-	v->co = co;
-	verts.push_back(v);
-
-	return v;
-}
-
-SubdFace *SubdMesh::add_face(int v0, int v1, int v2)
-{
-	int index[3] = {v0, v1, v2};
-	return add_face(index, 3);
-}
-
-SubdFace *SubdMesh::add_face(int v0, int v1, int v2, int v3)
-{
-	int index[4] = {v0, v1, v2, v3};
-	return add_face(index, 4);
-}
-
-SubdFace *SubdMesh::add_face(int *index, int num)
-{
-	/* skip ngons */
-	if(num < 3 || num > 4)
-		return NULL;
-
-	SubdFace *f = new SubdFace(faces.size());
-
-	for(int i = 0; i < num; i++)
-		f->verts[i] = index[i];
-
-	f->numverts = num;
-	faces.push_back(f);
-
-	return f;
-}
-
-bool SubdMesh::finish()
-{
-	return true;
-}
-
-void SubdMesh::tessellate(DiagSplit *split)
-{
-	int num_faces = faces.size();
-		        
-	for(int f = 0; f < num_faces; f++) {
-		SubdFace *face = faces[f];
-		Patch *patch;
-		float3 *hull;
-
-		if(face->numverts == 3) {
-			LinearTrianglePatch *lpatch = new LinearTrianglePatch();
-			hull = lpatch->hull;
-			patch = lpatch;
-		}
-		else if(face->numverts == 4) {
-			LinearQuadPatch *lpatch = new LinearQuadPatch();
-			hull = lpatch->hull;
-			patch = lpatch;
-		}
-		else {
-			assert(0); /* n-gons should have been split already */
-			continue;
-		}
-
-		for(int i = 0; i < face->numverts; i++)
-			hull[i] = verts[face->verts[i]]->co;
-
-		if(face->numverts == 4)
-			swap(hull[2], hull[3]);
-
-		if(patch->is_triangle())
-			split->split_triangle(patch);
-		else
-			split->split_quad(patch);
-
-		delete patch;
-	}
-}
-
-CCL_NAMESPACE_END
-
-#endif /* WITH_OPENSUBDIV */
-
diff --git a/intern/cycles/subd/subd_mesh.h b/intern/cycles/subd/subd_mesh.h
deleted file mode 100644
index f6aefc20318..00000000000
--- a/intern/cycles/subd/subd_mesh.h
+++ /dev/null
@@ -1,78 +0,0 @@
-/*
- * Original code in the public domain -- castanyo@yahoo.es
- *
- * Modifications copyright (c) 2011, Blender Foundation.
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions are
- * met:
- * * Redistributions of source code must retain the above copyright
- *   notice, this list of conditions and the following disclaimer.
- * * Redistributions in binary form must reproduce the above copyright
- *   notice, this list of conditions and the following disclaimer in the
- *   documentation and/or other materials provided with the distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- */
-
-#ifndef __SUBD_MESH_H__
-#define __SUBD_MESH_H__
-
-#include "util_map.h"
-#include "util_types.h"
-#include "util_vector.h"
-
-CCL_NAMESPACE_BEGIN
-
-#ifndef WITH_OPENSUBDIV
-class SubdVert;
-class SubdFace;
-#endif
-
-class DiagSplit;
-class Mesh;
-
-/* Subd Mesh with simple linear subdivision */
-
-class SubdMesh
-{
-public:
-	SubdMesh();
-	~SubdMesh();
-
-	SubdVert *add_vert(const float3& co);
-	
-	SubdFace *add_face(int v0, int v1, int v2);
-	SubdFace *add_face(int v0, int v1, int v2, int v3);
-	SubdFace *add_face(int *index, int num);
-
-	bool finish();
-	void tessellate(DiagSplit *split);
-
-protected:
-#ifdef WITH_OPENSUBDIV
-	void *_hbrmesh;
-	vector<float> positions;
-	int num_verts, num_ptex_faces;
-#else
-	vector<SubdVert*> verts;
-	vector<SubdFace*> faces;
-#endif
-
-};
-
-CCL_NAMESPACE_END
-
-#endif /* __SUBD_MESH_H__ */
-
diff --git a/intern/cycles/subd/subd_patch.cpp b/intern/cycles/subd/subd_patch.cpp
index 60a78016054..d3319c5ccf5 100644
--- a/intern/cycles/subd/subd_patch.cpp
+++ b/intern/cycles/subd/subd_patch.cpp
@@ -84,32 +84,6 @@ BoundBox LinearQuadPatch::bound()
 	return bbox;
 }
 
