Subsurf: Implement foreach traversal for subdivsion topology

This makes it more generic process to perform actions which
depend on ptex face + (u, v) and on subdivided vertex index.

Currently it is still just a subdivision calculation process,
but same foreach callbacks can easily be used to propagate
displacement from known vertex locations back to displacement
grids.

1 files changed, 2041 insertions, 0 deletions

diff --git a/source/blender/blenkernel/intern/subdiv_foreach.c b/source/blender/blenkernel/intern/subdiv_foreach.c
new file mode 100644
index 00000000000..b68b65475ac
--- /dev/null
+++ b/source/blender/blenkernel/intern/subdiv_foreach.c
@@ -0,0 +1,2041 @@
+/*
+ * ***** BEGIN GPL LICENSE BLOCK *****
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ *
+ * The Original Code is Copyright (C) 2018 by Blender Foundation.
+ * All rights reserved.
+ *
+ * Contributor(s): Sergey Sharybin.
+ *
+ * ***** END GPL LICENSE BLOCK *****
+ */
+
+/** \file blender/blenkernel/intern/subdiv_foreach.c
+ *  \ingroup bke
+ */
+
+#include "BKE_subdiv.h"
+
+#include "atomic_ops.h"
+
+#include "DNA_mesh_types.h"
+#include "DNA_meshdata_types.h"
+#include "DNA_key_types.h"
+
+#include "BLI_alloca.h"
+#include "BLI_bitmap.h"
+#include "BLI_math_vector.h"
+#include "BLI_task.h"
+
+#include "BKE_mesh.h"
+#include "BKE_key.h"
+
+#include "MEM_guardedalloc.h"
+
+/* =============================================================================
+ * General helpers.
+ */
+
+/* Number of ptex faces for a given polygon. */
+BLI_INLINE int num_ptex_faces_per_poly_get(const MPoly *poly)
+{
+	return (poly->totloop == 4) ? 1 : poly->totloop;
+}
+
+BLI_INLINE int num_edges_per_ptex_face_get(const int resolution)
+{
+	return 2 * (resolution - 1) * resolution;
+}
+
+BLI_INLINE int num_inner_edges_per_ptex_face_get(const int resolution)
+{
+	if (resolution < 2) {
+		return 0;
+	}
+	return (resolution - 2) * resolution +
+	       (resolution - 1) * (resolution - 1);
+}
+
+/* Number of subdivision polygons per ptex face. */
+BLI_INLINE int num_polys_per_ptex_get(const int resolution)
+{
+	return (resolution - 1) * (resolution - 1);
+}
+
+/* Subdivision resolution per given polygon's ptex faces. */
+BLI_INLINE int ptex_face_resolution_get(const MPoly *poly, int resolution)
+{
+	return (poly->totloop == 4) ? (resolution)
+	                            : ((resolution >> 1) + 1);
+}
+
+/* =============================================================================
+ * Context which is passed to all threaded tasks.
+ */
+
+typedef struct SubdivForeachTaskContext {
+	const Mesh *coarse_mesh;
+	const SubdivToMeshSettings *settings;
+	/* Callbacks. */
+	const SubdivForeachContext *foreach_context;
+	/* Counters of geometry in subdivided mesh, initialized as a part of
+	 * offsets calculation.
+	 */
+	int num_subdiv_vertices;
+	int num_subdiv_edges;
+	int num_subdiv_loops;
+	int num_subdiv_polygons;
+	/* Offsets of various geometry in the subdivision mesh arrays. */
+	int vertices_corner_offset;
+	int vertices_edge_offset;
+	int vertices_inner_offset;
+	int edge_boundary_offset;
+	int edge_inner_offset;
+	/* Indexed by coarse polygon index, indicates offset in subdivided mesh
+	 * vertices, edges and polygons arrays, where first element of the poly
+	 * begins.
+	 */
+	int *subdiv_vertex_offset;
+	int *subdiv_edge_offset;
+	int *subdiv_polygon_offset;
+	/* Indexed by base face index, element indicates total number of ptex faces
+	 * created for preceding base faces.
+	 */
+	int *face_ptex_offset;
+	/* Bitmap indicating whether vertex was used already or not.
+	 * - During patch evaluation indicates whether coarse vertex was already
+	 *   evaluated and its position on limit is already known.
+	 */
+	BLI_bitmap *coarse_vertices_used_map;
+	/* Bitmap indicating whether edge was used already or not. This includes:
+	 * - During context initialization it indicates whether subdivided verticies
+	 *   for corresponding edge were already calculated or not.
+	 * - During patch evaluation it indicates whether vertices along this edge
+	 *   were already evaluated.
+	 */
+	BLI_bitmap *coarse_edges_used_map;
+} SubdivForeachTaskContext;
+
+/* NOTE: Expects edge map to be zeroed. */
+static void subdiv_foreach_ctx_count(SubdivForeachTaskContext *ctx)
+{
+	/* Reset counters. */
+	ctx->num_subdiv_vertices = 0;
+	ctx->num_subdiv_edges = 0;
+	ctx->num_subdiv_loops = 0;
+	ctx->num_subdiv_polygons = 0;
+	/* Static geometry counters. */
+	const int resolution = ctx->settings->resolution;
+	const int no_quad_patch_resolution = ((resolution >> 1) + 1);
+	const int num_subdiv_vertices_per_coarse_edge = resolution - 2;
+	const int num_inner_vertices_per_quad = (resolution - 2) * (resolution - 2);
+	const int num_inner_vertices_per_noquad_patch =
+	        (no_quad_patch_resolution - 2) * (no_quad_patch_resolution - 2);
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MLoop *coarse_mloop = coarse_mesh->mloop;
+	const MPoly *coarse_mpoly = coarse_mesh->mpoly;
+	ctx->num_subdiv_vertices = coarse_mesh->totvert;
+	ctx->num_subdiv_edges =
+	        coarse_mesh->totedge * (num_subdiv_vertices_per_coarse_edge + 1);
+	/* Calculate extra vertices and edges createdd by non-loose geometry. */
+	for (int poly_index = 0; poly_index < coarse_mesh->totpoly; poly_index++) {
+		const MPoly *coarse_poly = &coarse_mpoly[poly_index];
+		const int num_ptex_faces_per_poly =
+		        num_ptex_faces_per_poly_get(coarse_poly);
+		for (int corner = 0; corner < coarse_poly->totloop; corner++) {
+			 const MLoop *loop = &coarse_mloop[coarse_poly->loopstart + corner];
+			 const bool is_edge_used =
+			         BLI_BITMAP_TEST_BOOL(ctx->coarse_edges_used_map, loop->e);
+			/* Edges which aren't counted yet. */
+			if (!is_edge_used) {
+				BLI_BITMAP_ENABLE(ctx->coarse_edges_used_map, loop->e);
+				ctx->num_subdiv_vertices += num_subdiv_vertices_per_coarse_edge;
+			}
+		}
+		/* Inner verticies of polygon. */
+		if (num_ptex_faces_per_poly == 1) {
+			ctx->num_subdiv_vertices += num_inner_vertices_per_quad;
+			ctx->num_subdiv_edges +=
+			        num_edges_per_ptex_face_get(resolution - 2) +
+			        4 * num_subdiv_vertices_per_coarse_edge;
+			ctx->num_subdiv_polygons += num_polys_per_ptex_get(resolution);
+		}
+		else {
+			ctx->num_subdiv_vertices +=
+			        1 +
+			        num_ptex_faces_per_poly * (no_quad_patch_resolution - 2) +
+			        num_ptex_faces_per_poly * num_inner_vertices_per_noquad_patch;
+			ctx->num_subdiv_edges +=
+			        num_ptex_faces_per_poly *
+			                (num_inner_edges_per_ptex_face_get(
+			                         no_quad_patch_resolution - 1) +
+			                 (no_quad_patch_resolution - 2) +
+			                 num_subdiv_vertices_per_coarse_edge);
+			if (no_quad_patch_resolution >= 3) {
+				ctx->num_subdiv_edges += coarse_poly->totloop;
+			}
+			ctx->num_subdiv_polygons +=
+			        num_ptex_faces_per_poly *
+			        num_polys_per_ptex_get(no_quad_patch_resolution);
+		}
+	}
+	/* Calculate extra vertices createdd by loose edges. */
+	for (int edge_index = 0; edge_index < coarse_mesh->totedge; edge_index++) {
+		if (!BLI_BITMAP_TEST_BOOL(ctx->coarse_edges_used_map, edge_index)) {
+			ctx->num_subdiv_vertices += num_subdiv_vertices_per_coarse_edge;
+		}
+	}
+	ctx->num_subdiv_loops = ctx->num_subdiv_polygons * 4;
+}
+
+static void subdiv_foreach_ctx_init_offsets(SubdivForeachTaskContext *ctx)
+{
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const int resolution = ctx->settings->resolution;
+	const int resolution_2 = resolution - 2;
+	const int resolution_2_squared = resolution_2 * resolution_2;
+	const int no_quad_patch_resolution = ((resolution >> 1) + 1);
+	const int num_irregular_vertices_per_patch =
+		(no_quad_patch_resolution - 2) * (no_quad_patch_resolution - 1);
+	const int num_subdiv_vertices_per_coarse_edge = resolution - 2;
+	const int num_subdiv_edges_per_coarse_edge = resolution - 1;
+	/* Constant offsets in arrays. */
+	ctx->vertices_corner_offset = 0;
+	ctx->vertices_edge_offset = coarse_mesh->totvert;
+	ctx->vertices_inner_offset =
+	        ctx->vertices_edge_offset +
+	        coarse_mesh->totedge * num_subdiv_vertices_per_coarse_edge;
+	ctx->edge_boundary_offset = 0;
+	ctx->edge_inner_offset =
+	        ctx->edge_boundary_offset +
+	        coarse_mesh->totedge * num_subdiv_edges_per_coarse_edge;
+	/* "Indexed" offsets. */
+	const MPoly *coarse_mpoly = coarse_mesh->mpoly;
+	int vertex_offset = 0;
+	int edge_offset = 0;
+	int polygon_offset = 0;
+	int face_ptex_offset = 0;
+	for (int poly_index = 0; poly_index < coarse_mesh->totpoly; poly_index++) {
+		const MPoly *coarse_poly = &coarse_mpoly[poly_index];
+		const int num_ptex_faces_per_poly =
+		        num_ptex_faces_per_poly_get(coarse_poly);
+		ctx->face_ptex_offset[poly_index] = face_ptex_offset;
+		ctx->subdiv_vertex_offset[poly_index] = vertex_offset;
+		ctx->subdiv_edge_offset[poly_index] = edge_offset;
+		ctx->subdiv_polygon_offset[poly_index] = polygon_offset;
+		face_ptex_offset += num_ptex_faces_per_poly;
+		if (num_ptex_faces_per_poly == 1) {
+			vertex_offset += resolution_2_squared;
+			edge_offset += num_edges_per_ptex_face_get(resolution - 2) +
