diff --git a/3rdparty/meshoptimizer/src/clusterizer.cpp b/3rdparty/meshoptimizer/src/clusterizer.cpp
index b01342781..26d2fb11c 100644
--- a/3rdparty/meshoptimizer/src/clusterizer.cpp
+++ b/3rdparty/meshoptimizer/src/clusterizer.cpp
@@ -13,12 +13,16 @@
 namespace meshopt
 {
 
-// This must be <= 255 since index 0xff is used internally to indice a vertex that doesn't belong to a meshlet
-const size_t kMeshletMaxVertices = 255;
+// This must be <= 256 since meshlet indices are stored as bytes
+const size_t kMeshletMaxVertices = 256;
 
 // A reasonable limit is around 2*max_vertices or less
 const size_t kMeshletMaxTriangles = 512;
 
+// We keep a limited number of seed triangles and add a few triangles per finished meshlet
+const size_t kMeshletMaxSeeds = 256;
+const size_t kMeshletAddSeeds = 4;
+
 struct TriangleAdjacency2
 {
 	unsigned int* counts;
@@ -97,10 +101,9 @@ static void buildTriangleAdjacencySparse(TriangleAdjacency2& adjacency, const un
 	for (size_t i = 0; i < index_count; ++i)
 		adjacency.counts[indices[i]]++;
 
-	// fill offset table
+	// fill offset table; uses sparse_seen bit to tag visited vertices
 	unsigned int offset = 0;
 
-	// when using sparse mode this pass uses sparse_seen bit to tag visited vertices
 	for (size_t i = 0; i < index_count; ++i)
 	{
 		unsigned int v = indices[i];
@@ -126,7 +129,7 @@ static void buildTriangleAdjacencySparse(TriangleAdjacency2& adjacency, const un
 	}
 
 	// fix offsets that have been disturbed by the previous pass
-	// when using sparse mode this pass also fixes counts (that were marked with sparse_seen)
+	// also fix counts (that were marked with sparse_seen by the first pass)
 	for (size_t i = 0; i < index_count; ++i)
 	{
 		unsigned int v = indices[i];
@@ -141,67 +144,82 @@ static void buildTriangleAdjacencySparse(TriangleAdjacency2& adjacency, const un
 	}
 }
 
-static void computeBoundingSphere(float result[4], const float points[][3], size_t count)
+static void computeBoundingSphere(float result[4], const float* points, size_t count, size_t points_stride, const float* radii, size_t radii_stride)
 {
 	assert(count > 0);
 
+	size_t points_stride_float = points_stride / sizeof(float);
+	size_t radii_stride_float = radii_stride / sizeof(float);
+
 	// find extremum points along all 3 axes; for each axis we get a pair of points with min/max coordinates
 	size_t pmin[3] = {0, 0, 0};
 	size_t pmax[3] = {0, 0, 0};
 
 	for (size_t i = 0; i < count; ++i)
 	{
-		const float* p = points[i];
+		const float* p = points + i * points_stride_float;
+		float r = radii[i * radii_stride_float];
 
 		for (int axis = 0; axis < 3; ++axis)
 		{
-			pmin[axis] = (p[axis] < points[pmin[axis]][axis]) ? i : pmin[axis];
-			pmax[axis] = (p[axis] > points[pmax[axis]][axis]) ? i : pmax[axis];
+			float bmin = points[pmin[axis] * points_stride_float + axis] - radii[pmin[axis] * radii_stride_float];
+			float bmax = points[pmax[axis] * points_stride_float + axis] + radii[pmax[axis] * radii_stride_float];
+
+			pmin[axis] = (p[axis] - r < bmin) ? i : pmin[axis];
+			pmax[axis] = (p[axis] + r > bmax) ? i : pmax[axis];
 		}
 	}
 
 	// find the pair of points with largest distance
-	float paxisd2 = 0;
 	int paxis = 0;
+	float paxisdr = 0;
 
 	for (int axis = 0; axis < 3; ++axis)
 	{
-		const float* p1 = points[pmin[axis]];
-		const float* p2 = points[pmax[axis]];
+		const float* p1 = points + pmin[axis] * points_stride_float;
+		const float* p2 = points + pmax[axis] * points_stride_float;
+		float r1 = radii[pmin[axis] * radii_stride_float];
+		float r2 = radii[pmax[axis] * radii_stride_float];
 
 		float d2 = (p2[0] - p1[0]) * (p2[0] - p1[0]) + (p2[1] - p1[1]) * (p2[1] - p1[1]) + (p2[2] - p1[2]) * (p2[2] - p1[2]);
+		float dr = sqrtf(d2) + r1 + r2;
 
-		if (d2 > paxisd2)
+		if (dr > paxisdr)
 		{
-			paxisd2 = d2;
+			paxisdr = dr;
 			paxis = axis;
 		}
 	}
 
 	// use the longest segment as the initial sphere diameter
-	const float* p1 = points[pmin[paxis]];
-	const float* p2 = points[pmax[paxis]];
+	const float* p1 = points + pmin[paxis] * points_stride_float;
+	const float* p2 = points + pmax[paxis] * points_stride_float;
+	float r1 = radii[pmin[paxis] * radii_stride_float];
+	float r2 = radii[pmax[paxis] * radii_stride_float];
 
-	float center[3] = {(p1[0] + p2[0]) / 2, (p1[1] + p2[1]) / 2, (p1[2] + p2[2]) / 2};
-	float radius = sqrtf(paxisd2) / 2;
+	float paxisd = sqrtf((p2[0] - p1[0]) * (p2[0] - p1[0]) + (p2[1] - p1[1]) * (p2[1] - p1[1]) + (p2[2] - p1[2]) * (p2[2] - p1[2]));
+	float paxisk = paxisd > 0 ? (paxisd + r2 - r1) / (2 * paxisd) : 0.f;
+
+	float center[3] = {p1[0] + (p2[0] - p1[0]) * paxisk, p1[1] + (p2[1] - p1[1]) * paxisk, p1[2] + (p2[2] - p1[2]) * paxisk};
+	float radius = paxisdr / 2;
 
 	// iteratively adjust the sphere up until all points fit
 	for (size_t i = 0; i < count; ++i)
 	{
-		const float* p = points[i];
+		const float* p = points + i * points_stride_float;
+		float r = radii[i * radii_stride_float];
+
 		float d2 = (p[0] - center[0]) * (p[0] - center[0]) + (p[1] - center[1]) * (p[1] - center[1]) + (p[2] - center[2]) * (p[2] - center[2]);
+		float d = sqrtf(d2);
 
-		if (d2 > radius * radius)
+		if (d + r > radius)
 		{
-			float d = sqrtf(d2);
-			assert(d > 0);
+			float k = d > 0 ? (d + r - radius) / (2 * d) : 0.f;
 
-			float k = 0.5f + (radius / d) / 2;
-
-			center[0] = center[0] * k + p[0] * (1 - k);
-			center[1] = center[1] * k + p[1] * (1 - k);
-			center[2] = center[2] * k + p[2] * (1 - k);
-			radius = (radius + d) / 2;
+			center[0] += k * (p[0] - center[0]);
+			center[1] += k * (p[1] - center[1]);
+			center[2] += k * (p[2] - center[2]);
+			radius = (radius + d + r) / 2;
 		}
 	}
 
@@ -217,12 +235,25 @@ struct Cone
 	float nx, ny, nz;
 };
 
-static float getMeshletScore(float distance2, float spread, float cone_weight, float expected_radius)
+static float getDistance(float dx, float dy, float dz, bool aa)
 {
+	if (!aa)
+		return sqrtf(dx * dx + dy * dy + dz * dz);
+
+	float rx = fabsf(dx), ry = fabsf(dy), rz = fabsf(dz);
+	float rxy = rx > ry ? rx : ry;
+	return rxy > rz ? rxy : rz;
+}
+
+static float getMeshletScore(float distance, float spread, float cone_weight, float expected_radius)
+{
+	if (cone_weight < 0)
+		return 1 + distance / expected_radius;
+
 	float cone = 1.f - spread * cone_weight;
 	float cone_clamped = cone < 1e-3f ? 1e-3f : cone;
 
-	return (1 + sqrtf(distance2) / expected_radius * (1 - cone_weight)) * cone_clamped;
+	return (1 + distance / expected_radius * (1 - cone_weight)) * cone_clamped;
 }
 
 static Cone getMeshletCone(const Cone& acc, unsigned int triangle_count)
@@ -296,22 +327,22 @@ static void finishMeshlet(meshopt_Meshlet& meshlet, unsigned char* meshlet_trian
 		meshlet_triangles[offset++] = 0;
 }
 
-static bool appendMeshlet(meshopt_Meshlet& meshlet, unsigned int a, unsigned int b, unsigned int c, unsigned char* used, meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, size_t meshlet_offset, size_t max_vertices, size_t max_triangles)
+static bool appendMeshlet(meshopt_Meshlet& meshlet, unsigned int a, unsigned int b, unsigned int c, short* used, meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, size_t meshlet_offset, size_t max_vertices, size_t max_triangles, bool split = false)
 {
-	unsigned char& av = used[a];
-	unsigned char& bv = used[b];
-	unsigned char& cv = used[c];
+	short& av = used[a];
+	short& bv = used[b];
+	short& cv = used[c];
 
 	bool result = false;
 