-/* Linear Triangle Patch */
-
-void LinearTrianglePatch::eval(float3 *P, float3 *dPdu, float3 *dPdv, float3 *N, float u, float v)
-{
-	*P = u*hull[0] + v*hull[1] + (1.0f - u - v)*hull[2];
-
-	if(dPdu && dPdv) {
-		*dPdu = hull[0] - hull[2];
-		*dPdv = hull[1] - hull[2];
-	}
-
-	if(N) {
-		*N = normalize(u*normals[0] + v*normals[1] + (1.0f - u - v)*normals[2]);
-	}
-}
-
-BoundBox LinearTrianglePatch::bound()
-{
-	BoundBox bbox = BoundBox::empty;
-
-	for(int i = 0; i < 3; i++)
-		bbox.grow(hull[i]);
-	
-	return bbox;
-}
-
 /* Bicubic Patch */
 
 void BicubicPatch::eval(float3 *P, float3 *dPdu, float3 *dPdv, float3 *N, float u, float v)
diff --git a/intern/cycles/subd/subd_patch.h b/intern/cycles/subd/subd_patch.h
index bfa04412c66..360c1abf27b 100644
--- a/intern/cycles/subd/subd_patch.h
+++ b/intern/cycles/subd/subd_patch.h
@@ -26,9 +26,11 @@ class Patch {
 public:
 	virtual ~Patch() {}
 	virtual void eval(float3 *P, float3 *dPdu, float3 *dPdv, float3 *N, float u, float v) = 0;
-	virtual bool is_triangle() { return false; }
 	virtual BoundBox bound() = 0;
 	virtual int ptex_face_id() { return -1; }
+
+	int patch_index;
+	int shader;
 };
 
 /* Linear Quad Patch */
@@ -39,19 +41,6 @@ public:
 	float3 normals[4];
 
 	void eval(float3 *P, float3 *dPdu, float3 *dPdv, float3 *N, float u, float v);
-	bool is_triangle() { return false; }
-	BoundBox bound();
-};
-
-/* Linear Triangle Patch */
-
-class LinearTrianglePatch : public Patch {
-public:
-	float3 hull[3];
-	float3 normals[3];
-
-	void eval(float3 *P, float3 *dPdu, float3 *dPdv, float3 *N, float u, float v);
-	bool is_triangle() { return true; }
 	BoundBox bound();
 };
 
@@ -62,7 +51,6 @@ public:
 	float3 hull[16];
 
 	void eval(float3 *P, float3 *dPdu, float3 *dPdv, float3 *N, float u, float v);
-	bool is_triangle() { return false; }
 	BoundBox bound();
 };
 
diff --git a/intern/cycles/subd/subd_split.cpp b/intern/cycles/subd/subd_split.cpp
index c4af8cc8c43..3c91ad8ab0d 100644
--- a/intern/cycles/subd/subd_split.cpp
+++ b/intern/cycles/subd/subd_split.cpp
@@ -40,12 +40,6 @@ void DiagSplit::dispatch(QuadDice::SubPatch& sub, QuadDice::EdgeFactors& ef)
 	edgefactors_quad.push_back(ef);
 }
 
-void DiagSplit::dispatch(TriangleDice::SubPatch& sub, TriangleDice::EdgeFactors& ef)
-{
-	subpatches_triangle.push_back(sub);
-	edgefactors_triangle.push_back(ef);
-}
-
 float3 DiagSplit::to_world(Patch *patch, float2 uv)
 {
 	float3 P;
@@ -112,34 +106,6 @@ void DiagSplit::partition_edge(Patch *patch, float2 *P, int *t0, int *t1, float2
 	}
 }
 