+			               4 * num_subdiv_vertices_per_coarse_edge;
+			polygon_offset += num_polys_per_ptex_get(resolution);
+		}
+		else {
+			vertex_offset +=
+			        1 +
+			        num_ptex_faces_per_poly * num_irregular_vertices_per_patch;
+			edge_offset +=
+			        num_ptex_faces_per_poly *
+			                (num_inner_edges_per_ptex_face_get(
+			                         no_quad_patch_resolution - 1) +
+			                 (no_quad_patch_resolution - 2) +
+			                 num_subdiv_vertices_per_coarse_edge);
+			if (no_quad_patch_resolution >= 3) {
+				edge_offset += coarse_poly->totloop;
+			}
+			polygon_offset +=
+			        num_ptex_faces_per_poly *
+			        num_polys_per_ptex_get(no_quad_patch_resolution);
+		}
+	}
+}
+
+static void subdiv_foreach_ctx_init(SubdivForeachTaskContext *ctx)
+{
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	/* Allocate maps and offsets. */
+	ctx->coarse_vertices_used_map =
+	        BLI_BITMAP_NEW(coarse_mesh->totvert, "vertices used map");
+	ctx->coarse_edges_used_map =
+	        BLI_BITMAP_NEW(coarse_mesh->totedge, "edges used map");
+	ctx->subdiv_vertex_offset = MEM_malloc_arrayN(
+	        coarse_mesh->totpoly,
+	        sizeof(*ctx->subdiv_vertex_offset),
+	        "vertex_offset");
+	ctx->subdiv_edge_offset = MEM_malloc_arrayN(
+	        coarse_mesh->totpoly,
+	        sizeof(*ctx->subdiv_edge_offset),
+	        "subdiv_edge_offset");
+	ctx->subdiv_polygon_offset = MEM_malloc_arrayN(
+	        coarse_mesh->totpoly,
+	        sizeof(*ctx->subdiv_polygon_offset),
+	        "subdiv_edge_offset");
+	ctx->face_ptex_offset = MEM_malloc_arrayN(coarse_mesh->totpoly,
+	                                          sizeof(*ctx->face_ptex_offset),
+	                                          "face_ptex_offset");
+	/* Initialize all offsets. */
+	subdiv_foreach_ctx_init_offsets(ctx);
+	/* Calculate number of geometry in the result subdivision mesh. */
+	subdiv_foreach_ctx_count(ctx);
+	/* Re-set maps which were used at this step. */
+	BLI_BITMAP_SET_ALL(ctx->coarse_edges_used_map, false, coarse_mesh->totedge);
+}
+
+static void subdiv_foreach_ctx_free(SubdivForeachTaskContext *ctx)
+{
+	MEM_freeN(ctx->coarse_vertices_used_map);
+	MEM_freeN(ctx->coarse_edges_used_map);
+	MEM_freeN(ctx->subdiv_vertex_offset);
+	MEM_freeN(ctx->subdiv_edge_offset);
+	MEM_freeN(ctx->subdiv_polygon_offset);
+	MEM_freeN(ctx->face_ptex_offset);
+}
+
+/* =============================================================================
+ * Vertex traversal process.
+ */
+
+/* Traversal of corner vertices. They are coming from coarse vertices. */
+
+static void subdiv_foreach_corner_vertices_regular_do(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly,
+        SubdivForeachVertexFromCornerCb vertex_corner,
+        bool check_usage)
+{
+	const float weights[4][2] = {{0.0f, 0.0f},
+	                             {1.0f, 0.0f},
+	                             {1.0f, 1.0f},
+	                             {0.0f, 1.0f}};
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MLoop *coarse_mloop = coarse_mesh->mloop;
+	const int coarse_poly_index = coarse_poly - coarse_mesh->mpoly;
+	const int ptex_face_index = ctx->face_ptex_offset[coarse_poly_index];
+	for (int corner = 0; corner < coarse_poly->totloop; corner++) {
+		const MLoop *coarse_loop =
+		    &coarse_mloop[coarse_poly->loopstart + corner];
+		if (check_usage &&
+		    BLI_BITMAP_TEST_AND_SET_ATOMIC(ctx->coarse_vertices_used_map,
+		                                   coarse_loop->v))
+		{
+			continue;
+		}
+		const int coarse_vertex_index = coarse_loop->v;
+		const int subdiv_vertex_index = 
+		        ctx->vertices_corner_offset + coarse_vertex_index;
+		const float u = weights[corner][0];
+		const float v = weights[corner][1];
+		vertex_corner(
+		        ctx->foreach_context,
+		        tls,
+		        ptex_face_index,
+		        u, v,
+		        coarse_vertex_index,
+		        coarse_poly_index,
+		        0,
+		        subdiv_vertex_index);
+	}
+}
+
+static void subdiv_foreach_corner_vertices_regular(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly)
+{
+	subdiv_foreach_corner_vertices_regular_do(
+	        ctx, tls, coarse_poly, ctx->foreach_context->vertex_corner, true);
+}
+
+static void subdiv_foreach_corner_vertices_special_do(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly,
+        SubdivForeachVertexFromCornerCb vertex_corner,
+        bool check_usage)
+{
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MLoop *coarse_mloop = coarse_mesh->mloop;
+	const int coarse_poly_index = coarse_poly - coarse_mesh->mpoly;
+	int ptex_face_index = ctx->face_ptex_offset[coarse_poly_index];
+	for (int corner = 0;
+	     corner < coarse_poly->totloop;
+	     corner++, ptex_face_index++)
+	{
+		const MLoop *coarse_loop =
+		    &coarse_mloop[coarse_poly->loopstart + corner];
+		if (check_usage &&
+		    BLI_BITMAP_TEST_AND_SET_ATOMIC(ctx->coarse_vertices_used_map,
+		                                   coarse_loop->v))
+		{
+			continue;
+		}
+		const int coarse_vertex_index = coarse_loop->v;
+		const int subdiv_vertex_index =
+		        ctx->vertices_corner_offset + coarse_vertex_index;
+		vertex_corner(
+		        ctx->foreach_context,
+		        tls,
+		        ptex_face_index,
+		        0.0f, 0.0f,
+		        coarse_vertex_index,
+		        coarse_poly_index,
+		        corner,
+		        subdiv_vertex_index);
+	}
+}
+
+static void subdiv_foreach_corner_vertices_special(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly)
+{
+	subdiv_foreach_corner_vertices_special_do(
+	        ctx, tls, coarse_poly, ctx->foreach_context->vertex_corner, true);
+}
+
+static void subdiv_foreach_corner_vertices(SubdivForeachTaskContext *ctx,
+                                           void *tls,
+                                           const MPoly *coarse_poly)
+{
+	if (coarse_poly->totloop == 4) {
+		subdiv_foreach_corner_vertices_regular(ctx, tls, coarse_poly);
+	}
+	else {
+		subdiv_foreach_corner_vertices_special(ctx, tls, coarse_poly);
+	}
+}
+
+static void subdiv_foreach_every_corner_vertices_regular(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly)
+{
+	subdiv_foreach_corner_vertices_regular_do(
+	        ctx, tls, coarse_poly,
+	        ctx->foreach_context->vertex_every_corner,
+	        false);
+}
+
+static void subdiv_foreach_every_corner_vertices_special(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly)
+{
+	subdiv_foreach_corner_vertices_special_do(
+	        ctx, tls, coarse_poly,
+	        ctx->foreach_context->vertex_every_corner,
+	        false);
+}
+
+static void subdiv_foreach_every_corner_vertices(SubdivForeachTaskContext *ctx)
+{
+	if (ctx->foreach_context->vertex_every_corner == NULL) {
+		return;
+	}
+	const SubdivForeachContext *foreach_context = ctx->foreach_context;
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MPoly *coarse_mpoly = coarse_mesh->mpoly;
+	void *tls = NULL;
+	if (foreach_context->user_data_tls_size != 0) {
+		tls = MEM_mallocN(foreach_context->user_data_tls_size, "tls");
+		memcpy(tls,
+		       foreach_context->user_data_tls,
+		       foreach_context->user_data_tls_size);
+	}
+	for (int poly_index = 0; poly_index < coarse_mesh->totpoly; poly_index++) {
+		const MPoly *coarse_poly = &coarse_mpoly[poly_index];
+		if (coarse_poly->totloop == 4) {
+			subdiv_foreach_every_corner_vertices_regular(ctx, tls, coarse_poly);
+		}
+		else {
+			subdiv_foreach_every_corner_vertices_special(ctx, tls, coarse_poly);
+		}
+	}
+	if (tls != NULL) {
+		MEM_freeN(tls);
+	}
+}
+
+/* Traverse of edge vertices. They are coming from coarse edges. */
+
+static void subdiv_foreach_edge_vertices_regular_do(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly,
+        SubdivForeachVertexFromEdgeCb vertex_edge,
+        bool check_usage)
+{
+	const int resolution = ctx->settings->resolution;
+	const int resolution_1 = resolution - 1;
+	const float inv_resolution_1 = 1.0f / (float)resolution_1;
+	const int num_subdiv_vertices_per_coarse_edge = resolution - 2;
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MEdge *coarse_medge = coarse_mesh->medge;
+	const MLoop *coarse_mloop = coarse_mesh->mloop;
+	const MPoly *coarse_mpoly = coarse_mesh->mpoly;
+	const int coarse_poly_index = coarse_poly - coarse_mpoly;
+	const int poly_index = coarse_poly - coarse_mesh->mpoly;
+	const int ptex_face_index = ctx->face_ptex_offset[poly_index];
+	for (int corner = 0; corner < coarse_poly->totloop; corner++) {
+		const MLoop *coarse_loop =
+		    &coarse_mloop[coarse_poly->loopstart + corner];
+		const int coarse_edge_index = coarse_loop->e;
+		if (check_usage &&
+		    BLI_BITMAP_TEST_AND_SET_ATOMIC(ctx->coarse_edges_used_map,
+		                                   coarse_edge_index))
+		{
+			continue;
+		}
+		const MEdge *coarse_edge = &coarse_medge[coarse_edge_index];
+		const bool flip = (coarse_edge->v2 == coarse_loop->v);
+		int subdiv_vertex_index =
+		        ctx->vertices_edge_offset +
+		        coarse_edge_index * num_subdiv_vertices_per_coarse_edge;
+		for (int vertex_index = 0;
+		     vertex_index < num_subdiv_vertices_per_coarse_edge;
+		     vertex_index++, subdiv_vertex_index++)
+		{
+			float fac = (vertex_index + 1) * inv_resolution_1;
+			if (flip) {