-	int used_extra = (av == 0xff) + (bv == 0xff) + (cv == 0xff);
+	int used_extra = (av < 0) + (bv < 0) + (cv < 0);
 
-	if (meshlet.vertex_count + used_extra > max_vertices || meshlet.triangle_count >= max_triangles)
+	if (meshlet.vertex_count + used_extra > max_vertices || meshlet.triangle_count >= max_triangles || split)
 	{
 		meshlets[meshlet_offset] = meshlet;
 
 		for (size_t j = 0; j < meshlet.vertex_count; ++j)
-			used[meshlet_vertices[meshlet.vertex_offset + j]] = 0xff;
+			used[meshlet_vertices[meshlet.vertex_offset + j]] = -1;
 
 		finishMeshlet(meshlet, meshlet_triangles);
 
@@ -323,33 +354,33 @@ static bool appendMeshlet(meshopt_Meshlet& meshlet, unsigned int a, unsigned int
 		result = true;
 	}
 
-	if (av == 0xff)
+	if (av < 0)
 	{
-		av = (unsigned char)meshlet.vertex_count;
+		av = short(meshlet.vertex_count);
 		meshlet_vertices[meshlet.vertex_offset + meshlet.vertex_count++] = a;
 	}
 
-	if (bv == 0xff)
+	if (bv < 0)
 	{
-		bv = (unsigned char)meshlet.vertex_count;
+		bv = short(meshlet.vertex_count);
 		meshlet_vertices[meshlet.vertex_offset + meshlet.vertex_count++] = b;
 	}
 
-	if (cv == 0xff)
+	if (cv < 0)
 	{
-		cv = (unsigned char)meshlet.vertex_count;
+		cv = short(meshlet.vertex_count);
 		meshlet_vertices[meshlet.vertex_offset + meshlet.vertex_count++] = c;
 	}
 
-	meshlet_triangles[meshlet.triangle_offset + meshlet.triangle_count * 3 + 0] = av;
-	meshlet_triangles[meshlet.triangle_offset + meshlet.triangle_count * 3 + 1] = bv;
-	meshlet_triangles[meshlet.triangle_offset + meshlet.triangle_count * 3 + 2] = cv;
+	meshlet_triangles[meshlet.triangle_offset + meshlet.triangle_count * 3 + 0] = (unsigned char)av;
+	meshlet_triangles[meshlet.triangle_offset + meshlet.triangle_count * 3 + 1] = (unsigned char)bv;
+	meshlet_triangles[meshlet.triangle_offset + meshlet.triangle_count * 3 + 2] = (unsigned char)cv;
 	meshlet.triangle_count++;
 
 	return result;
 }
 
-static unsigned int getNeighborTriangle(const meshopt_Meshlet& meshlet, const Cone* meshlet_cone, unsigned int* meshlet_vertices, const unsigned int* indices, const TriangleAdjacency2& adjacency, const Cone* triangles, const unsigned int* live_triangles, const unsigned char* used, float meshlet_expected_radius, float cone_weight)
+static unsigned int getNeighborTriangle(const meshopt_Meshlet& meshlet, const Cone& meshlet_cone, const unsigned int* meshlet_vertices, const unsigned int* indices, const TriangleAdjacency2& adjacency, const Cone* triangles, const unsigned int* live_triangles, const short* used, float meshlet_expected_radius, float cone_weight)
 {
 	unsigned int best_triangle = ~0u;
 	int best_priority = 5;
@@ -367,7 +398,7 @@ static unsigned int getNeighborTriangle(const meshopt_Meshlet& meshlet, const Co
 			unsigned int triangle = neighbors[j];
 			unsigned int a = indices[triangle * 3 + 0], b = indices[triangle * 3 + 1], c = indices[triangle * 3 + 2];
 
-			int extra = (used[a] == 0xff) + (used[b] == 0xff) + (used[c] == 0xff);
+			int extra = (used[a] < 0) + (used[b] < 0) + (used[c] < 0);
 			assert(extra <= 2);
 
 			int priority = -1;
@@ -389,27 +420,13 @@ static unsigned int getNeighborTriangle(const meshopt_Meshlet& meshlet, const Co
 			if (priority > best_priority)
 				continue;
 
-			float score = 0;
+			const Cone& tri_cone = triangles[triangle];
 
-			// caller selects one of two scoring functions: geometrical (based on meshlet cone) or topological (based on remaining triangles)
-			if (meshlet_cone)
-			{
-				const Cone& tri_cone = triangles[triangle];
+			float dx = tri_cone.px - meshlet_cone.px, dy = tri_cone.py - meshlet_cone.py, dz = tri_cone.pz - meshlet_cone.pz;
+			float distance = getDistance(dx, dy, dz, cone_weight < 0);
+			float spread = tri_cone.nx * meshlet_cone.nx + tri_cone.ny * meshlet_cone.ny + tri_cone.nz * meshlet_cone.nz;
 
-				float distance2 =
-				    (tri_cone.px - meshlet_cone->px) * (tri_cone.px - meshlet_cone->px) +
-				    (tri_cone.py - meshlet_cone->py) * (tri_cone.py - meshlet_cone->py) +
-				    (tri_cone.pz - meshlet_cone->pz) * (tri_cone.pz - meshlet_cone->pz);
-
-				float spread = tri_cone.nx * meshlet_cone->nx + tri_cone.ny * meshlet_cone->ny + tri_cone.nz * meshlet_cone->nz;
-
-				score = getMeshletScore(distance2, spread, cone_weight, meshlet_expected_radius);
-			}
-			else
-			{
-				// each live_triangles entry is >= 1 since it includes the current triangle we're processing
-				score = float(live_triangles[a] + live_triangles[b] + live_triangles[c] - 3);
-			}
+			float score = getMeshletScore(distance, spread, cone_weight, meshlet_expected_radius);
 
 			// note that topology-based priority is always more important than the score
 			// this helps maintain reasonable effectiveness of meshlet data and reduces scoring cost
@@ -425,6 +442,113 @@ static unsigned int getNeighborTriangle(const meshopt_Meshlet& meshlet, const Co
 	return best_triangle;
 }
 
+static size_t appendSeedTriangles(unsigned int* seeds, const meshopt_Meshlet& meshlet, const unsigned int* meshlet_vertices, const unsigned int* indices, const TriangleAdjacency2& adjacency, const Cone* triangles, const unsigned int* live_triangles, float cornerx, float cornery, float cornerz)
+{
+	unsigned int best_seeds[kMeshletAddSeeds];
+	unsigned int best_live[kMeshletAddSeeds];
+	float best_score[kMeshletAddSeeds];
+
+	for (size_t i = 0; i < kMeshletAddSeeds; ++i)
+	{
+		best_seeds[i] = ~0u;
+		best_live[i] = ~0u;
+		best_score[i] = FLT_MAX;
+	}
+
+	for (size_t i = 0; i < meshlet.vertex_count; ++i)
+	{
+		unsigned int index = meshlet_vertices[meshlet.vertex_offset + i];
+
+		unsigned int best_neighbor = ~0u;
+		unsigned int best_neighbor_live = ~0u;
+
+		// find the neighbor with the smallest live metric
+		unsigned int* neighbors = &adjacency.data[0] + adjacency.offsets[index];
+		size_t neighbors_size = adjacency.counts[index];
+
+		for (size_t j = 0; j < neighbors_size; ++j)
+		{
+			unsigned int triangle = neighbors[j];
+			unsigned int a = indices[triangle * 3 + 0], b = indices[triangle * 3 + 1], c = indices[triangle * 3 + 2];
+
+			unsigned int live = live_triangles[a] + live_triangles[b] + live_triangles[c];
+
+			if (live < best_neighbor_live)
+			{
+				best_neighbor = triangle;
+				best_neighbor_live = live;
+			}
+		}
+
+		// add the neighbor to the list of seeds; the list is unsorted and the replacement criteria is approximate
+		if (best_neighbor == ~0u)
+			continue;
+
+		float best_neighbor_score = getDistance(triangles[best_neighbor].px - cornerx, triangles[best_neighbor].py - cornery, triangles[best_neighbor].pz - cornerz, false);
+
+		for (size_t j = 0; j < kMeshletAddSeeds; ++j)
+		{
+			// non-strict comparison reduces the number of duplicate seeds (triangles adjacent to multiple vertices)
+			if (best_neighbor_live < best_live[j] || (best_neighbor_live == best_live[j] && best_neighbor_score <= best_score[j]))
+			{
+				best_seeds[j] = best_neighbor;
+				best_live[j] = best_neighbor_live;
+				best_score[j] = best_neighbor_score;
+				break;
+			}
+		}
+	}
+
+	// add surviving seeds to the meshlet
+	size_t seed_count = 0;
+
+	for (size_t i = 0; i < kMeshletAddSeeds; ++i)
+		if (best_seeds[i] != ~0u)
+			seeds[seed_count++] = best_seeds[i];
+
+	return seed_count;
+}
+
+static size_t pruneSeedTriangles(unsigned int* seeds, size_t seed_count, const unsigned char* emitted_flags)
+{
+	size_t result = 0;
+
+	for (size_t i = 0; i < seed_count; ++i)
+	{
+		unsigned int index = seeds[i];
+
+		seeds[result] = index;
+		result += emitted_flags[index] == 0;
+	}
+
+	return result;
+}
+
+static unsigned int selectSeedTriangle(const unsigned int* seeds, size_t seed_count, const unsigned int* indices, const Cone* triangles, const unsigned int* live_triangles, float cornerx, float cornery, float cornerz)
+{
+	unsigned int best_seed = ~0u;
+	unsigned int best_live = ~0u;
+	float best_score = FLT_MAX;
+
+	for (size_t i = 0; i < seed_count; ++i)
+	{
+		unsigned int index = seeds[i];
+		unsigned int a = indices[index * 3 + 0], b = indices[index * 3 + 1], c = indices[index * 3 + 2];
+
+		unsigned int live = live_triangles[a] + live_triangles[b] + live_triangles[c];
+		float score = getDistance(triangles[index].px - cornerx, triangles[index].py - cornery, triangles[index].pz - cornerz, false);
+
+		if (live < best_live || (live == best_live && score < best_score))
+		{
+			best_seed = index;
+			best_live = live;
+			best_score = score;
+		}
+	}
+
+	return best_seed;
+}
+
 struct KDNode
 {
 	union
@@ -533,7 +657,7 @@ static size_t kdtreeBuild(size_t offset, KDNode* nodes, size_t node_count, const
 	return kdtreeBuild(next_offset, nodes, node_count, points, stride, indices + middle, count - middle, leaf_size);
 }
 