-static float2 right_to_equilateral(float2 P)
-{
-	static const float2 A = make_float2(1.0f, 0.5f);
-	static const float2 B = make_float2(0.0f, sinf(M_PI_F/3.0f));
-	return make_float2(dot(P, A), dot(P, B));
-}
-
-static void limit_edge_factors(const TriangleDice::SubPatch& sub, TriangleDice::EdgeFactors& ef, int max_t)
-{
-	float2 Pu = sub.Pu;
-	float2 Pv = sub.Pv;
-	float2 Pw = sub.Pw;
-
-	if(sub.patch->is_triangle()) {
-		Pu = right_to_equilateral(Pu);
-		Pv = right_to_equilateral(Pv);
-		Pw = right_to_equilateral(Pw);
-	}
-
-	int tu = int(max_t * len(Pw - Pv));
-	int tv = int(max_t * len(Pw - Pu));
-	int tw = int(max_t * len(Pv - Pu));
-
-	ef.tu = tu <= 1 ? 1 : min(ef.tu, tu);
-	ef.tv = tv <= 1 ? 1 : min(ef.tv, tv);
-	ef.tw = tw <= 1 ? 1 : min(ef.tw, tw);
-}
-
 static void limit_edge_factors(const QuadDice::SubPatch& sub, QuadDice::EdgeFactors& ef, int max_t)
 {
 	float2 P00 = sub.P00;
@@ -147,13 +113,6 @@ static void limit_edge_factors(const QuadDice::SubPatch& sub, QuadDice::EdgeFact
 	float2 P10 = sub.P10;
 	float2 P11 = sub.P11;
 
-	if(sub.patch->is_triangle()) {
-		P00 = right_to_equilateral(P00);
-		P01 = right_to_equilateral(P01);
-		P10 = right_to_equilateral(P10);
-		P11 = right_to_equilateral(P11);
-	}
-
 	int tu0 = int(max_t * len(P10 - P00));
 	int tu1 = int(max_t * len(P11 - P01));
 	int tv0 = int(max_t * len(P01 - P00));
@@ -165,84 +124,6 @@ static void limit_edge_factors(const QuadDice::SubPatch& sub, QuadDice::EdgeFact
 	ef.tv1 = tv1 <= 1 ? 1 : min(ef.tv1, tv1);
 }
 
-void DiagSplit::split(TriangleDice::SubPatch& sub, TriangleDice::EdgeFactors& ef, int depth)
-{
-	if(depth > 32) {
-		/* We should never get here, but just in case end recursion safely. */
-		ef.tu = 1;
-		ef.tv = 1;
-		ef.tw = 1;
-
-		dispatch(sub, ef);
-		return;
-	}
-
-	assert(ef.tu == T(sub.patch, sub.Pv, sub.Pw));
-	assert(ef.tv == T(sub.patch, sub.Pw, sub.Pu));
-	assert(ef.tw == T(sub.patch, sub.Pu, sub.Pv));
-
-	int non_uniform_count = int(ef.tu == DSPLIT_NON_UNIFORM) +
-	                        int(ef.tv == DSPLIT_NON_UNIFORM) +
-                            int(ef.tw == DSPLIT_NON_UNIFORM);
-
-	switch(non_uniform_count) {
-		case 1: {
-			/* TODO(mai): one edge is non-uniform, split into two triangles */
-			// fallthru
-		}
-		case 2: {
-			/* TODO(mai): two edges are non-uniform, split into triangle and quad */
-			// fallthru
-		}
-		case 3: {
-			/* all three edges are non-uniform, split into three quads */
-
-			/* partition edges */
-			QuadDice::EdgeFactors ef0, ef1, ef2;
-			float2 Pu, Pv, Pw, Pcenter;
-
-			partition_edge(sub.patch, &Pu, &ef1.tv0, &ef2.tu0, sub.Pw, sub.Pv, ef.tu);
-			partition_edge(sub.patch, &Pv, &ef0.tv0, &ef1.tu0, sub.Pu, sub.Pw, ef.tv);
-			partition_edge(sub.patch, &Pw, &ef2.tv0, &ef0.tu0, sub.Pv, sub.Pu, ef.tw);
-			Pcenter = (Pu + Pv + Pw) * (1.0f / 3.0f);
-
-			/* split */
-			int tsplit01 = T(sub.patch, Pv, Pcenter);
-			int tsplit12 = T(sub.patch, Pu, Pcenter);
-			int tsplit20 = T(sub.patch, Pw, Pcenter);
-
-			ef0.tu1 = tsplit01;
-			ef0.tv1 = tsplit20;
-
-			ef1.tu1 = tsplit12;
-			ef1.tv1 = tsplit01;
-
-			ef2.tu1 = tsplit20;
-			ef2.tv1 = tsplit12;
-
-			/* create subpatches */
-			QuadDice::SubPatch sub0 = {sub.patch, sub.Pu, Pw, Pv, Pcenter};
-			QuadDice::SubPatch sub1 = {sub.patch, sub.Pw, Pv, Pu, Pcenter};
-			QuadDice::SubPatch sub2 = {sub.patch, sub.Pv, Pu, Pw, Pcenter};
-
-			limit_edge_factors(sub0, ef0, 1 << params.max_level);
-			limit_edge_factors(sub1, ef1, 1 << params.max_level);
-			limit_edge_factors(sub2, ef2, 1 << params.max_level);
-
-			split(sub0, ef0, depth+1);
-			split(sub1, ef1, depth+1);
-			split(sub2, ef2, depth+1);
-
-			break;
-		}
-		default: {
-			/* all edges uniform, no splitting needed */
-			dispatch(sub, ef);
-			break;
-		}
-	}
-}
-
 void DiagSplit::split(QuadDice::SubPatch& sub, QuadDice::EdgeFactors& ef, int depth)
 {
 	if(depth > 32) {
@@ -259,6 +140,16 @@ void DiagSplit::split(QuadDice::SubPatch& sub, QuadDice::EdgeFactors& ef, int de
 	bool split_u = (ef.tu0 == DSPLIT_NON_UNIFORM || ef.tu1 == DSPLIT_NON_UNIFORM);
 	bool split_v = (ef.tv0 == DSPLIT_NON_UNIFORM || ef.tv1 == DSPLIT_NON_UNIFORM);
 