+				fac = 1.0f - fac;
+			}
+			if (corner >= 2) {
+				fac = 1.0f - fac;
+			}
+			float u, v;
+			if ((corner & 1) == 0) {
+				u = fac;
+				v = (corner == 2) ? 1.0f : 0.0f;
+			}
+			else {
+				u = (corner == 1) ? 1.0f : 0.0f;
+				v = fac;
+			}
+			vertex_edge(
+			        ctx->foreach_context,
+			        tls,
+			        ptex_face_index,
+			        u, v,
+			        coarse_edge_index,
+			        coarse_poly_index,
+			        0,
+			        subdiv_vertex_index);
+		}
+	}
+}
+
+static void subdiv_foreach_edge_vertices_regular(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly)
+{
+	subdiv_foreach_edge_vertices_regular_do(
+	        ctx, tls, coarse_poly,
+	        ctx->foreach_context->vertex_edge,
+	        true);
+}
+
+static void subdiv_foreach_edge_vertices_special_do(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly,
+        SubdivForeachVertexFromEdgeCb vertex_edge,
+        bool check_usage)
+{
+	const int resolution = ctx->settings->resolution;
+	const int num_subdiv_vertices_per_coarse_edge = resolution - 2;
+	const int num_vertices_per_ptex_edge = ((resolution >> 1) + 1);
+	const float inv_ptex_resolution_1 =
+	        1.0f / (float)(num_vertices_per_ptex_edge - 1);
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MEdge *coarse_medge = coarse_mesh->medge;
+	const MLoop *coarse_mloop = coarse_mesh->mloop;
+	const MPoly *coarse_mpoly = coarse_mesh->mpoly;
+	const int coarse_poly_index = coarse_poly - coarse_mpoly;
+	const int poly_index = coarse_poly - coarse_mesh->mpoly;
+	const int ptex_face_start_index = ctx->face_ptex_offset[poly_index];
+	int ptex_face_index = ptex_face_start_index;
+	for (int corner = 0;
+	     corner < coarse_poly->totloop;
+	     corner++, ptex_face_index++)
+	{
+		const MLoop *coarse_loop =
+		        &coarse_mloop[coarse_poly->loopstart + corner];
+		const int coarse_edge_index = coarse_loop->e;
+		if (check_usage &&
+		    BLI_BITMAP_TEST_AND_SET_ATOMIC(ctx->coarse_edges_used_map,
+		                                   coarse_edge_index))
+		{
+			continue;
+		}
+		const MEdge *coarse_edge = &coarse_medge[coarse_edge_index];
+		const bool flip = (coarse_edge->v2 == coarse_loop->v);
+		int subdiv_vertex_index =
+		        ctx->vertices_edge_offset +
+		        coarse_edge_index * num_subdiv_vertices_per_coarse_edge;
+		int veretx_delta = 1;
+		if (flip) {
+			subdiv_vertex_index += num_subdiv_vertices_per_coarse_edge - 1;
+			veretx_delta = -1;
+		}
+		for (int vertex_index = 1;
+		     vertex_index < num_vertices_per_ptex_edge;
+		     vertex_index++, subdiv_vertex_index += veretx_delta)
+		{
+			const float u = vertex_index * inv_ptex_resolution_1;
+			vertex_edge(
+			        ctx->foreach_context,
+			        tls,
+			        ptex_face_index,
+			        u, 0.0f,
+			        coarse_edge_index,
+			        coarse_poly_index,
+			        corner,
+			        subdiv_vertex_index);
+		}
+		const int next_ptex_face_index =
+		        ptex_face_start_index + (corner + 1) % coarse_poly->totloop;
+		for (int vertex_index = 1;
+		     vertex_index < num_vertices_per_ptex_edge - 1;
+		     vertex_index++, subdiv_vertex_index += veretx_delta)
+		{
+			const float v = 1.0f - vertex_index * inv_ptex_resolution_1;
+			vertex_edge(
+			        ctx->foreach_context,
+			        tls,
+			        next_ptex_face_index,
+			        0.0f, v,
+			        coarse_edge_index,
+			        coarse_poly_index,
+			        corner,
+			        subdiv_vertex_index);
+		}
+	}
+}
+
+static void subdiv_foreach_edge_vertices_special(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly)
+{
+	subdiv_foreach_edge_vertices_special_do(
+	        ctx, tls, coarse_poly,
+	        ctx->foreach_context->vertex_edge,
+	        true);
+}
+
+static void subdiv_foreach_edge_vertices(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly)
+{
+	if (coarse_poly->totloop == 4) {
+		subdiv_foreach_edge_vertices_regular(ctx, tls, coarse_poly);
+	}
+	else {
+		subdiv_foreach_edge_vertices_special(ctx, tls, coarse_poly);
+	}
+}
+
+static void subdiv_foreach_every_edge_vertices_regular(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly)
+{
+	subdiv_foreach_edge_vertices_regular_do(
+	        ctx, tls, coarse_poly,
+	        ctx->foreach_context->vertex_every_edge,
+	        false);
+}
+
+static void subdiv_foreach_every_edge_vertices_special(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly)
+{
+	subdiv_foreach_edge_vertices_special_do(
+	        ctx, tls, coarse_poly,
+	        ctx->foreach_context->vertex_every_edge,
+	        false);
+}
+
+static void subdiv_foreach_every_edge_vertices(SubdivForeachTaskContext *ctx)
+{
+	if (ctx->foreach_context->vertex_every_edge == NULL) {
+		return;
+	}
+	const SubdivForeachContext *foreach_context = ctx->foreach_context;
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MPoly *coarse_mpoly = coarse_mesh->mpoly;
+	void *tls = NULL;
+	if (foreach_context->user_data_tls_size != 0) {
+		tls = MEM_mallocN(foreach_context->user_data_tls_size, "tls");
+		memcpy(tls,
+		       foreach_context->user_data_tls,
+		       foreach_context->user_data_tls_size);
+	}
+	for (int poly_index = 0; poly_index < coarse_mesh->totpoly; poly_index++) {
+		const MPoly *coarse_poly = &coarse_mpoly[poly_index];
+		if (coarse_poly->totloop == 4) {
+			subdiv_foreach_every_edge_vertices_regular(ctx, tls, coarse_poly);
+		}
+		else {
+			subdiv_foreach_every_edge_vertices_special(ctx, tls, coarse_poly);
+		}
+	}
+	if (tls != NULL) {
+		MEM_freeN(tls);
+	}
+}
+
+/* Traversal of inner vertices, they are coming from ptex patches. */
+
+static void subdiv_foreach_inner_vertices_regular(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly)
+{
+	const int resolution = ctx->settings->resolution;
+	const float inv_resolution_1 = 1.0f / (float)(resolution - 1);
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const int coarse_poly_index = coarse_poly - coarse_mesh->mpoly;
+	const int ptex_face_index = ctx->face_ptex_offset[coarse_poly_index];
+	const int start_vertex_index = ctx->subdiv_vertex_offset[coarse_poly_index];
+	int subdiv_vertex_index =
+	        ctx->vertices_inner_offset + start_vertex_index;
+	for (int y = 1; y < resolution - 1; y++) {
+		const float v = y * inv_resolution_1;
+		for (int x = 1; x < resolution - 1; x++, subdiv_vertex_index++) {
+			const float u = x * inv_resolution_1;
+			ctx->foreach_context->vertex_inner(
+			        ctx->foreach_context,
+			        tls,
+			        ptex_face_index,
+			        u, v,
+			        coarse_poly_index, 0,
+			        subdiv_vertex_index);
+		}
+	}
+}
+
+static void subdiv_foreach_inner_vertices_special(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly)
+{
+	const int resolution = ctx->settings->resolution;
+	const int ptex_face_resolution = ptex_face_resolution_get(
+	        coarse_poly, resolution);
+	const float inv_ptex_face_resolution_1 =
+	        1.0f / (float)(ptex_face_resolution - 1);
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const int coarse_poly_index = coarse_poly - coarse_mesh->mpoly;
+	int ptex_face_index = ctx->face_ptex_offset[coarse_poly_index];
+	const int start_vertex_index = ctx->subdiv_vertex_offset[coarse_poly_index];
+	int subdiv_vertex_index =
+	        ctx->vertices_inner_offset + start_vertex_index;
+	ctx->foreach_context->vertex_inner(
+			ctx->foreach_context,
+	        tls,
+			ptex_face_index,
+			1.0f, 1.0f,
+			coarse_poly_index, 0,
+			subdiv_vertex_index);
+	subdiv_vertex_index++;
+	for (int corner = 0;
+	     corner < coarse_poly->totloop;
+	     corner++, ptex_face_index++)
+	{
+		for (int y = 1; y < ptex_face_resolution - 1; y++) {
+			const float v = y * inv_ptex_face_resolution_1;
+			for (int x = 1;
+			     x < ptex_face_resolution; x++,
+			     subdiv_vertex_index++)
+			{
+				const float u = x * inv_ptex_face_resolution_1;
+				ctx->foreach_context->vertex_inner(
+				        ctx->foreach_context,
+				        tls,
+				        ptex_face_index,
+				        u, v,
+				        coarse_poly_index,  corner,
+				        subdiv_vertex_index);
+			}
+		}
+	}
+}
+
+static void subdiv_foreach_inner_vertices(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly)
+{
+	if (coarse_poly->totloop == 4) {
+		subdiv_foreach_inner_vertices_regular(ctx, tls, coarse_poly);
+	}
+	else {
+		subdiv_foreach_inner_vertices_special(ctx, tls, coarse_poly);
+	}
+}
+
+/* Traverse all vertices which are emitted from given coarse polygon. */
+static void subdiv_foreach_vertices(SubdivForeachTaskContext *ctx,
+                                    void *tls,
+                                    const int poly_index)
+{
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MPoly *coarse_mpoly = coarse_mesh->mpoly;
+	const MPoly *coarse_poly = &coarse_mpoly[poly_index];
+	if (ctx->foreach_context->vertex_corner != NULL) {
+		subdiv_foreach_corner_vertices(ctx, tls, coarse_poly);
+	}
+	if (ctx->foreach_context->vertex_edge != NULL) {
+		subdiv_foreach_edge_vertices(ctx, tls, coarse_poly);
+	}
+	if (ctx->foreach_context->vertex_inner != NULL) {
+		subdiv_foreach_inner_vertices(ctx, tls, coarse_poly);
+	}
+}
+
+/* =============================================================================
+ * Edge traversal process.