-static void kdtreeNearest(KDNode* nodes, unsigned int root, const float* points, size_t stride, const unsigned char* emitted_flags, const float* position, unsigned int& result, float& limit)
+static void kdtreeNearest(KDNode* nodes, unsigned int root, const float* points, size_t stride, const unsigned char* emitted_flags, const float* position, bool aa, unsigned int& result, float& limit)
 {
 	const KDNode& node = nodes[root];
 
@@ -549,11 +673,8 @@ static void kdtreeNearest(KDNode* nodes, unsigned int root, const float* points,
 
 			const float* point = points + index * stride;
 
-			float distance2 =
-			    (point[0] - position[0]) * (point[0] - position[0]) +
-			    (point[1] - position[1]) * (point[1] - position[1]) +
-			    (point[2] - position[2]) * (point[2] - position[2]);
-			float distance = sqrtf(distance2);
+			float dx = point[0] - position[0], dy = point[1] - position[1], dz = point[2] - position[2];
+			float distance = getDistance(dx, dy, dz, aa);
 
 			if (distance < limit)
 			{
@@ -569,11 +690,11 @@ static void kdtreeNearest(KDNode* nodes, unsigned int root, const float* points,
 		unsigned int first = (delta <= 0) ? 0 : node.children;
 		unsigned int second = first ^ node.children;
 
-		kdtreeNearest(nodes, root + 1 + first, points, stride, emitted_flags, position, result, limit);
+		kdtreeNearest(nodes, root + 1 + first, points, stride, emitted_flags, position, aa, result, limit);
 
 		// only process the other node if it can have a match based on closest distance so far
 		if (fabsf(delta) <= limit)
-			kdtreeNearest(nodes, root + 1 + second, points, stride, emitted_flags, position, result, limit);
+			kdtreeNearest(nodes, root + 1 + second, points, stride, emitted_flags, position, aa, result, limit);
 	}
 }
 
@@ -601,7 +722,7 @@ size_t meshopt_buildMeshletsBound(size_t index_count, size_t max_vertices, size_
 	return meshlet_limit_vertices > meshlet_limit_triangles ? meshlet_limit_vertices : meshlet_limit_triangles;
 }
 
-size_t meshopt_buildMeshlets(meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t max_vertices, size_t max_triangles, float cone_weight)
+size_t meshopt_buildMeshletsFlex(meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t max_vertices, size_t min_triangles, size_t max_triangles, float cone_weight, float split_factor)
 {
 	using namespace meshopt;
 
@@ -610,10 +731,14 @@ size_t meshopt_buildMeshlets(meshopt_Meshlet* meshlets, unsigned int* meshlet_ve
 	assert(vertex_positions_stride % sizeof(float) == 0);
 
 	assert(max_vertices >= 3 && max_vertices <= kMeshletMaxVertices);
-	assert(max_triangles >= 1 && max_triangles <= kMeshletMaxTriangles);
-	assert(max_triangles % 4 == 0); // ensures the caller will compute output space properly as index data is 4b aligned
+	assert(min_triangles >= 1 && min_triangles <= max_triangles && max_triangles <= kMeshletMaxTriangles);
+	assert(min_triangles % 4 == 0 && max_triangles % 4 == 0); // ensures the caller will compute output space properly as index data is 4b aligned
 
-	assert(cone_weight >= 0 && cone_weight <= 1);
+	assert(cone_weight <= 1); // negative cone weight switches metric to optimize for axis-aligned meshlets
+	assert(split_factor >= 0);
+
+	if (index_count == 0)
+		return 0;
 
 	meshopt_Allocator allocator;
 
@@ -647,9 +772,42 @@ size_t meshopt_buildMeshlets(meshopt_Meshlet* meshlets, unsigned int* meshlet_ve
 	KDNode* nodes = allocator.allocate<KDNode>(face_count * 2);
 	kdtreeBuild(0, nodes, face_count * 2, &triangles[0].px, sizeof(Cone) / sizeof(float), kdindices, face_count, /* leaf_size= */ 8);
 
-	// index of the vertex in the meshlet, 0xff if the vertex isn't used
-	unsigned char* used = allocator.allocate<unsigned char>(vertex_count);
-	memset(used, -1, vertex_count);
+	// find a specific corner of the mesh to use as a starting point for meshlet flow
+	float cornerx = FLT_MAX, cornery = FLT_MAX, cornerz = FLT_MAX;
+
+	for (size_t i = 0; i < face_count; ++i)
+	{
+		const Cone& tri = triangles[i];
+
+		cornerx = cornerx > tri.px ? tri.px : cornerx;
+		cornery = cornery > tri.py ? tri.py : cornery;
+		cornerz = cornerz > tri.pz ? tri.pz : cornerz;
+	}
+
+	// index of the vertex in the meshlet, -1 if the vertex isn't used
+	short* used = allocator.allocate<short>(vertex_count);
+	memset(used, -1, vertex_count * sizeof(short));
+
+	// initial seed triangle is the one closest to the corner
+	unsigned int initial_seed = ~0u;
+	float initial_score = FLT_MAX;
+
+	for (size_t i = 0; i < face_count; ++i)
+	{
+		const Cone& tri = triangles[i];
+
+		float score = getDistance(tri.px - cornerx, tri.py - cornery, tri.pz - cornerz, false);
+
+		if (initial_seed == ~0u || score < initial_score)
+		{
+			initial_seed = unsigned(i);
+			initial_score = score;
+		}
+	}
+
+	// seed triangles to continue meshlet flow
+	unsigned int seeds[kMeshletMaxSeeds] = {};
+	size_t seed_count = 0;
 
 	meshopt_Meshlet meshlet = {};
 	size_t meshlet_offset = 0;
@@ -660,35 +818,53 @@ size_t meshopt_buildMeshlets(meshopt_Meshlet* meshlets, unsigned int* meshlet_ve
 	{
 		Cone meshlet_cone = getMeshletCone(meshlet_cone_acc, meshlet.triangle_count);
 
-		unsigned int best_triangle = getNeighborTriangle(meshlet, &meshlet_cone, meshlet_vertices, indices, adjacency, triangles, live_triangles, used, meshlet_expected_radius, cone_weight);
-		int best_extra = best_triangle == ~0u ? -1 : (used[indices[best_triangle * 3 + 0]] == 0xff) + (used[indices[best_triangle * 3 + 1]] == 0xff) + (used[indices[best_triangle * 3 + 2]] == 0xff);
+		unsigned int best_triangle = ~0u;
 
-		// if the best triangle doesn't fit into current meshlet, the spatial scoring we've used is not very meaningful, so we re-select using topological scoring
-		if (best_triangle != ~0u && (meshlet.vertex_count + best_extra > max_vertices || meshlet.triangle_count >= max_triangles))
-		{
-			best_triangle = getNeighborTriangle(meshlet, NULL, meshlet_vertices, indices, adjacency, triangles, live_triangles, used, meshlet_expected_radius, 0.f);
-		}
+		// for the first triangle, we don't have a meshlet cone yet, so we use the initial seed
+		// to continue the meshlet, we select an adjacent triangle based on connectivity and spatial scoring
+		if (meshlet_offset == 0 && meshlet.triangle_count == 0)
+			best_triangle = initial_seed;
+		else
+			best_triangle = getNeighborTriangle(meshlet, meshlet_cone, meshlet_vertices, indices, adjacency, triangles, live_triangles, used, meshlet_expected_radius, cone_weight);
 