+	/* Split subpatches such that the ratio of T for opposite edges doesn't
+     * exceed 1.5, this reduces over tessellation for some patches
+	 */
+	bool tmp_split_v = split_v;
+	if(!split_u && min(ef.tu0, ef.tu1) > 8 && min(ef.tu0, ef.tu1)*1.5f < max(ef.tu0, ef.tu1))
+		split_v = true;
+	if(!tmp_split_v && min(ef.tu0, ef.tu1) > 8 && min(ef.tv0, ef.tv1)*1.5f < max(ef.tv0, ef.tv1))
+		split_u = true;
+
+	/* alternate axis */
 	if(split_u && split_v) {
 		split_u = depth % 2;
 	}
@@ -324,69 +215,21 @@ void DiagSplit::split(QuadDice::SubPatch& sub, QuadDice::EdgeFactors& ef, int de
 	}
 }
 
-void DiagSplit::split_triangle(Patch *patch)
-{
-	TriangleDice::SubPatch sub_split;
-	TriangleDice::EdgeFactors ef_split;
-
-	sub_split.patch = patch;
-	sub_split.Pu = make_float2(1.0f, 0.0f);
-	sub_split.Pv = make_float2(0.0f, 1.0f);
-	sub_split.Pw = make_float2(0.0f, 0.0f);
-
-	ef_split.tu = T(patch, sub_split.Pv, sub_split.Pw);
-	ef_split.tv = T(patch, sub_split.Pw, sub_split.Pu);
-	ef_split.tw = T(patch, sub_split.Pu, sub_split.Pv);
-
-	limit_edge_factors(sub_split, ef_split, 1 << params.max_level);
-
-	split(sub_split, ef_split);
-
-	TriangleDice dice(params);
-
-	for(size_t i = 0; i < subpatches_triangle.size(); i++) {
-		TriangleDice::SubPatch& sub = subpatches_triangle[i];
-		TriangleDice::EdgeFactors& ef = edgefactors_triangle[i];
-
-		ef.tu = max(ef.tu, 1);
-		ef.tv = max(ef.tv, 1);
-		ef.tw = max(ef.tw, 1);
-
-		dice.dice(sub, ef);
-	}
-
-	subpatches_triangle.clear();
-	edgefactors_triangle.clear();
-
-	/* triangle might be split into quads so dice quad subpatches as well */
-	QuadDice qdice(params);
-
-	for(size_t i = 0; i < subpatches_quad.size(); i++) {
-		QuadDice::SubPatch& sub = subpatches_quad[i];
-		QuadDice::EdgeFactors& ef = edgefactors_quad[i];
-
-		ef.tu0 = max(ef.tu0, 1);
-		ef.tu1 = max(ef.tu1, 1);
-		ef.tv0 = max(ef.tv0, 1);
-		ef.tv1 = max(ef.tv1, 1);
-
-		qdice.dice(sub, ef);
-	}
-
-	subpatches_quad.clear();
-	edgefactors_quad.clear();
-}
-
-void DiagSplit::split_quad(Patch *patch)
+void DiagSplit::split_quad(Patch *patch, QuadDice::SubPatch *subpatch)
 {
 	QuadDice::SubPatch sub_split;
 	QuadDice::EdgeFactors ef_split;
 