+ */
+
+/* TODO(sergey): Coarse edge are always NONE, consider getting rid of it. */
+static int subdiv_foreach_edges_row(SubdivForeachTaskContext *ctx,
+                                    void *tls,
+                                    const int coarse_edge_index,
+                                    const int start_subdiv_edge_index,
+                                    const int start_vertex_index,
+                                    const int num_edges_per_row)
+{
+	int subdiv_edge_index = start_subdiv_edge_index;
+	int vertex_index = start_vertex_index;
+	for (int edge_index = 0;
+	     edge_index < num_edges_per_row - 1;
+	     edge_index++, subdiv_edge_index++)
+	{
+		const int v1 = vertex_index;
+		const int v2 = vertex_index + 1;
+		ctx->foreach_context->edge(
+		        ctx->foreach_context,
+		        tls,
+		        coarse_edge_index,
+		        subdiv_edge_index,
+		        v1, v2);
+		vertex_index += 1;
+	}
+	return subdiv_edge_index;
+}
+
+/* TODO(sergey): Coarse edges are always NONE, consider getting rid of them. */
+static int subdiv_foreach_edges_column(SubdivForeachTaskContext *ctx,
+                                       void *tls,
+                                       const int coarse_start_edge_index,
+                                       const int coarse_end_edge_index,
+                                       const int start_subdiv_edge_index,
+                                       const int start_vertex_index,
+                                       const int num_edges_per_row)
+{
+	int subdiv_edge_index = start_subdiv_edge_index;
+	int vertex_index = start_vertex_index;
+	for (int edge_index = 0;
+	     edge_index < num_edges_per_row;
+	     edge_index++, subdiv_edge_index++)
+	{
+		int coarse_edge_index = ORIGINDEX_NONE;
+		if (edge_index == 0) {
+			coarse_edge_index = coarse_start_edge_index;
+		}
+		else if (edge_index == num_edges_per_row - 1) {
+			coarse_edge_index = coarse_end_edge_index;
+		}
+		const int v1 = vertex_index;
+		const int v2 = vertex_index + num_edges_per_row;
+		ctx->foreach_context->edge(
+		        ctx->foreach_context,
+		        tls,
+		        coarse_edge_index,
+		        subdiv_edge_index,
+		        v1, v2);
+		vertex_index += 1;
+	}
+	return subdiv_edge_index;
+}
+
+/* Defines edges between inner vertices of patch, and also edges to the
+ * boundary.
+ */
+
+/* Consider a subdivision of base face at level 1:
+ *
+ *  y
+ *  ^
+ *  |   (6) ---- (7) ---- (8)
+ *  |    |        |        |
+ *  |   (3) ---- (4) ---- (5)
+ *  |    |        |        |
+ *  |   (0) ---- (1) ---- (2)
+ *  o---------------------------> x
+ *
+ * This is illustrate which parts of geometry is created by code below.
+ */
+
+static void subdiv_foreach_edges_all_patches_regular(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly)
+{
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MEdge *coarse_medge = coarse_mesh->medge;
+	const MLoop *coarse_mloop = coarse_mesh->mloop;
+	const MPoly *coarse_mpoly = coarse_mesh->mpoly;
+	const int poly_index = coarse_poly - coarse_mpoly;
+	const int resolution = ctx->settings->resolution;
+	const int start_vertex_index =
+	        ctx->vertices_inner_offset +
+	        ctx->subdiv_vertex_offset[poly_index];
+	const int num_subdiv_vertices_per_coarse_edge = resolution - 2;
+	int subdiv_edge_index =
+	        ctx->edge_inner_offset + ctx->subdiv_edge_offset[poly_index];
+	/* Traverse bottom row of edges (0-1, 1-2). */
+	subdiv_edge_index = subdiv_foreach_edges_row(
+	        ctx,
+	        tls,
+	        ORIGINDEX_NONE,
+	        subdiv_edge_index,
+	        start_vertex_index,
+	        resolution - 2);
+	/* Traverse remaining edges. */
+	for (int row = 0; row < resolution - 3; row++) {
+		const int start_row_vertex_index =
+		        start_vertex_index + row * (resolution - 2);
+		/* Traverse vertical columns.
+		 *
+		 * At first iteration it will be edges (0-3. 1-4, 2-5), then it
+		 * will be (3-6, 4-7, 5-8) and so on.
+		 */
+		subdiv_edge_index = subdiv_foreach_edges_column(
+		        ctx,
+		        tls,
+		        ORIGINDEX_NONE,
+		        ORIGINDEX_NONE,
+		        subdiv_edge_index,
+		        start_row_vertex_index,
+		        resolution - 2);
+		/* Create horizontal edge row.
+		 *
+		 * At first iteration it will be edges (3-4, 4-5), then it will be
+		 * (6-7, 7-8) and so on.
+		 */
+		subdiv_edge_index = subdiv_foreach_edges_row(
+		        ctx,
+		        tls,
+		        ORIGINDEX_NONE,
+		        subdiv_edge_index,
+		        start_row_vertex_index + resolution - 2,
+		        resolution - 2);
+	}
+	/* Connect inner part of patch to boundary. */
+	for (int corner = 0; corner < coarse_poly->totloop; corner++) {
+		const MLoop *coarse_loop =
+		        &coarse_mloop[coarse_poly->loopstart + corner];
+		const MEdge *coarse_edge = &coarse_medge[coarse_loop->e];
+		const int start_edge_vertex = ctx->vertices_edge_offset +
+		        coarse_loop->e * num_subdiv_vertices_per_coarse_edge;
+		const bool flip = (coarse_edge->v2 == coarse_loop->v);
+		int side_start_index = start_vertex_index;
+		int side_stride = 0;
+		/* Calculate starting veretx of corresponding inner part of ptex. */
+		if (corner == 0) {
+			side_stride = 1;
+		}
+		else if (corner == 1) {
+			side_start_index += resolution - 3;
+			side_stride = resolution - 2;
+		}
+		else if (corner == 2) {
+			side_start_index += num_subdiv_vertices_per_coarse_edge *
+			                    num_subdiv_vertices_per_coarse_edge - 1;
+			side_stride = -1;
+		}
+		else if (corner == 3) {
+			side_start_index += num_subdiv_vertices_per_coarse_edge *
+			                    (num_subdiv_vertices_per_coarse_edge - 1);
+			side_stride = -(resolution - 2);
+		}
+		for (int i = 0; i < resolution - 2; i++, subdiv_edge_index++) {
+			const int v1 = (flip)
+			                 ? (start_edge_vertex + (resolution - i - 3))
+			                 : (start_edge_vertex + i);
+			const int v2 = side_start_index + side_stride * i;
+			ctx->foreach_context->edge(
+			        ctx->foreach_context,
+			        tls,
+			        ORIGINDEX_NONE,
+			        subdiv_edge_index,
+			        v1, v2);
+		}
+	}
+}
+
+static void subdiv_foreach_edges_all_patches_special(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly)
+{
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MEdge *coarse_medge = coarse_mesh->medge;
+	const MLoop *coarse_mloop = coarse_mesh->mloop;
+	const MPoly *coarse_mpoly = coarse_mesh->mpoly;
+	const int poly_index = coarse_poly - coarse_mpoly;
+	const int resolution = ctx->settings->resolution;
+	const int ptex_face_resolution =
+	        ptex_face_resolution_get(coarse_poly, resolution);
+	const int ptex_face_inner_resolution = ptex_face_resolution - 2;
+	const int num_inner_vertices_per_ptex =
+	        (ptex_face_resolution - 1) * (ptex_face_resolution - 2);
+	const int num_subdiv_vertices_per_coarse_edge = resolution - 2;
+	const int center_vertex_index =
+	        ctx->vertices_inner_offset +
+	        ctx->subdiv_vertex_offset[poly_index];
+	const int start_vertex_index = center_vertex_index + 1;
+	int subdiv_edge_index =
+	        ctx->edge_inner_offset + ctx->subdiv_edge_offset[poly_index];
+	/* Traverse inner ptex edges. */
+	for (int corner = 0; corner < coarse_poly->totloop; corner++) {
+		const int start_ptex_face_vertex_index =
+		        start_vertex_index + corner * num_inner_vertices_per_ptex;
+		/* Similar steps to regular patch case. */
+		subdiv_edge_index = subdiv_foreach_edges_row(
+		        ctx,
+		        tls,
+		        ORIGINDEX_NONE,
+		        subdiv_edge_index,
+		        start_ptex_face_vertex_index,
+		        ptex_face_inner_resolution + 1);
+		for (int row = 0; row < ptex_face_inner_resolution - 1; row++) {
+			const int start_row_vertex_index =
+			        start_ptex_face_vertex_index +
+			        row * (ptex_face_inner_resolution + 1);
+			subdiv_edge_index = subdiv_foreach_edges_column(
+			        ctx,
+			        tls,
+			        ORIGINDEX_NONE,
+			        ORIGINDEX_NONE,
+			        subdiv_edge_index,
+			        start_row_vertex_index,
+			        ptex_face_inner_resolution + 1);
+			subdiv_edge_index = subdiv_foreach_edges_row(
+			        ctx,
+			        tls,
+			        ORIGINDEX_NONE,
+			        subdiv_edge_index,
+			        start_row_vertex_index + ptex_face_inner_resolution + 1,
+			        ptex_face_inner_resolution + 1);
+		}
+	}
+	/* Create connections between ptex faces. */
+	for (int corner = 0; corner < coarse_poly->totloop; corner++) {
+		const int next_corner = (corner + 1) % coarse_poly->totloop;
+		int current_patch_vertex_index =
+		        start_vertex_index + corner * num_inner_vertices_per_ptex +
+		        ptex_face_inner_resolution;
+		int next_path_vertex_index =
+		        start_vertex_index + next_corner * num_inner_vertices_per_ptex +
+		        num_inner_vertices_per_ptex - ptex_face_resolution + 1;
+		for (int row = 0;
+		     row < ptex_face_inner_resolution;
+		     row++, subdiv_edge_index++)
+		{
+			const int v1 = current_patch_vertex_index;
+			const int v2 = next_path_vertex_index;
+			ctx->foreach_context->edge(
+			        ctx->foreach_context,
+			        tls,
+			        ORIGINDEX_NONE,
+			        subdiv_edge_index,
+			        v1, v2);
+			current_patch_vertex_index += ptex_face_inner_resolution + 1;