-		// when we run out of neighboring triangles we need to switch to spatial search; we currently just pick the closest triangle irrespective of connectivity
+		bool split = false;
+
+		// when we run out of adjacent triangles we need to switch to spatial search; we currently just pick the closest triangle irrespective of connectivity
 		if (best_triangle == ~0u)
 		{
 			float position[3] = {meshlet_cone.px, meshlet_cone.py, meshlet_cone.pz};
 			unsigned int index = ~0u;
-			float limit = FLT_MAX;
+			float distance = FLT_MAX;
 
-			kdtreeNearest(nodes, 0, &triangles[0].px, sizeof(Cone) / sizeof(float), emitted_flags, position, index, limit);
+			kdtreeNearest(nodes, 0, &triangles[0].px, sizeof(Cone) / sizeof(float), emitted_flags, position, cone_weight < 0.f, index, distance);
 
 			best_triangle = index;
+			split = meshlet.triangle_count >= min_triangles && split_factor > 0 && distance > meshlet_expected_radius * split_factor;
 		}
 
 		if (best_triangle == ~0u)
 			break;
 
+		int best_extra = (used[indices[best_triangle * 3 + 0]] < 0) + (used[indices[best_triangle * 3 + 1]] < 0) + (used[indices[best_triangle * 3 + 2]] < 0);
+
+		// if the best triangle doesn't fit into current meshlet, we re-select using seeds to maintain global flow
+		if (split || (meshlet.vertex_count + best_extra > max_vertices || meshlet.triangle_count >= max_triangles))
+		{
+			seed_count = pruneSeedTriangles(seeds, seed_count, emitted_flags);
+			seed_count = (seed_count + kMeshletAddSeeds <= kMeshletMaxSeeds) ? seed_count : kMeshletMaxSeeds - kMeshletAddSeeds;
+			seed_count += appendSeedTriangles(seeds + seed_count, meshlet, meshlet_vertices, indices, adjacency, triangles, live_triangles, cornerx, cornery, cornerz);
+
+			unsigned int best_seed = selectSeedTriangle(seeds, seed_count, indices, triangles, live_triangles, cornerx, cornery, cornerz);
+
+			// we may not find a valid seed triangle if the mesh is disconnected as seeds are based on adjacency
+			best_triangle = best_seed != ~0u ? best_seed : best_triangle;
+		}
+
 		unsigned int a = indices[best_triangle * 3 + 0], b = indices[best_triangle * 3 + 1], c = indices[best_triangle * 3 + 2];
 		assert(a < vertex_count && b < vertex_count && c < vertex_count);
 
 		// add meshlet to the output; when the current meshlet is full we reset the accumulated bounds
-		if (appendMeshlet(meshlet, a, b, c, used, meshlets, meshlet_vertices, meshlet_triangles, meshlet_offset, max_vertices, max_triangles))
+		if (appendMeshlet(meshlet, a, b, c, used, meshlets, meshlet_vertices, meshlet_triangles, meshlet_offset, max_vertices, max_triangles, split))
 		{
 			meshlet_offset++;
 			memset(&meshlet_cone_acc, 0, sizeof(meshlet_cone_acc));
@@ -725,6 +901,7 @@ size_t meshopt_buildMeshlets(meshopt_Meshlet* meshlets, unsigned int* meshlet_ve
 		meshlet_cone_acc.ny += triangles[best_triangle].ny;
 		meshlet_cone_acc.nz += triangles[best_triangle].nz;
 
+		assert(!emitted_flags[best_triangle]);
 		emitted_flags[best_triangle] = 1;
 	}
 
@@ -735,10 +912,17 @@ size_t meshopt_buildMeshlets(meshopt_Meshlet* meshlets, unsigned int* meshlet_ve
 		meshlets[meshlet_offset++] = meshlet;
 	}
 
-	assert(meshlet_offset <= meshopt_buildMeshletsBound(index_count, max_vertices, max_triangles));
+	assert(meshlet_offset <= meshopt_buildMeshletsBound(index_count, max_vertices, min_triangles));
 	return meshlet_offset;
 }
 
+size_t meshopt_buildMeshlets(meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t max_vertices, size_t max_triangles, float cone_weight)
+{
+	assert(cone_weight >= 0); // to use negative cone weight, use meshopt_buildMeshletsFlex
+
+	return meshopt_buildMeshletsFlex(meshlets, meshlet_vertices, meshlet_triangles, indices, index_count, vertex_positions, vertex_count, vertex_positions_stride, max_vertices, max_triangles, max_triangles, cone_weight, 0.0f);
+}
+
 size_t meshopt_buildMeshletsScan(meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const unsigned int* indices, size_t index_count, size_t vertex_count, size_t max_vertices, size_t max_triangles)
 {
 	using namespace meshopt;
@@ -751,9 +935,9 @@ size_t meshopt_buildMeshletsScan(meshopt_Meshlet* meshlets, unsigned int* meshle
 
 	meshopt_Allocator allocator;
 
-	// index of the vertex in the meshlet, 0xff if the vertex isn't used
-	unsigned char* used = allocator.allocate<unsigned char>(vertex_count);
-	memset(used, -1, vertex_count);
+	// index of the vertex in the meshlet, -1 if the vertex isn't used
+	short* used = allocator.allocate<short>(vertex_count);
+	memset(used, -1, vertex_count * sizeof(short));
 
 	meshopt_Meshlet meshlet = {};
 	size_t meshlet_offset = 0;
@@ -834,15 +1018,17 @@ meshopt_Bounds meshopt_computeClusterBounds(const unsigned int* indices, size_t
 	if (triangles == 0)
 		return bounds;
 
+	const float rzero = 0.f;
+
 	// compute cluster bounding sphere; we'll use the center to determine normal cone apex as well
 	float psphere[4] = {};
-	computeBoundingSphere(psphere, corners[0], triangles * 3);
+	computeBoundingSphere(psphere, corners[0][0], triangles * 3, sizeof(float) * 3, &rzero, 0);
 
 	float center[3] = {psphere[0], psphere[1], psphere[2]};
 
 	// treating triangle normals as points, find the bounding sphere - the sphere center determines the optimal cone axis
 	float nsphere[4] = {};
-	computeBoundingSphere(nsphere, normals, triangles);
+	computeBoundingSphere(nsphere, normals[0], triangles, sizeof(float) * 3, &rzero, 0);
 
 	float axis[3] = {nsphere[0], nsphere[1], nsphere[2]};
 	float axislength = sqrtf(axis[0] * axis[0] + axis[1] * axis[1] + axis[2] * axis[2]);
@@ -952,6 +1138,33 @@ meshopt_Bounds meshopt_computeMeshletBounds(const unsigned int* meshlet_vertices
 	return meshopt_computeClusterBounds(indices, triangle_count * 3, vertex_positions, vertex_count, vertex_positions_stride);
 }
 
+meshopt_Bounds meshopt_computeSphereBounds(const float* positions, size_t count, size_t positions_stride, const float* radii, size_t radii_stride)
+{
+	using namespace meshopt;
+
+	assert(positions_stride >= 12 && positions_stride <= 256);
+	assert(positions_stride % sizeof(float) == 0);
+	assert((radii_stride >= 4 && radii_stride <= 256) || radii == NULL);
+	assert(radii_stride % sizeof(float) == 0);
+
+	meshopt_Bounds bounds = {};
+
+	if (count == 0)
+		return bounds;
+
+	const float rzero = 0.f;
+
+	float psphere[4] = {};
+	computeBoundingSphere(psphere, positions, count, positions_stride, radii ? radii : &rzero, radii ? radii_stride : 0);
+
+	bounds.center[0] = psphere[0];
+	bounds.center[1] = psphere[1];
+	bounds.center[2] = psphere[2];
+	bounds.radius = psphere[3];
+
+	return bounds;
+}
+
 void meshopt_optimizeMeshlet(unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, size_t triangle_count, size_t vertex_count)
 {
 	using namespace meshopt;
@@ -1019,23 +1232,23 @@ void meshopt_optimizeMeshlet(unsigned int* meshlet_vertices, unsigned char* mesh
 	// reorder meshlet vertices for access locality assuming index buffer is scanned sequentially
 	unsigned int order[kMeshletMaxVertices];
 
-	unsigned char remap[kMeshletMaxVertices];
-	memset(remap, -1, vertex_count);
+	short remap[kMeshletMaxVertices];
+	memset(remap, -1, vertex_count * sizeof(short));
 
 	size_t vertex_offset = 0;
 
 	for (size_t i = 0; i < triangle_count * 3; ++i)
 	{
-		unsigned char& r = remap[indices[i]];
+		short& r = remap[indices[i]];
 