-	sub_split.patch = patch;
-	sub_split.P00 = make_float2(0.0f, 0.0f);
-	sub_split.P10 = make_float2(1.0f, 0.0f);
-	sub_split.P01 = make_float2(0.0f, 1.0f);
-	sub_split.P11 = make_float2(1.0f, 1.0f);
+	if(subpatch) {
+		sub_split = *subpatch;
+	}
+	else {
+		sub_split.patch = patch;
+		sub_split.P00 = make_float2(0.0f, 0.0f);
+		sub_split.P10 = make_float2(1.0f, 0.0f);
+		sub_split.P01 = make_float2(0.0f, 1.0f);
+		sub_split.P11 = make_float2(1.0f, 1.0f);
+	}
 
 	ef_split.tu0 = T(patch, sub_split.P00, sub_split.P10);
 	ef_split.tu1 = T(patch, sub_split.P01, sub_split.P11);
diff --git a/intern/cycles/subd/subd_split.h b/intern/cycles/subd/subd_split.h
index bbe921f739c..a2f76dd2e03 100644
--- a/intern/cycles/subd/subd_split.h
+++ b/intern/cycles/subd/subd_split.h
@@ -38,8 +38,6 @@ class DiagSplit {
 public:
 	vector<QuadDice::SubPatch> subpatches_quad;
 	vector<QuadDice::EdgeFactors> edgefactors_quad;
-	vector<TriangleDice::SubPatch> subpatches_triangle;
-	vector<TriangleDice::EdgeFactors> edgefactors_triangle;
 
 	SubdParams params;
 
@@ -53,11 +51,7 @@ public:
 	void dispatch(QuadDice::SubPatch& sub, QuadDice::EdgeFactors& ef);
 	void split(QuadDice::SubPatch& sub, QuadDice::EdgeFactors& ef, int depth=0);
 
-	void dispatch(TriangleDice::SubPatch& sub, TriangleDice::EdgeFactors& ef);
-	void split(TriangleDice::SubPatch& sub, TriangleDice::EdgeFactors& ef, int depth=0);
-
-	void split_triangle(Patch *patch);
-	void split_quad(Patch *patch);
+	void split_quad(Patch *patch, QuadDice::SubPatch *subpatch=NULL);
 };
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/util/util_math.h b/intern/cycles/util/util_math.h
index cfe6fa65143..e070b51ae32 100644
--- a/intern/cycles/util/util_math.h
+++ b/intern/cycles/util/util_math.h
@@ -174,6 +174,11 @@ ccl_device_inline float clamp(float a, float mn, float mx)
 	return min(max(a, mn), mx);
 }
 
+ccl_device_inline float mix(float a, float b, float t)
+{
+    return a + t*(b - a);
+}
+
 #endif
 
 #ifndef __KERNEL_CUDA__
@@ -219,6 +224,11 @@ ccl_device_inline float smoothstepf(float f)
 	return (3.0f*ff - 2.0f*ff*f);
 }
 
+ccl_device_inline int mod(int x, int m)
+{
+	return (x % m + m) % m;
+}
+
 /* Float2 Vector */
 
 #ifndef __KERNEL_OPENCL__
@@ -652,6 +662,15 @@ ccl_device_inline float3 interp(float3 a, float3 b, float t)
 	return a + t*(b - a);
 }
 
+#ifndef __KERNEL_OPENCL__
+
+ccl_device_inline float3 mix(float3 a, float3 b, float t)
+{
+	return a + t*(b - a);
+}
+
+#endif
+
 ccl_device_inline bool is_zero(const float3 a)
 {
 #ifdef __KERNEL_SSE__