+			next_path_vertex_index += 1;
+		}
+	}
+	/* Create edges from center. */
+	if (ptex_face_resolution >= 3) {
+		for (int corner = 0;
+		     corner < coarse_poly->totloop;
+		     corner++, subdiv_edge_index++)
+		{
+			const int current_patch_end_vertex_index =
+			        start_vertex_index + corner * num_inner_vertices_per_ptex +
+			        num_inner_vertices_per_ptex - 1;
+			const int v1 = center_vertex_index;
+			const int v2 = current_patch_end_vertex_index;
+			ctx->foreach_context->edge(
+			        ctx->foreach_context,
+			        tls,
+			        ORIGINDEX_NONE,
+			        subdiv_edge_index,
+			        v1, v2);
+		}
+	}
+	/* Connect inner path of patch to boundary. */
+	const MLoop *prev_coarse_loop =
+	        &coarse_mloop[coarse_poly->loopstart + coarse_poly->totloop - 1];
+	for (int corner = 0; corner < coarse_poly->totloop; corner++) {
+		const MLoop *coarse_loop =
+		        &coarse_mloop[coarse_poly->loopstart + corner];
+		{
+			const MEdge *coarse_edge = &coarse_medge[coarse_loop->e];
+			const int start_edge_vertex = ctx->vertices_edge_offset +
+			        coarse_loop->e * num_subdiv_vertices_per_coarse_edge;
+			const bool flip = (coarse_edge->v2 == coarse_loop->v);
+			int side_start_index;
+			if (ptex_face_resolution >= 3) {
+				side_start_index =
+				        start_vertex_index + num_inner_vertices_per_ptex * corner;
+			}
+			else {
+				side_start_index = center_vertex_index;
+			}
+			for (int i = 0; i < ptex_face_resolution - 1;
+			     i++,
+			     subdiv_edge_index++)
+			{
+				const int v1 = (flip)
+				                   ? (start_edge_vertex + (resolution - i - 3))
+				                   : (start_edge_vertex + i);
+				const int v2 = side_start_index + i;
+				ctx->foreach_context->edge(
+				        ctx->foreach_context,
+				        tls,
+				        ORIGINDEX_NONE,
+				        subdiv_edge_index,
+				        v1, v2);
+			}
+		}
+		if (ptex_face_resolution >= 3) {
+			const MEdge *coarse_edge = &coarse_medge[prev_coarse_loop->e];
+			const int start_edge_vertex = ctx->vertices_edge_offset +
+			        prev_coarse_loop->e * num_subdiv_vertices_per_coarse_edge;
+			const bool flip = (coarse_edge->v2 == coarse_loop->v);
+			int side_start_index =
+			        start_vertex_index + num_inner_vertices_per_ptex * corner;
+			for (int i = 0; i < ptex_face_resolution - 2;
+			     i++,
+				 subdiv_edge_index++)
+			{
+				const int v1 = (flip)
+				                   ? (start_edge_vertex + (resolution - i - 3))
+				                   : (start_edge_vertex + i);
+				const int v2 = side_start_index +
+				                  (ptex_face_inner_resolution + 1) * i;
+				ctx->foreach_context->edge(
+				        ctx->foreach_context,
+				        tls,
+				        ORIGINDEX_NONE,
+				        subdiv_edge_index,
+				        v1, v2);
+			}
+		}
+		prev_coarse_loop = coarse_loop;
+	}
+}
+
+static void subdiv_foreach_edges_all_patches(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        const MPoly *coarse_poly)
+{
+	if (coarse_poly->totloop == 4) {
+		subdiv_foreach_edges_all_patches_regular(ctx, tls, coarse_poly);
+	}
+	else {
+		subdiv_foreach_edges_all_patches_special(ctx, tls, coarse_poly);
+	}
+}
+
+static void subdiv_foreach_edges(SubdivForeachTaskContext *ctx,
+                                 void *tls,
+                                 int poly_index)
+{
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MPoly *coarse_mpoly = coarse_mesh->mpoly;
+	const MPoly *coarse_poly = &coarse_mpoly[poly_index];
+	subdiv_foreach_edges_all_patches(ctx, tls, coarse_poly);
+}
+
+static void subdiv_foreach_boundary_edges(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        int coarse_edge_index)
+{
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MEdge *coarse_medge = coarse_mesh->medge;
+	const MEdge *coarse_edge = &coarse_medge[coarse_edge_index];
+	const int resolution = ctx->settings->resolution;
+	const int num_subdiv_vertices_per_coarse_edge = resolution - 2;
+	const int num_subdiv_edges_per_coarse_edge = resolution - 1;
+	int subdiv_edge_index =
+	        ctx->edge_boundary_offset +
+	        coarse_edge_index * num_subdiv_edges_per_coarse_edge;
+	int last_vertex_index = ctx->vertices_corner_offset + coarse_edge->v1;
+	for (int i = 0;
+	     i < num_subdiv_edges_per_coarse_edge - 1;
+	     i++, subdiv_edge_index++)
+	{
+		const int v1 = last_vertex_index;
+		const int v2 =
+		        ctx->vertices_edge_offset +
+		        coarse_edge_index * num_subdiv_vertices_per_coarse_edge +
+		        i;
+		ctx->foreach_context->edge(
+		        ctx->foreach_context,
+		        tls,
+		        coarse_edge_index,
+		        subdiv_edge_index,
+		        v1, v2);
+		last_vertex_index = v2;
+	}
+	const int v1 = last_vertex_index;
+	const int v2 = ctx->vertices_corner_offset + coarse_edge->v2;
+	ctx->foreach_context->edge(
+	        ctx->foreach_context,
+	        tls,
+	        coarse_edge_index,
+	        subdiv_edge_index,
+	        v1, v2);
+}
+
+/* =============================================================================
+ * Loops traversal.
+ */
+
+static void rotate_indices(const int rot, int *a, int *b, int *c, int *d)
+{
+	int values[4] = {*a, *b, *c, *d};
+	*a = values[(0 - rot + 4) % 4];
+	*b = values[(1 - rot + 4) % 4];
+	*c = values[(2 - rot + 4) % 4];
+	*d = values[(3 - rot + 4) % 4];
+}
+
+static void subdiv_foreach_loops_of_poly(
+        SubdivForeachTaskContext *ctx,
+        void *tls,
+        int subdiv_loop_start_index,
+        const int ptex_face_index,
+        const int coarse_poly_index,
+		const int coarse_corner_index,
+        const int rotation,
+        /*const*/ int v0, /*const*/ int e0,
+        /*const*/ int v1, /*const*/ int e1,
+        /*const*/ int v2, /*const*/ int e2,
+        /*const*/ int v3, /*const*/ int e3,
+        const float u, const float v,
+        const float du, const float dv)
+{
+	rotate_indices(rotation, &v0, &v1, &v2, &v3);
+	rotate_indices(rotation, &e0, &e1, &e2, &e3);
+	ctx->foreach_context->loop(
+	        ctx->foreach_context,
+	        tls,
+	        ptex_face_index, u, v,
+	        ORIGINDEX_NONE,
+	        coarse_poly_index,
+	        coarse_corner_index,
+	        subdiv_loop_start_index + 0,
+	        v0, e0);
+	ctx->foreach_context->loop(
+	        ctx->foreach_context,
+	        tls,
+	        ptex_face_index, u + du, v,
+	        ORIGINDEX_NONE,
+	        coarse_poly_index,
+	        coarse_corner_index,
+	        subdiv_loop_start_index + 1,
+	        v1, e1);
+	ctx->foreach_context->loop(
+	        ctx->foreach_context,
+	        tls,
+	        ptex_face_index, u + du, v + dv,
+	        ORIGINDEX_NONE,
+	        coarse_poly_index,
+	        coarse_corner_index,
+	        subdiv_loop_start_index + 2,
+	        v2, e2);
+	ctx->foreach_context->loop(
+	        ctx->foreach_context,
+	        tls,
+	        ptex_face_index, u, v + dv,
+	        ORIGINDEX_NONE,
+	        coarse_poly_index,
+	        coarse_corner_index,
+	        subdiv_loop_start_index + 3,
+	        v3, e3);
+}
+
+static void subdiv_foreach_loops_regular(SubdivForeachTaskContext *ctx,
+                                         void *tls,
+                                         const MPoly *coarse_poly)
+{
+	const int resolution = ctx->settings->resolution;
+	/* Base/coarse mesh information. */
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MEdge *coarse_medge = coarse_mesh->medge;
+	const MLoop *coarse_mloop = coarse_mesh->mloop;
+	const MPoly *coarse_mpoly = coarse_mesh->mpoly;
+	const int coarse_poly_index = coarse_poly - coarse_mpoly;
+	const int ptex_resolution =
+	        ptex_face_resolution_get(coarse_poly, resolution);
+	const int ptex_inner_resolution = ptex_resolution - 2;
+	const int num_subdiv_edges_per_coarse_edge = resolution - 1;
+	const int num_subdiv_vertices_per_coarse_edge = resolution - 2;
+	const float inv_ptex_resolution_1 = 1.0f / (float)(ptex_resolution - 1);
+	const int ptex_face_index = ctx->face_ptex_offset[coarse_poly_index];
+	const int start_vertex_index =
+	        ctx->vertices_inner_offset +
+	        ctx->subdiv_vertex_offset[coarse_poly_index];
+	const int start_edge_index =
+	        ctx->edge_inner_offset +
+	        ctx->subdiv_edge_offset[coarse_poly_index];
+	const int start_poly_index = ctx->subdiv_polygon_offset[coarse_poly_index];
+	const int start_loop_index = 4 * start_poly_index;
+	const float du = inv_ptex_resolution_1;
+	const float dv = inv_ptex_resolution_1;
+	/* Hi-poly subdivided mesh. */
+	int subdiv_loop_index = start_loop_index;
+	/* Loops for inner part of ptex. */
+	for (int y = 1; y < ptex_resolution - 2; y++) {
+		const float v = y * inv_ptex_resolution_1;
+		const int inner_y = y - 1;
+		for (int x = 1; x < ptex_resolution - 2; x++, subdiv_loop_index += 4) {
+			const int inner_x = x - 1;
+			const float u = x * inv_ptex_resolution_1;
+			/* Vertex indicies ordered counter-clockwise. */
+			const int v0 = start_vertex_index +
+			               (inner_y * ptex_inner_resolution + inner_x);
+			const int v1 = v0 + 1;
+			const int v2 = v0 + ptex_inner_resolution + 1;
+			const int v3 = v0 + ptex_inner_resolution;
+			/* Edge indicies ordered counter-clockwise. */
+			const int e0 = start_edge_index +
+			        (inner_y * (2 * ptex_inner_resolution - 1) + inner_x);