-		if (r == 0xff)
+		if (r < 0)
 		{
-			r = (unsigned char)(vertex_offset);
+			r = short(vertex_offset);
 			order[vertex_offset] = vertices[indices[i]];
 			vertex_offset++;
 		}
 
-		indices[i] = r;
+		indices[i] = (unsigned char)r;
 	}
 
 	assert(vertex_offset <= vertex_count);
diff --git a/3rdparty/meshoptimizer/src/indexcodec.cpp b/3rdparty/meshoptimizer/src/indexcodec.cpp
index b30046005..b4fdfe16d 100644
--- a/3rdparty/meshoptimizer/src/indexcodec.cpp
+++ b/3rdparty/meshoptimizer/src/indexcodec.cpp
@@ -14,6 +14,7 @@ const unsigned char kIndexHeader = 0xe0;
 const unsigned char kSequenceHeader = 0xd0;
 
 static int gEncodeIndexVersion = 1;
+const int kDecodeIndexVersion = 1;
 
 typedef unsigned int VertexFifo[16];
 typedef unsigned int EdgeFifo[16][2];
@@ -354,11 +355,28 @@ size_t meshopt_encodeIndexBufferBound(size_t index_count, size_t vertex_count)
 
 void meshopt_encodeIndexVersion(int version)
 {
-	assert(unsigned(version) <= 1);
+	assert(unsigned(version) <= unsigned(meshopt::kDecodeIndexVersion));
 
 	meshopt::gEncodeIndexVersion = version;
 }
 
+int meshopt_decodeIndexVersion(const unsigned char* buffer, size_t buffer_size)
+{
+	if (buffer_size < 1)
+		return -1;
+
+	unsigned char header = buffer[0];
+
+	if ((header & 0xf0) != meshopt::kIndexHeader && (header & 0xf0) != meshopt::kSequenceHeader)
+		return -1;
+
+	int version = header & 0x0f;
+	if (version > meshopt::kDecodeIndexVersion)
+		return -1;
+
+	return version;
+}
+
 int meshopt_decodeIndexBuffer(void* destination, size_t index_count, size_t index_size, const unsigned char* buffer, size_t buffer_size)
 {
 	using namespace meshopt;
@@ -374,7 +392,7 @@ int meshopt_decodeIndexBuffer(void* destination, size_t index_count, size_t inde
 		return -1;
 
 	int version = buffer[0] & 0x0f;
-	if (version > 1)
+	if (version > kDecodeIndexVersion)
 		return -1;
 
 	EdgeFifo edgefifo;
@@ -627,7 +645,7 @@ int meshopt_decodeIndexSequence(void* destination, size_t index_count, size_t in
 		return -1;
 
 	int version = buffer[0] & 0x0f;
-	if (version > 1)
+	if (version > kDecodeIndexVersion)
 		return -1;
 
 	const unsigned char* data = buffer + 1;
diff --git a/3rdparty/meshoptimizer/src/meshoptimizer.h b/3rdparty/meshoptimizer/src/meshoptimizer.h
index baeba1522..295324c78 100644
--- a/3rdparty/meshoptimizer/src/meshoptimizer.h
+++ b/3rdparty/meshoptimizer/src/meshoptimizer.h
@@ -1,5 +1,5 @@
 /**
- * meshoptimizer - version 0.22
+ * meshoptimizer - version 0.23
  *
  * Copyright (C) 2016-2025, by Arseny Kapoulkine (arseny.kapoulkine@gmail.com)
  * Report bugs and download new versions at https://github.com/zeux/meshoptimizer
@@ -12,7 +12,7 @@
 #include <stddef.h>
 
 /* Version macro; major * 1000 + minor * 10 + patch */
-#define MESHOPTIMIZER_VERSION 220 /* 0.22 */
+#define MESHOPTIMIZER_VERSION 230 /* 0.23 */
 
 /* If no API is defined, assume default */
 #ifndef MESHOPTIMIZER_API
@@ -243,6 +243,13 @@ MESHOPTIMIZER_API void meshopt_encodeIndexVersion(int version);
  */
 MESHOPTIMIZER_API int meshopt_decodeIndexBuffer(void* destination, size_t index_count, size_t index_size, const unsigned char* buffer, size_t buffer_size);
 
+/**
+ * Get encoded index format version
+ * Returns format version of the encoded index buffer/sequence, or -1 if the buffer header is invalid
+ * Note that a non-negative value doesn't guarantee that the buffer will be decoded correctly if the input is malformed.
+ */
+MESHOPTIMIZER_API int meshopt_decodeIndexVersion(const unsigned char* buffer, size_t buffer_size);
+
 /**
  * Index sequence encoder
  * Encodes index sequence into an array of bytes that is generally smaller and compresses better compared to original.
@@ -285,7 +292,7 @@ MESHOPTIMIZER_API size_t meshopt_encodeVertexBufferBound(size_t vertex_count, si
  *
  * level should be in the range [0, 3] with 0 being the fastest and 3 being the slowest and producing the best compression ratio.
  */
-MESHOPTIMIZER_API size_t meshopt_encodeVertexBufferLevel(unsigned char* buffer, size_t buffer_size, const void* vertices, size_t vertex_count, size_t vertex_size, int level);
+MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_encodeVertexBufferLevel(unsigned char* buffer, size_t buffer_size, const void* vertices, size_t vertex_count, size_t vertex_size, int level);
 
 /**
  * Set vertex encoder format version
@@ -303,6 +310,13 @@ MESHOPTIMIZER_API void meshopt_encodeVertexVersion(int version);
  */
 MESHOPTIMIZER_API int meshopt_decodeVertexBuffer(void* destination, size_t vertex_count, size_t vertex_size, const unsigned char* buffer, size_t buffer_size);
 
+/**
+ * Get encoded vertex format version
+ * Returns format version of the encoded vertex buffer, or -1 if the buffer header is invalid
+ * Note that a non-negative value doesn't guarantee that the buffer will be decoded correctly if the input is malformed.
+ */
+MESHOPTIMIZER_API int meshopt_decodeVertexVersion(const unsigned char* buffer, size_t buffer_size);
+
 /**
  * Vertex buffer filters
  * These functions can be used to filter output of meshopt_decodeVertexBuffer in-place.
@@ -344,7 +358,7 @@ enum meshopt_EncodeExpMode
 	meshopt_EncodeExpSharedVector,
 	/* When encoding exponents, use shared value for each component of all vectors (best compression) */
 	meshopt_EncodeExpSharedComponent,
-	/* Experimental: When encoding exponents, use separate values for each component, but clamp to 0 (good quality if very small values are not important) */
+	/* When encoding exponents, use separate values for each component, but clamp to 0 (good quality if very small values are not important) */
 	meshopt_EncodeExpClamped,
 };
 
@@ -385,7 +399,7 @@ enum
 MESHOPTIMIZER_API size_t meshopt_simplify(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error, unsigned int options, float* result_error);
 
 /**
- * Experimental: Mesh simplifier with attribute metric
+ * Mesh simplifier with attribute metric
  * The algorithm enhances meshopt_simplify by incorporating attribute values into the error metric used to prioritize simplification order; see meshopt_simplify documentation for details.
  * Note that the number of attributes affects memory requirements and running time; this algorithm requires ~1.5x more memory and time compared to meshopt_simplify when using 4 scalar attributes.
  *
@@ -394,7 +408,7 @@ MESHOPTIMIZER_API size_t meshopt_simplify(unsigned int* destination, const unsig
  * attribute_count must be <= 32
  * vertex_lock can be NULL; when it's not NULL, it should have a value for each vertex; 1 denotes vertices that can't be moved
  */
-MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_simplifyWithAttributes(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, const float* vertex_attributes, size_t vertex_attributes_stride, const float* attribute_weights, size_t attribute_count, const unsigned char* vertex_lock, size_t target_index_count, float target_error, unsigned int options, float* result_error);
+MESHOPTIMIZER_API size_t meshopt_simplifyWithAttributes(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, const float* vertex_attributes, size_t vertex_attributes_stride, const float* attribute_weights, size_t attribute_count, const unsigned char* vertex_lock, size_t target_index_count, float target_error, unsigned int options, float* result_error);
 
 /**
  * Experimental: Mesh simplifier (sloppy)
@@ -412,7 +426,7 @@ MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_simplifyWithAttributes(unsigned int* d
 MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_simplifySloppy(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t target_index_count, float target_error, float* result_error);
 
 /**
- * Experimental: Point cloud simplifier
+ * Point cloud simplifier
  * Reduces the number of points in the cloud to reach the given target
  * Returns the number of points after simplification, with destination containing new index data
  * The resulting index buffer references vertices from the original vertex buffer.
@@ -420,10 +434,10 @@ MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_simplifySloppy(unsigned int* destinati
  *
  * destination must contain enough space for the target index buffer (target_vertex_count elements)
  * vertex_positions should have float3 position in the first 12 bytes of each vertex
- * vertex_colors should can be NULL; when it's not NULL, it should have float3 color in the first 12 bytes of each vertex
+ * vertex_colors can be NULL; when it's not NULL, it should have float3 color in the first 12 bytes of each vertex
  * color_weight determines relative priority of color wrt position; 1.0 is a safe default
  */
-MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_simplifyPoints(unsigned int* destination, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, const float* vertex_colors, size_t vertex_colors_stride, float color_weight, size_t target_vertex_count);
+MESHOPTIMIZER_API size_t meshopt_simplifyPoints(unsigned int* destination, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, const float* vertex_colors, size_t vertex_colors_stride, float color_weight, size_t target_vertex_count);
 