+			const int e1 = e0 + ptex_inner_resolution;
+			const int e2 = e0 + (2 * ptex_inner_resolution - 1);
+			const int e3 = e0 + ptex_inner_resolution - 1;
+			subdiv_foreach_loops_of_poly(
+			        ctx, tls, subdiv_loop_index, ptex_face_index,
+			        coarse_poly_index, 0,
+			        0,
+			        v0, e0, v1, e1, v2, e2, v3, e3,
+			        u, v, du, dv);
+		}
+	}
+	/* Loops for faces connecting inner ptex part with boundary. */
+	const MLoop *prev_coarse_loop =
+	        &coarse_mloop[coarse_poly->loopstart + coarse_poly->totloop - 1];
+	for (int corner = 0; corner < coarse_poly->totloop; corner++) {
+		const MLoop *coarse_loop =
+		        &coarse_mloop[coarse_poly->loopstart + corner];
+		const MEdge *coarse_edge = &coarse_medge[coarse_loop->e];
+		const MEdge *prev_coarse_edge = &coarse_medge[prev_coarse_loop->e];
+		const int start_edge_vertex = ctx->vertices_edge_offset +
+		        coarse_loop->e * num_subdiv_vertices_per_coarse_edge;
+		const bool flip = (coarse_edge->v2 == coarse_loop->v);
+		int side_start_index = start_vertex_index;
+		int side_stride = 0;
+		int v0 = ctx->vertices_corner_offset + coarse_loop->v;
+		int v3, e3;
+		int e2_offset, e2_stride;
+		float u, v, delta_u, delta_v;
+		if (prev_coarse_loop->v == prev_coarse_edge->v1) {
+			v3 = ctx->vertices_edge_offset +
+		        prev_coarse_loop->e * num_subdiv_vertices_per_coarse_edge +
+		        num_subdiv_vertices_per_coarse_edge - 1;
+			e3 = ctx->edge_boundary_offset +
+			         prev_coarse_loop->e * num_subdiv_edges_per_coarse_edge +
+			         num_subdiv_edges_per_coarse_edge - 1;
+		}
+		else {
+			v3 = ctx->vertices_edge_offset +
+		        prev_coarse_loop->e * num_subdiv_vertices_per_coarse_edge;
+			e3 = ctx->edge_boundary_offset +
+			         prev_coarse_loop->e * num_subdiv_edges_per_coarse_edge;
+		}
+		/* Calculate starting veretx of corresponding inner part of ptex. */
+		if (corner == 0) {
+			side_stride = 1;
+			e2_offset = 0;
+			e2_stride = 1;
+			u = 0.0f;
+			v = 0.0f;
+			delta_u = du;
+			delta_v = 0.0f;
+		}
+		else if (corner == 1) {
+			side_start_index += resolution - 3;
+			side_stride = resolution - 2;
+			e2_offset = 2 * num_subdiv_edges_per_coarse_edge - 4;
+			e2_stride = 2 * num_subdiv_edges_per_coarse_edge - 3;
+			u = 1.0f - du;
+			v = 0;
+			delta_u = 0.0f;
+			delta_v = dv;
+		}
+		else if (corner == 2) {
+			side_start_index += num_subdiv_vertices_per_coarse_edge *
+			                    num_subdiv_vertices_per_coarse_edge - 1;
+			side_stride = -1;
+			e2_offset = num_edges_per_ptex_face_get(resolution - 2) - 1;
+			e2_stride = -1;
+			u = 1.0f - du;
+			v = 1.0f - dv;
+			delta_u = -du;
+			delta_v = 0.0f;
+		}
+		else if (corner == 3) {
+			side_start_index += num_subdiv_vertices_per_coarse_edge *
+			                    (num_subdiv_vertices_per_coarse_edge - 1);
+			side_stride = -(resolution - 2);
+			e2_offset = num_edges_per_ptex_face_get(resolution - 2) -
+			            (2 * num_subdiv_edges_per_coarse_edge - 3);
+			e2_stride = -(2 * num_subdiv_edges_per_coarse_edge - 3);
+			u = 0.0f;
+			v = 1.0f - dv;
+			delta_u = 0.0f;
+			delta_v = -dv;
+		}
+		for (int i = 0; i < resolution - 2; i++, subdiv_loop_index += 4) {
+			int v1;
+			if (flip) {
+				v1 = start_edge_vertex + (resolution - i - 3);
+			}
+			else {
+				v1 = start_edge_vertex + i;
+			}
+			const int v2 = side_start_index + side_stride * i;
+			int e0;
+			if (flip) {
+				e0 = ctx->edge_boundary_offset +
+				         coarse_loop->e * num_subdiv_edges_per_coarse_edge +
+				         num_subdiv_edges_per_coarse_edge - i - 1;
+			}
+			else {
+				e0 = ctx->edge_boundary_offset +
+				         coarse_loop->e * num_subdiv_edges_per_coarse_edge +
+				         i;
+			}
+			int e1 = start_edge_index +
+			        num_edges_per_ptex_face_get(resolution - 2) +
+			        corner * num_subdiv_vertices_per_coarse_edge +
+			        i;
+			int e2;
+			if (i == 0) {
+				e2 = start_edge_index +
+				        num_edges_per_ptex_face_get(resolution - 2) +
+				        ((corner - 1 + coarse_poly->totloop) %
+				                coarse_poly->totloop) *
+				                        num_subdiv_vertices_per_coarse_edge +
+				        num_subdiv_vertices_per_coarse_edge - 1;
+			}
+			else {
+				e2 = start_edge_index + e2_offset + e2_stride * (i - 1);
+			}
+			subdiv_foreach_loops_of_poly(
+			        ctx, tls, subdiv_loop_index, ptex_face_index,
+			        coarse_poly_index, corner,
+			        corner,
+			        v0, e0, v1, e1, v2, e2, v3, e3,
+			        u + delta_u * i, v + delta_v * i, du, dv);
+			v0 = v1;
+			v3 = v2;
+			e3 = e1;
+		}
+		prev_coarse_loop = coarse_loop;
+	}
+}
+
+static void subdiv_foreach_loops_special(SubdivForeachTaskContext *ctx,
+                                         void *tls,
+                                         const MPoly *coarse_poly)
+{
+	const int resolution = ctx->settings->resolution;
+	/* Base/coarse mesh information. */
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MEdge *coarse_medge = coarse_mesh->medge;
+	const MLoop *coarse_mloop = coarse_mesh->mloop;
+	const MPoly *coarse_mpoly = coarse_mesh->mpoly;
+	const int coarse_poly_index = coarse_poly - coarse_mpoly;
+	const int ptex_face_resolution =
+	        ptex_face_resolution_get(coarse_poly, resolution);
+	const int ptex_face_inner_resolution = ptex_face_resolution - 2;
+	const float inv_ptex_resolution_1 =
+	       1.0f / (float)(ptex_face_resolution - 1);
+	const int num_inner_vertices_per_ptex =
+	        (ptex_face_resolution - 1) * (ptex_face_resolution - 2);
+	const int num_inner_edges_per_ptex_face =
+	        num_inner_edges_per_ptex_face_get(
+	                ptex_face_inner_resolution + 1);
+	const int num_subdiv_vertices_per_coarse_edge = resolution - 2;
+	const int num_subdiv_edges_per_coarse_edge = resolution - 1;
+	const int ptex_face_index = ctx->face_ptex_offset[coarse_poly_index];
+	const int center_vertex_index =
+	        ctx->vertices_inner_offset +
+	        ctx->subdiv_vertex_offset[coarse_poly_index];
+	const int start_vertex_index = center_vertex_index + 1;
+	const int start_inner_vertex_index = center_vertex_index + 1;
+	const int start_edge_index = ctx->edge_inner_offset +
+	                             ctx->subdiv_edge_offset[coarse_poly_index];
+	const int start_poly_index = ctx->subdiv_polygon_offset[coarse_poly_index];
+	const int start_loop_index = 4 * start_poly_index;
+	const float du = inv_ptex_resolution_1;
+	const float dv = inv_ptex_resolution_1;
+	/* Hi-poly subdivided mesh. */
+	int subdiv_loop_index = start_loop_index;
+	for (int corner = 0; corner < coarse_poly->totloop; corner++) {
+		const int corner_vertex_index =
+		        start_vertex_index + corner * num_inner_vertices_per_ptex;
+		const int corner_edge_index =
+		        start_edge_index + corner * num_inner_edges_per_ptex_face;
+		for (int y = 1; y < ptex_face_inner_resolution; y++) {
+			const float v = y * inv_ptex_resolution_1;
+			const int inner_y = y - 1;
+			for (int x = 1;
+			     x < ptex_face_inner_resolution + 1;
+			     x++, subdiv_loop_index += 4)
+			{
+				const int inner_x = x - 1;
+				const float u = x * inv_ptex_resolution_1;
+				/* Vertex indicies ordered counter-clockwise. */
+				const int v0 =
+				        corner_vertex_index +
+				        (inner_y * (ptex_face_inner_resolution + 1) + inner_x);
+				const int v1 = v0 + 1;
+				const int v2 = v0 + ptex_face_inner_resolution + 2;
+				const int v3 = v0 + ptex_face_inner_resolution + 1;
+				/* Edge indicies ordered counter-clockwise. */
+				const int e0 = corner_edge_index +
+				          (inner_y * (2 * ptex_face_inner_resolution + 1) + inner_x);
+				const int e1 = e0 + ptex_face_inner_resolution + 1;
+				const int e2 = e0 + (2 * ptex_face_inner_resolution + 1);
+				const int e3 = e0 + ptex_face_inner_resolution;
+				subdiv_foreach_loops_of_poly(
+				        ctx, tls, subdiv_loop_index, ptex_face_index + corner,
+				        coarse_poly_index, corner,
+				        0,
+				        v0, e0, v1, e1, v2, e2, v3, e3,
+				        u, v, du, dv);
+			}
+		}
+	}
+	/* Create connections between ptex faces. */
+	for (int corner = 0; corner < coarse_poly->totloop; corner++) {
+		const int next_corner = (corner + 1) % coarse_poly->totloop;
+		const int corner_edge_index =
+		        start_edge_index + corner * num_inner_edges_per_ptex_face;
+		const int next_corner_edge_index =
+		        start_edge_index + next_corner * num_inner_edges_per_ptex_face;
+		int current_patch_vertex_index =
+		        start_inner_vertex_index +
+		        corner * num_inner_vertices_per_ptex +
+		        ptex_face_inner_resolution;
+		int next_path_vertex_index =
+		        start_inner_vertex_index +
+		        next_corner * num_inner_vertices_per_ptex +
+		        num_inner_vertices_per_ptex - ptex_face_resolution + 1;
+		int v0 = current_patch_vertex_index;
+		int v1 = next_path_vertex_index;
+		current_patch_vertex_index += ptex_face_inner_resolution + 1;
+		next_path_vertex_index += 1;
+		int e0 = start_edge_index +
+		         coarse_poly->totloop * num_inner_edges_per_ptex_face +
+		         corner * (ptex_face_resolution - 2);
+		int e1 = next_corner_edge_index + num_inner_edges_per_ptex_face -
+		         ptex_face_resolution + 2;
+		int e3 = corner_edge_index + 2 * ptex_face_resolution - 4;
+		for (int row = 1;
+		     row < ptex_face_inner_resolution;
+		     row++, subdiv_loop_index += 4)
+		{
+			const int v2 = next_path_vertex_index;
+			const int v3 = current_patch_vertex_index;
+			const int e2 = e0 + 1;
+			const float u = row * du;
+			const float v = 1.0f - dv;
+			subdiv_foreach_loops_of_poly(
+			        ctx, tls, subdiv_loop_index, ptex_face_index + next_corner,
+			        coarse_poly_index, next_corner,
+			        3,
+			        v0, e0, v1, e1, v2, e2, v3, e3,
+			        u, v, du, dv);
+			current_patch_vertex_index += ptex_face_inner_resolution + 1;
+			next_path_vertex_index += 1;
+			v0 = v3;
+			v1 = v2;
+			e0 = e2;
+			e1 += 1;
+			e3 += 2 * ptex_face_resolution - 3;
+		}
+	}
+	/* Create loops from center. */
+	if (ptex_face_resolution >= 3) {
+		const int start_center_edge_index =
+		        start_edge_index +
+		        (num_inner_edges_per_ptex_face +
+		         ptex_face_inner_resolution) * coarse_poly->totloop;
+		const int start_boundary_edge =
+		        start_edge_index +
+		        coarse_poly->totloop * num_inner_edges_per_ptex_face +
+		        ptex_face_inner_resolution - 1;
+		for (int corner = 0, prev_corner = coarse_poly->totloop - 1;
+		     corner < coarse_poly->totloop;
+		     prev_corner = corner, corner++, subdiv_loop_index += 4)
+		{
+			const int corner_edge_index =
+			        start_edge_index +
+			        corner * num_inner_edges_per_ptex_face;
+			const int current_patch_end_vertex_index =
+			        start_vertex_index + corner * num_inner_vertices_per_ptex +
+			        num_inner_vertices_per_ptex - 1;
+			const int prev_current_patch_end_vertex_index =
+			        start_vertex_index + prev_corner *
+			                num_inner_vertices_per_ptex +
+			        num_inner_vertices_per_ptex - 1;
+			const int v0 = center_vertex_index;
+			const int v1 = prev_current_patch_end_vertex_index;
+			const int v2 = current_patch_end_vertex_index - 1;
+			const int v3 = current_patch_end_vertex_index;
+			const int e0 = start_center_edge_index + prev_corner;
+			const int e1 = start_boundary_edge +
+			               prev_corner * (ptex_face_inner_resolution);
+			const int e2 = corner_edge_index +
+			               num_inner_edges_per_ptex_face - 1;
+			const int e3 = start_center_edge_index + corner;
+			const float u = 1.0f - du;
+			const float v = 1.0f - dv;
+			subdiv_foreach_loops_of_poly(
+			        ctx, tls, subdiv_loop_index,
+			        ptex_face_index + corner,
+			        coarse_poly_index, corner,
+			        2,
+			        v0, e0, v1, e1, v2, e2, v3, e3,
+			        u, v, du, dv);
+		}
+	}
+	/* Loops for faces connecting inner ptex part with boundary. */
+	const MLoop *prev_coarse_loop =
+	        &coarse_mloop[coarse_poly->loopstart + coarse_poly->totloop - 1];
+	for (int prev_corner = coarse_poly->totloop - 1, corner = 0;
+	     corner < coarse_poly->totloop;
+	     prev_corner = corner, corner++)
+	{
+		const MLoop *coarse_loop =
+		        &coarse_mloop[coarse_poly->loopstart + corner];
+		const MEdge *coarse_edge = &coarse_medge[coarse_loop->e];
+		const MEdge *prev_coarse_edge = &coarse_medge[prev_coarse_loop->e];
+		const bool flip = (coarse_edge->v2 == coarse_loop->v);
+		const int start_edge_vertex = ctx->vertices_edge_offset +
+		        coarse_loop->e * num_subdiv_vertices_per_coarse_edge;
+		const int corner_vertex_index =
+		        start_vertex_index + corner * num_inner_vertices_per_ptex;
+		const int corner_edge_index =
+		        start_edge_index + corner * num_inner_edges_per_ptex_face;
+		/* Create loops for polygons along U axis. */
+		int v0 = ctx->vertices_corner_offset + coarse_loop->v;
+		int v3, e3;
+		if (prev_coarse_loop->v == prev_coarse_edge->v1) {
+			v3 = ctx->vertices_edge_offset +
+		        prev_coarse_loop->e * num_subdiv_vertices_per_coarse_edge +
+		        num_subdiv_vertices_per_coarse_edge - 1;
+			e3 = ctx->edge_boundary_offset +
+			         prev_coarse_loop->e * num_subdiv_edges_per_coarse_edge +
+			         num_subdiv_edges_per_coarse_edge - 1;
+		}
+		else {
+			v3 = ctx->vertices_edge_offset +
+		        prev_coarse_loop->e * num_subdiv_vertices_per_coarse_edge;
+			e3 = ctx->edge_boundary_offset +
+			         prev_coarse_loop->e * num_subdiv_edges_per_coarse_edge;
+		}
+		for (int i = 0;
+		     i <= ptex_face_inner_resolution;
+		     i++, subdiv_loop_index += 4)
+		{
+			int v1;
+			if (flip) {
+				v1 = start_edge_vertex + (resolution - i - 3);
+			}
+			else {
+				v1 = start_edge_vertex + i;
+			}
+			int v2;
+			if (ptex_face_inner_resolution >= 1) {
+				v2 = corner_vertex_index + i;
+			}
+			else {
+				v2 = center_vertex_index;
+			}
+			int e0;
+			if (flip) {
+				e0 = ctx->edge_boundary_offset +
+				         coarse_loop->e * num_subdiv_edges_per_coarse_edge +
+				         num_subdiv_edges_per_coarse_edge - i - 1;
+			}
+			else {
+				e0 = ctx->edge_boundary_offset +
+				         coarse_loop->e * num_subdiv_edges_per_coarse_edge +
+				         i;
+			}
+			int e1 = start_edge_index +
+			         corner * (2 * ptex_face_inner_resolution + 1);
+			if (ptex_face_resolution >= 3) {
+				e1 += coarse_poly->totloop * (num_inner_edges_per_ptex_face +
+				                              ptex_face_inner_resolution + 1) +
+				      i;
+			}
+			int e2 = 0;
+			if (i == 0 && ptex_face_resolution >= 3) {
+				e2 = start_edge_index +
+				         coarse_poly->totloop *
+				                 (num_inner_edges_per_ptex_face +
+				                  ptex_face_inner_resolution + 1) +
+				         corner * (2 * ptex_face_inner_resolution + 1) +
+				         ptex_face_inner_resolution + 1;
+			}
+			else if (i == 0 && ptex_face_resolution < 3) {
+				e2 = start_edge_index +
+			         prev_corner * (2 * ptex_face_inner_resolution + 1);
+			}
+			else {
+				e2 = corner_edge_index + i - 1;
+			}
+			const float u = du * i;
+			const float v = 0.0f;
+			subdiv_foreach_loops_of_poly(
+			        ctx, tls, subdiv_loop_index, ptex_face_index + corner,
+			        coarse_poly_index, corner,
+			        0,
+			        v0, e0, v1, e1, v2, e2, v3, e3,
+			        u, v, du, dv);
+			v0 = v1;
+			v3 = v2;
+			e3 = e1;
+		}
+		/* Create loops for polygons along V axis. */
+		const bool flip_prev = (prev_coarse_edge->v2 == coarse_loop->v);
+		v0 = corner_vertex_index;
+		if (prev_coarse_loop->v == prev_coarse_edge->v1) {
+			v3 = ctx->vertices_edge_offset +
+		        prev_coarse_loop->e * num_subdiv_vertices_per_coarse_edge +
+		        num_subdiv_vertices_per_coarse_edge - 1;
+		}
+		else {
+			v3 = ctx->vertices_edge_offset +
+		        prev_coarse_loop->e * num_subdiv_vertices_per_coarse_edge;
+		}
+		e3 = start_edge_index +
+		         coarse_poly->totloop *
+		                 (num_inner_edges_per_ptex_face +
+		                  ptex_face_inner_resolution + 1) +
+		         corner * (2 * ptex_face_inner_resolution + 1) +
+		         ptex_face_inner_resolution + 1;
+		for (int i = 0;
+		     i <= ptex_face_inner_resolution - 1;
+		     i++, subdiv_loop_index += 4)
+		{
+			int v1;
+			int e0, e1;
+			if (i == ptex_face_inner_resolution - 1) {
+				v1 = start_vertex_index +
+				     prev_corner * num_inner_vertices_per_ptex +
+				     ptex_face_inner_resolution;
+				e1 = start_edge_index +
+				         coarse_poly->totloop *
+				                 (num_inner_edges_per_ptex_face +
+				                  ptex_face_inner_resolution + 1) +
+				         prev_corner * (2 * ptex_face_inner_resolution + 1) +
+				         ptex_face_inner_resolution;
+				e0 = start_edge_index +
+				      coarse_poly->totloop * num_inner_edges_per_ptex_face +
+				      prev_corner * ptex_face_inner_resolution;
+			}
+			else {
+				v1 = v0 + ptex_face_inner_resolution + 1;
+				e0 = corner_edge_index + ptex_face_inner_resolution +
+				         i * (2 * ptex_face_inner_resolution + 1);
+				e1 = e3 + 1;
+			}
+			int v2 = flip_prev ? v3 - 1 : v3 + 1;
+			int e2;
+			if (flip_prev) {
+				e2 = ctx->edge_boundary_offset +
+				         prev_coarse_loop->e *
+				                 num_subdiv_edges_per_coarse_edge +
+				         num_subdiv_edges_per_coarse_edge - 2 - i;
+			}
+			else {
+				e2 = ctx->edge_boundary_offset +
+				         prev_coarse_loop->e *
+				                 num_subdiv_edges_per_coarse_edge + 1 + i;
+			}
+			const float u = 0.0f;
+			const float v = du * (i + 1);
+			subdiv_foreach_loops_of_poly(
+			        ctx, tls, subdiv_loop_index, ptex_face_index + corner,
+			        coarse_poly_index, corner,
+			        1,
+			        v0, e0, v1, e1, v2, e2, v3, e3,
+			        u, v, du, dv);
+			v0 = v1;
+			v3 = v2;
+			e3 = e1;
+		}
+		prev_coarse_loop = coarse_loop;
+	}
+}
+
+static void subdiv_foreach_loops(SubdivForeachTaskContext *ctx,
+                                 void *tls,
+                                 int poly_index)
+{
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MPoly *coarse_mpoly = coarse_mesh->mpoly;
+	const MPoly *coarse_poly = &coarse_mpoly[poly_index];
+	if (coarse_poly->totloop == 4) {
+		subdiv_foreach_loops_regular(ctx, tls, coarse_poly);
+	}
+	else {
+		subdiv_foreach_loops_special(ctx, tls, coarse_poly);
+	}
+}
+
+/* =============================================================================
+ * Polygons traverse process.