 /**
  * Returns the error scaling factor used by the simplifier to convert between absolute and relative extents
@@ -530,7 +544,7 @@ struct meshopt_Meshlet
  * meshlet_vertices must contain enough space for all meshlets, worst case size is equal to max_meshlets * max_vertices
  * meshlet_triangles must contain enough space for all meshlets, worst case size is equal to max_meshlets * max_triangles * 3
  * vertex_positions should have float3 position in the first 12 bytes of each vertex
- * max_vertices and max_triangles must not exceed implementation limits (max_vertices <= 255 - not 256!, max_triangles <= 512; max_triangles must be divisible by 4)
+ * max_vertices and max_triangles must not exceed implementation limits (max_vertices <= 256, max_triangles <= 512; max_triangles must be divisible by 4)
  * cone_weight should be set to 0 when cone culling is not used, and a value between 0 and 1 otherwise to balance between cluster size and cone culling efficiency
  */
 MESHOPTIMIZER_API size_t meshopt_buildMeshlets(struct meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t max_vertices, size_t max_triangles, float cone_weight);
@@ -538,14 +552,30 @@ MESHOPTIMIZER_API size_t meshopt_buildMeshletsScan(struct meshopt_Meshlet* meshl
 MESHOPTIMIZER_API size_t meshopt_buildMeshletsBound(size_t index_count, size_t max_vertices, size_t max_triangles);
 
 /**
- * Experimental: Meshlet optimizer
+ * Experimental: Meshlet builder with flexible cluster sizes
+ * Splits the mesh into a set of meshlets, similarly to meshopt_buildMeshlets, but allows to specify minimum and maximum number of triangles per meshlet.
+ * Clusters between min and max triangle counts are split when the cluster size would have exceeded the expected cluster size by more than split_factor.
+ * Additionally, allows to switch to axis aligned clusters by setting cone_weight to a negative value.
+ *
+ * meshlets must contain enough space for all meshlets, worst case size can be computed with meshopt_buildMeshletsBound using min_triangles (not max!)
+ * meshlet_vertices must contain enough space for all meshlets, worst case size is equal to max_meshlets * max_vertices
+ * meshlet_triangles must contain enough space for all meshlets, worst case size is equal to max_meshlets * max_triangles * 3
+ * vertex_positions should have float3 position in the first 12 bytes of each vertex
+ * max_vertices, min_triangles and max_triangles must not exceed implementation limits (max_vertices <= 256, max_triangles <= 512; min_triangles <= max_triangles; both min_triangles and max_triangles must be divisible by 4)
+ * cone_weight should be set to 0 when cone culling is not used, and a value between 0 and 1 otherwise to balance between cluster size and cone culling efficiency; additionally, cone_weight can be set to a negative value to prioritize axis aligned clusters (for raytracing) instead
+ * split_factor should be set to a non-negative value; when greater than 0, clusters that have large bounds may be split unless they are under the min_triangles threshold
+ */
+MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_buildMeshletsFlex(struct meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t max_vertices, size_t min_triangles, size_t max_triangles, float cone_weight, float split_factor);
+
+/**
+ * Meshlet optimizer
  * Reorders meshlet vertices and triangles to maximize locality to improve rasterizer throughput
  *
  * meshlet_triangles and meshlet_vertices must refer to meshlet triangle and vertex index data; when buildMeshlets* is used, these
  * need to be computed from meshlet's vertex_offset and triangle_offset
- * triangle_count and vertex_count must not exceed implementation limits (vertex_count <= 255 - not 256!, triangle_count <= 512)
+ * triangle_count and vertex_count must not exceed implementation limits (vertex_count <= 256, triangle_count <= 512)
  */
-MESHOPTIMIZER_EXPERIMENTAL void meshopt_optimizeMeshlet(unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, size_t triangle_count, size_t vertex_count);
+MESHOPTIMIZER_API void meshopt_optimizeMeshlet(unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, size_t triangle_count, size_t vertex_count);
 
 struct meshopt_Bounds
 {
@@ -589,6 +619,27 @@ struct meshopt_Bounds
 MESHOPTIMIZER_API struct meshopt_Bounds meshopt_computeClusterBounds(const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride);
 MESHOPTIMIZER_API struct meshopt_Bounds meshopt_computeMeshletBounds(const unsigned int* meshlet_vertices, const unsigned char* meshlet_triangles, size_t triangle_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride);
 
+/**
+ * Experimental: Sphere bounds generator
+ * Creates bounding sphere around a set of points or a set of spheres; returns the center and radius of the sphere, with other fields of the result set to 0.
+ *
+ * positions should have float3 position in the first 12 bytes of each element
+ * radii can be NULL; when it's not NULL, it should have a non-negative float radius in the first 4 bytes of each element
+ */
+MESHOPTIMIZER_EXPERIMENTAL struct meshopt_Bounds meshopt_computeSphereBounds(const float* positions, size_t count, size_t positions_stride, const float* radii, size_t radii_stride);
+
+/**
+ * Experimental: Cluster partitioner
+ * Partitions clusters into groups of similar size, prioritizing grouping clusters that share vertices.
+ *
+ * destination must contain enough space for the resulting partiotion data (cluster_count elements)
+ * destination[i] will contain the partition id for cluster i, with the total number of partitions returned by the function
+ * cluster_indices should have the vertex indices referenced by each cluster, stored sequentially
+ * cluster_index_counts should have the number of indices in each cluster; sum of all cluster_index_counts must be equal to total_index_count
+ * target_partition_size is a target size for each partition, in clusters; the resulting partitions may be smaller or larger
+ */
+MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_partitionClusters(unsigned int* destination, const unsigned int* cluster_indices, size_t total_index_count, const unsigned int* cluster_index_counts, size_t cluster_count, size_t vertex_count, size_t target_partition_size);
+
 /**
  * Spatial sorter
  * Generates a remap table that can be used to reorder points for spatial locality.
@@ -724,8 +775,12 @@ inline size_t meshopt_buildMeshlets(meshopt_Meshlet* meshlets, unsigned int* mes
 template <typename T>
 inline size_t meshopt_buildMeshletsScan(meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const T* indices, size_t index_count, size_t vertex_count, size_t max_vertices, size_t max_triangles);
 template <typename T>
+inline size_t meshopt_buildMeshletsFlex(meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t max_vertices, size_t min_triangles, size_t max_triangles, float cone_weight, float split_factor);
+template <typename T>
 inline meshopt_Bounds meshopt_computeClusterBounds(const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride);
 template <typename T>
+inline size_t meshopt_partitionClusters(unsigned int* destination, const T* cluster_indices, size_t total_index_count, const unsigned int* cluster_index_counts, size_t cluster_count, size_t vertex_count, size_t target_partition_size);
+template <typename T>
 inline void meshopt_spatialSortTriangles(T* destination, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride);
 #endif
 
@@ -1104,6 +1159,14 @@ inline size_t meshopt_buildMeshletsScan(meshopt_Meshlet* meshlets, unsigned int*
 	return meshopt_buildMeshletsScan(meshlets, meshlet_vertices, meshlet_triangles, in.data, index_count, vertex_count, max_vertices, max_triangles);
 }
 
+template <typename T>
+inline size_t meshopt_buildMeshletsFlex(meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t max_vertices, size_t min_triangles, size_t max_triangles, float cone_weight, float split_factor)
+{
+	meshopt_IndexAdapter<T> in(NULL, indices, index_count);
+
+	return meshopt_buildMeshletsFlex(meshlets, meshlet_vertices, meshlet_triangles, in.data, index_count, vertex_positions, vertex_count, vertex_positions_stride, max_vertices, min_triangles, max_triangles, cone_weight, split_factor);
+}
+
 template <typename T>
 inline meshopt_Bounds meshopt_computeClusterBounds(const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride)
 {
@@ -1112,6 +1175,14 @@ inline meshopt_Bounds meshopt_computeClusterBounds(const T* indices, size_t inde
 	return meshopt_computeClusterBounds(in.data, index_count, vertex_positions, vertex_count, vertex_positions_stride);
 }
 
+template <typename T>
+inline size_t meshopt_partitionClusters(unsigned int* destination, const T* cluster_indices, size_t total_index_count, const unsigned int* cluster_index_counts, size_t cluster_count, size_t vertex_count, size_t target_partition_size)
+{
+	meshopt_IndexAdapter<T> in(NULL, cluster_indices, total_index_count);
+
+	return meshopt_partitionClusters(destination, in.data, total_index_count, cluster_index_counts, cluster_count, vertex_count, target_partition_size);
+}
+
 template <typename T>
 inline void meshopt_spatialSortTriangles(T* destination, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride)
 {
diff --git a/3rdparty/meshoptimizer/src/partition.cpp b/3rdparty/meshoptimizer/src/partition.cpp
new file mode 100644
index 000000000..9c2299805
--- /dev/null
+++ b/3rdparty/meshoptimizer/src/partition.cpp
@@ -0,0 +1,429 @@
+// This file is part of meshoptimizer library; see meshoptimizer.h for version/license details
+#include "meshoptimizer.h"
+
+#include <assert.h>
+#include <math.h>
+#include <string.h>
+
+namespace meshopt
+{
+
+struct ClusterAdjacency
+{
+	unsigned int* offsets;
+	unsigned int* clusters;
+	unsigned int* shared;
+};
+
+static void buildClusterAdjacency(ClusterAdjacency& adjacency, const unsigned int* cluster_indices, const unsigned int* cluster_offsets, size_t cluster_count, unsigned char* used, size_t vertex_count, meshopt_Allocator& allocator)
+{
+	unsigned int* ref_offsets = allocator.allocate<unsigned int>(vertex_count + 1);
+
+	// compute number of clusters referenced by each vertex
+	memset(ref_offsets, 0, vertex_count * sizeof(unsigned int));
+
+	for (size_t i = 0; i < cluster_count; ++i)
+	{
+		for (size_t j = cluster_offsets[i]; j < cluster_offsets[i + 1]; ++j)
+		{
+			unsigned int v = cluster_indices[j];
+			assert(v < vertex_count);
+
+			ref_offsets[v] += 1 - used[v];
+			used[v] = 1;
+		}
+
+		for (size_t j = cluster_offsets[i]; j < cluster_offsets[i + 1]; ++j)
+			used[cluster_indices[j]] = 0;
+	}
+
+	// compute (worst-case) number of adjacent clusters for each cluster
+	size_t total_adjacency = 0;
+
+	for (size_t i = 0; i < cluster_count; ++i)
+	{
+		size_t count = 0;
+		for (size_t j = cluster_offsets[i]; j < cluster_offsets[i + 1]; ++j)
+		{
+			unsigned int v = cluster_indices[j];
+			assert(v < vertex_count);
+
+			// worst case is every vertex has a disjoint cluster list
+			count += used[v] ? 0 : ref_offsets[v] - 1;
+			used[v] = 1;
+		}
+
+		// ... but only every other cluster can be adjacent in the end
+		total_adjacency += count < cluster_count - 1 ? count : cluster_count - 1;
+
+		for (size_t j = cluster_offsets[i]; j < cluster_offsets[i + 1]; ++j)
+			used[cluster_indices[j]] = 0;
+	}
+
+	// we can now allocate adjacency buffers
+	adjacency.offsets = allocator.allocate<unsigned int>(cluster_count + 1);
+	adjacency.clusters = allocator.allocate<unsigned int>(total_adjacency);
+	adjacency.shared = allocator.allocate<unsigned int>(total_adjacency);
+
+	// convert ref counts to offsets
+	size_t total_refs = 0;
+
+	for (size_t i = 0; i < vertex_count; ++i)
+	{
+		size_t count = ref_offsets[i];
+		ref_offsets[i] = unsigned(total_refs);
+		total_refs += count;
+	}
+
+	unsigned int* ref_data = allocator.allocate<unsigned int>(total_refs);
+
+	// fill cluster refs for each vertex
+	for (size_t i = 0; i < cluster_count; ++i)
+	{
+		for (size_t j = cluster_offsets[i]; j < cluster_offsets[i + 1]; ++j)
+		{
+			unsigned int v = cluster_indices[j];
+			assert(v < vertex_count);
+
+			if (used[v])
+				continue;
+
+			ref_data[ref_offsets[v]++] = unsigned(i);
+			used[v] = 1;
+		}
+
+		for (size_t j = cluster_offsets[i]; j < cluster_offsets[i + 1]; ++j)
+			used[cluster_indices[j]] = 0;
+	}
+
+	// after the previous pass, ref_offsets contain the end of the data for each vertex; shift it forward to get the start
+	memmove(ref_offsets + 1, ref_offsets, vertex_count * sizeof(unsigned int));
+	ref_offsets[0] = 0;
+
+	// fill cluster adjacency for each cluster...
+	adjacency.offsets[0] = 0;
+
+	for (size_t i = 0; i < cluster_count; ++i)
+	{
+		unsigned int* adj = adjacency.clusters + adjacency.offsets[i];
+		unsigned int* shd = adjacency.shared + adjacency.offsets[i];
+		size_t count = 0;
+
+		for (size_t j = cluster_offsets[i]; j < cluster_offsets[i + 1]; ++j)
+		{
+			unsigned int v = cluster_indices[j];
+			assert(v < vertex_count);
+
+			if (used[v])
+				continue;
+
+			// merge the entire cluster list of each vertex into current list
+			for (size_t k = ref_offsets[v]; k < ref_offsets[v + 1]; ++k)
+			{
+				unsigned int c = ref_data[k];
+				assert(c < cluster_count);
+
+				if (c == unsigned(i))
+					continue;
+
+				// if the cluster is already in the list, increment the shared count
+				bool found = false;
+				for (size_t l = 0; l < count; ++l)
+					if (adj[l] == c)
+					{
+						found = true;
+						shd[l]++;
+						break;
+					}
+
+				// .. or append a new cluster
+				if (!found)
+				{
+					adj[count] = c;
+					shd[count] = 1;
+					count++;
+				}
+			}
+
+			used[v] = 1;
+		}
+
+		for (size_t j = cluster_offsets[i]; j < cluster_offsets[i + 1]; ++j)
+			used[cluster_indices[j]] = 0;
+
+		// mark the end of the adjacency list; the next cluster will start there as well
+		adjacency.offsets[i + 1] = adjacency.offsets[i] + unsigned(count);
+	}
+
+	assert(adjacency.offsets[cluster_count] <= total_adjacency);
+
+	// ref_offsets can't be deallocated as it was allocated before adjacency
+	allocator.deallocate(ref_data);
+}
+
+struct ClusterGroup
+{
+	int group;
+	int next;
+	unsigned int size; // 0 unless root
+	unsigned int vertices;
+};
+
+struct GroupOrder
+{
+	unsigned int id;
+	int order;
+};
+
+static void heapPush(GroupOrder* heap, size_t size, GroupOrder item)
+{
+	// insert a new element at the end (breaks heap invariant)
+	heap[size++] = item;
+
+	// bubble up the new element to its correct position
+	size_t i = size - 1;
+	while (i > 0 && heap[i].order < heap[(i - 1) / 2].order)
+	{
+		size_t p = (i - 1) / 2;
+
+		GroupOrder temp = heap[i];
+		heap[i] = heap[p];
+		heap[p] = temp;
+		i = p;
+	}
+}
+
+static GroupOrder heapPop(GroupOrder* heap, size_t size)
+{
+	assert(size > 0);
+	GroupOrder top = heap[0];
+
+	// move the last element to the top (breaks heap invariant)
+	heap[0] = heap[--size];
+
+	// bubble down the new top element to its correct position
+	size_t i = 0;
+	while (i * 2 + 1 < size)
+	{
+		// find the smallest child
+		size_t j = i * 2 + 1;
+		j += (j + 1 < size && heap[j + 1].order < heap[j].order);
+
+		// if the parent is already smaller than both children, we're done
+		if (heap[j].order >= heap[i].order)
+			break;
+
+		// otherwise, swap the parent and child and continue
+		GroupOrder temp = heap[i];
+		heap[i] = heap[j];
+		heap[j] = temp;
+		i = j;
+	}
+
+	return top;
+}
+
+static unsigned int countTotal(const ClusterGroup* groups, int id, const unsigned int* cluster_indices, const unsigned int* cluster_offsets, unsigned char* used)
+{
+	unsigned int total = 0;
+
+	for (int i = id; i >= 0; i = groups[i].next)
+	{
+		for (size_t j = cluster_offsets[i]; j < cluster_offsets[i + 1]; ++j)
+		{
+			unsigned int v = cluster_indices[j];
+			total += 1 - used[v];
+			used[v] = 1;
+		}
+	}
+
+	for (int i = id; i >= 0; i = groups[i].next)
+	{
+		for (size_t j = cluster_offsets[i]; j < cluster_offsets[i + 1]; ++j)
+			used[cluster_indices[j]] = 0;
+	}
+
+	return total;
+}
+
+static unsigned int countShared(const ClusterGroup* groups, int group1, int group2, const ClusterAdjacency& adjacency)
+{
+	unsigned int total = 0;
+
+	for (int i1 = group1; i1 >= 0; i1 = groups[i1].next)
+		for (int i2 = group2; i2 >= 0; i2 = groups[i2].next)
+		{
+			for (unsigned int adj = adjacency.offsets[i1]; adj < adjacency.offsets[i1 + 1]; ++adj)
+				if (adjacency.clusters[adj] == unsigned(i2))
+				{
+					total += adjacency.shared[adj];
+					break;
+				}
+		}
+
+	return total;
+}
+
+static int pickGroupToMerge(const ClusterGroup* groups, int id, const ClusterAdjacency& adjacency, size_t max_partition_size)
+{
+	assert(groups[id].size > 0);
+
+	float group_rsqrt = 1.f / sqrtf(float(int(groups[id].vertices)));
+
+	int best_group = -1;
+	float best_score = 0;
+
+	for (int ci = id; ci >= 0; ci = groups[ci].next)
+	{
+		for (unsigned int adj = adjacency.offsets[ci]; adj != adjacency.offsets[ci + 1]; ++adj)
+		{
+			int other = groups[adjacency.clusters[adj]].group;
+			if (other < 0)
+				continue;
+
+			assert(groups[other].size > 0);
+			if (groups[id].size + groups[other].size > max_partition_size)
+				continue;
+
+			unsigned int shared = countShared(groups, id, other, adjacency);
+			float other_rsqrt = 1.f / sqrtf(float(int(groups[other].vertices)));
+
+			// normalize shared count by the expected boundary of each group (+ keeps scoring symmetric)
+			float score = float(int(shared)) * (group_rsqrt + other_rsqrt);
+
+			if (score > best_score)
+			{
+				best_group = other;
+				best_score = score;
+			}
+		}
+	}
+
+	return best_group;
+}
+
+} // namespace meshopt
+
+size_t meshopt_partitionClusters(unsigned int* destination, const unsigned int* cluster_indices, size_t total_index_count, const unsigned int* cluster_index_counts, size_t cluster_count, size_t vertex_count, size_t target_partition_size)
+{
+	using namespace meshopt;
+
+	assert(target_partition_size > 0);
+
+	size_t max_partition_size = target_partition_size + target_partition_size * 3 / 8;
+
+	meshopt_Allocator allocator;
+
+	unsigned char* used = allocator.allocate<unsigned char>(vertex_count);
+	memset(used, 0, vertex_count);
+
+	// build cluster index offsets as a prefix sum
+	unsigned int* cluster_offsets = allocator.allocate<unsigned int>(cluster_count + 1);
+	unsigned int cluster_nextoffset = 0;
+
+	for (size_t i = 0; i < cluster_count; ++i)
+	{
+		assert(cluster_index_counts[i] > 0);
+
+		cluster_offsets[i] = cluster_nextoffset;
+		cluster_nextoffset += cluster_index_counts[i];
+	}
+
+	assert(cluster_nextoffset == total_index_count);
+	cluster_offsets[cluster_count] = unsigned(total_index_count);
+
+	// build cluster adjacency along with edge weights (shared vertex count)
+	ClusterAdjacency adjacency = {};
+	buildClusterAdjacency(adjacency, cluster_indices, cluster_offsets, cluster_count, used, vertex_count, allocator);
+
+	ClusterGroup* groups = allocator.allocate<ClusterGroup>(cluster_count);
+
+	GroupOrder* order = allocator.allocate<GroupOrder>(cluster_count);
+	size_t pending = 0;
+
+	// create a singleton group for each cluster and order them by priority
+	for (size_t i = 0; i < cluster_count; ++i)
+	{
+		groups[i].group = int(i);
+		groups[i].next = -1;
+		groups[i].size = 1;
+		groups[i].vertices = countTotal(groups, int(i), cluster_indices, cluster_offsets, used);
+
+		GroupOrder item = {};
+		item.id = unsigned(i);
+		item.order = groups[i].vertices;
+
+		heapPush(order, pending++, item);
+	}
+
+	// iteratively merge the smallest group with the best group
+	while (pending)
+	{
+		GroupOrder top = heapPop(order, pending--);
+
+		// this group was merged into another group earlier
+		if (groups[top.id].size == 0)
+			continue;
+
+		// disassociate clusters from the group to prevent them from being merged again; we will re-associate them if the group is reinserted
+		for (int i = top.id; i >= 0; i = groups[i].next)
+		{
+			assert(groups[i].group == int(top.id));
+			groups[i].group = -1;
+		}
+
+		// the group is large enough, emit as is
+		if (groups[top.id].size >= target_partition_size)
+			continue;
+
+		int best_group = pickGroupToMerge(groups, top.id, adjacency, max_partition_size);
+
+		// we can't grow the group any more, emit as is
+		if (best_group == -1)
+			continue;
+
+		// compute shared vertices to adjust the total vertices estimate after merging
+		unsigned int shared = countShared(groups, top.id, best_group, adjacency);
+
+		// combine groups by linking them together
+		assert(groups[best_group].size > 0);
+
+		for (int i = top.id; i >= 0; i = groups[i].next)
+			if (groups[i].next < 0)
+			{
+				groups[i].next = best_group;
+				break;
+			}
+
+		// update group sizes; note, the vertex update is an approximation which avoids recomputing the true size via countTotal
+		groups[top.id].size += groups[best_group].size;
+		groups[top.id].vertices += groups[best_group].vertices;
+		groups[top.id].vertices = (groups[top.id].vertices > shared) ? groups[top.id].vertices - shared : 1;
+
+		groups[best_group].size = 0;
+		groups[best_group].vertices = 0;
+
+		// re-associate all clusters back to the merged group
+		for (int i = top.id; i >= 0; i = groups[i].next)
+			groups[i].group = int(top.id);
+
+		top.order = groups[top.id].vertices;
+		heapPush(order, pending++, top);
+	}
+
+	size_t next_group = 0;
+
+	for (size_t i = 0; i < cluster_count; ++i)
+	{
+		if (groups[i].size == 0)
+			continue;
+
+		for (int j = int(i); j >= 0; j = groups[j].next)
+			destination[j] = unsigned(next_group);
+
+		next_group++;
+	}
+
+	assert(next_group <= cluster_count);
+	return next_group;
+}
diff --git a/3rdparty/meshoptimizer/src/simplifier.cpp b/3rdparty/meshoptimizer/src/simplifier.cpp
index d464fc607..cf0a8a187 100644
--- a/3rdparty/meshoptimizer/src/simplifier.cpp
+++ b/3rdparty/meshoptimizer/src/simplifier.cpp
@@ -437,8 +437,13 @@ static void classifyVertices(unsigned char* result, unsigned int* loop, unsigned
 	{
 		// vertex_lock may lock any wedge, not just the primary vertex, so we need to lock the primary vertex and relock any wedges
 		for (size_t i = 0; i < vertex_count; ++i)
-			if (vertex_lock[sparse_remap ? sparse_remap[i] : i])
+		{
+			unsigned int ri = sparse_remap ? sparse_remap[i] : unsigned(i);
+			assert(vertex_lock[ri] <= 1); // values other than 0/1 are reserved for future use
+
+			if (vertex_lock[ri])
 				result[remap[i]] = Kind_Locked;
+		}
 