+ */
+
+static void subdiv_foreach_polys(SubdivForeachTaskContext *ctx,
+                                 void *tls,
+                                 int poly_index)
+{
+	const int resolution = ctx->settings->resolution;
+	const int start_poly_index = ctx->subdiv_polygon_offset[poly_index];
+	/* Base/coarse mesh information. */
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MPoly *coarse_mpoly = coarse_mesh->mpoly;
+	const MPoly *coarse_poly = &coarse_mpoly[poly_index];
+	const int num_ptex_faces_per_poly =
+	        num_ptex_faces_per_poly_get(coarse_poly);
+	const int ptex_resolution =
+	        ptex_face_resolution_get(coarse_poly, resolution);
+	const int num_polys_per_ptex = num_polys_per_ptex_get(ptex_resolution);
+	const int num_loops_per_ptex = 4 * num_polys_per_ptex;
+	const int start_loop_index = 4 * start_poly_index;
+	/* Hi-poly subdivided mesh. */
+	int subdiv_polyon_index = start_poly_index;
+	for (int ptex_of_poly_index = 0;
+	     ptex_of_poly_index < num_ptex_faces_per_poly;
+	     ptex_of_poly_index++)
+	{
+		for (int subdiv_poly_index = 0;
+		     subdiv_poly_index < num_polys_per_ptex;
+		     subdiv_poly_index++, subdiv_polyon_index++)
+		{
+			const int loopstart = start_loop_index +
+			                       (ptex_of_poly_index * num_loops_per_ptex) +
+			                       (subdiv_poly_index * 4);
+			ctx->foreach_context->poly(
+			        ctx->foreach_context,
+			        tls,
+			        poly_index,
+			        subdiv_polyon_index, 
+			        loopstart, 4);
+		}
+	}
+}
+
+/* =============================================================================
+ * Loose elements traverse process.
+ */
+
+static void subdiv_foreach_loose_vertices_task(
+        void *__restrict userdata,
+        const int coarse_vertex_index,
+        const ParallelRangeTLS *__restrict tls)
+{
+	SubdivForeachTaskContext *ctx = userdata;
+	if (BLI_BITMAP_TEST_BOOL(ctx->coarse_vertices_used_map,
+	                         coarse_vertex_index))
+	{
+		/* Vertex is not loose, was handled when handling polygons. */
+		return;
+	}
+	const int subdiv_vertex_index =
+	        ctx->vertices_corner_offset + coarse_vertex_index;
+	ctx->foreach_context->vertex_loose(
+	        ctx->foreach_context,
+	        tls->userdata_chunk,
+	        coarse_vertex_index,
+	        subdiv_vertex_index);
+}
+
+static void subdiv_foreach_vertices_of_loose_edges_task(
+        void *__restrict userdata,
+        const int coarse_edge_index,
+        const ParallelRangeTLS *__restrict tls)
+{
+	SubdivForeachTaskContext *ctx = userdata;
+	if (BLI_BITMAP_TEST_BOOL(ctx->coarse_edges_used_map, coarse_edge_index)) {
+		/* Vertex is not loose, was handled when handling polygons. */
+		return;
+	}
+	const int resolution = ctx->settings->resolution;
+	const int resolution_1 = resolution - 1;
+	const float inv_resolution_1 = 1.0f / (float)resolution_1;
+	const int num_subdiv_vertices_per_coarse_edge = resolution - 2;
+	const Mesh *coarse_mesh = ctx->coarse_mesh;
+	const MEdge *coarse_edge = &coarse_mesh->medge[coarse_edge_index];
+	/* Subdivion verticies which corresponds to edge's v1 and v2. */
+	const int subdiv_v1_index =
+	        ctx->vertices_corner_offset + coarse_edge->v1;
+	const int subdiv_v2_index =
+	        ctx->vertices_corner_offset + coarse_edge->v2;
+	/* First subdivided inner vertex of the edge.  */
+	const int subdiv_start_vertex =
+	        ctx->vertices_edge_offset +
+	        coarse_edge_index * num_subdiv_vertices_per_coarse_edge;
+	/* Perform interpolation. */
+	for (int i = 0; i < resolution; i++) {
+		const float u = i * inv_resolution_1;
+		int subdiv_vertex_index;
+		if (i == 0) {
+			subdiv_vertex_index = subdiv_v1_index;
+		}
+		else if (i == resolution - 1) {
+			subdiv_vertex_index = subdiv_v2_index;
+		}
+		else {
+			subdiv_vertex_index = subdiv_start_vertex + (i - 1);
+		}
+		ctx->foreach_context->vertex_of_loose_edge(
+		        ctx->foreach_context,
+		        tls->userdata_chunk,
+		        coarse_edge_index,
+		        u,
+		        subdiv_vertex_index);
+	}
+}
+
+/* =============================================================================
+ * Subdivision process entry points.
+ */
+
+static void subdiv_foreach_task(
+        void *__restrict userdata,
+        const int poly_index,
+        const ParallelRangeTLS *__restrict tls)
+{
+	SubdivForeachTaskContext *ctx = userdata;
+	/* Traverse hi-poly vertex coordinates and normals. */
+	subdiv_foreach_vertices(ctx, tls->userdata_chunk, poly_index);
+	/* Traverse mesh geometry for the given base poly index. */
+	if (ctx->foreach_context->edge != NULL) {
+		subdiv_foreach_edges(ctx, tls->userdata_chunk, poly_index);
+	}
+	if (ctx->foreach_context->loop != NULL) {
+		subdiv_foreach_loops(ctx, tls->userdata_chunk, poly_index);
+	}
+	if (ctx->foreach_context->poly != NULL) {
+		subdiv_foreach_polys(ctx, tls->userdata_chunk, poly_index);
+	}
+}
+
+static void subdiv_foreach_boundary_edges_task(
+        void *__restrict userdata,
+        const int edge_index,
+        const ParallelRangeTLS *__restrict tls)
+{
+	SubdivForeachTaskContext *ctx = userdata;
+	subdiv_foreach_boundary_edges(ctx, tls->userdata_chunk, edge_index);
+}
+
+static void subdiv_foreach_finalize(void *__restrict userdata,
+                                    void *__restrict userdata_chunk)
+{
+	SubdivForeachTaskContext *ctx = userdata;
+	ctx->foreach_context->user_data_tls_free(userdata_chunk);
+}
+
+bool BKE_subdiv_foreach_subdiv_geometry(
+        struct Subdiv *UNUSED(subdiv),
+        const SubdivForeachContext *context,
+        const SubdivToMeshSettings *mesh_settings,
+        const struct Mesh *coarse_mesh)
+{
+	SubdivForeachTaskContext ctx = {0};
+	ctx.coarse_mesh = coarse_mesh;
+	ctx.settings = mesh_settings;
+	ctx.foreach_context = context;
+	subdiv_foreach_ctx_init(&ctx);
+	if (context->topology_info != NULL) {
+		if (!context->topology_info(context,
+		                            ctx.num_subdiv_vertices,
+		                            ctx.num_subdiv_edges,
+		                            ctx.num_subdiv_loops,
+		                            ctx.num_subdiv_polygons))
+		{
+			return false;
+		}
+	}
+	/* Single threaded passes to average displacement on the corner vertices
+	 * and boundary edges.
+	 */
+	subdiv_foreach_every_corner_vertices(&ctx);
+	subdiv_foreach_every_edge_vertices(&ctx);
+	/* Threaded traversal of the rest of topology. */
+	ParallelRangeSettings parallel_range_settings;
+	BLI_parallel_range_settings_defaults(&parallel_range_settings);
+	parallel_range_settings.userdata_chunk = context->user_data_tls;
+	parallel_range_settings.userdata_chunk_size = context->user_data_tls_size;
+	if (context->user_data_tls_free != NULL) {
+		parallel_range_settings.func_finalize = subdiv_foreach_finalize;
+	}
+	BLI_task_parallel_range(0, coarse_mesh->totpoly,
+	                        &ctx,
+	                        subdiv_foreach_task,
+	                        &parallel_range_settings);
+	if (context->vertex_loose != NULL) {
+		BLI_task_parallel_range(0, coarse_mesh->totvert,
+		                        &ctx,
+		                        subdiv_foreach_loose_vertices_task,
+		                        &parallel_range_settings);
+	}
+	if (context->vertex_of_loose_edge != NULL) {
+		BLI_task_parallel_range(0, coarse_mesh->totedge,
+		                        &ctx,
+		                        subdiv_foreach_vertices_of_loose_edges_task,
+		                        &parallel_range_settings);
+	}
+	if (context->edge != NULL) {
+		BLI_task_parallel_range(0, coarse_mesh->totedge,
+		                        &ctx,
+		                        subdiv_foreach_boundary_edges_task,
+		                        &parallel_range_settings);
+	}
+	subdiv_foreach_ctx_free(&ctx);
+	return true;
+}