 		for (size_t i = 0; i < vertex_count; ++i)
 			if (result[remap[i]] == Kind_Locked)
diff --git a/3rdparty/meshoptimizer/src/vertexcodec.cpp b/3rdparty/meshoptimizer/src/vertexcodec.cpp
index b0ea61f5b..53cf9d753 100644
--- a/3rdparty/meshoptimizer/src/vertexcodec.cpp
+++ b/3rdparty/meshoptimizer/src/vertexcodec.cpp
@@ -123,6 +123,7 @@ namespace meshopt
 const unsigned char kVertexHeader = 0xa0;
 
 static int gEncodeVertexVersion = 0;
+const int kDecodeVertexVersion = 1;
 
 const size_t kVertexBlockSizeBytes = 8192;
 const size_t kVertexBlockMaxSize = 256;
@@ -1803,11 +1804,28 @@ size_t meshopt_encodeVertexBufferBound(size_t vertex_count, size_t vertex_size)
 
 void meshopt_encodeVertexVersion(int version)
 {
-	assert(unsigned(version) <= 1);
+	assert(unsigned(version) <= unsigned(meshopt::kDecodeVertexVersion));
 
 	meshopt::gEncodeVertexVersion = version;
 }
 
+int meshopt_decodeVertexVersion(const unsigned char* buffer, size_t buffer_size)
+{
+	if (buffer_size < 1)
+		return -1;
+
+	unsigned char header = buffer[0];
+
+	if ((header & 0xf0) != meshopt::kVertexHeader)
+		return -1;
+
+	int version = header & 0x0f;
+	if (version > meshopt::kDecodeVertexVersion)
+		return -1;
+
+	return version;
+}
+
 int meshopt_decodeVertexBuffer(void* destination, size_t vertex_count, size_t vertex_size, const unsigned char* buffer, size_t buffer_size)
 {
 	using namespace meshopt;
@@ -1844,7 +1862,7 @@ int meshopt_decodeVertexBuffer(void* destination, size_t vertex_count, size_t ve
 		return -1;
 
 	int version = data_header & 0x0f;
-	if (version > 1)
+	if (version > kDecodeVertexVersion)
 		return -1;
 
 	size_t tail_size = vertex_size + (version == 0 ? 0 : vertex_size / 4);