init

@@ -104,0 +132,4 @@
+  });
+}
+
+static Array<int> reverse_indices_in_groups(const Span<int> group_indices,

@@ -177,0 +302,4 @@
+      const float total = elem_island.size();
+      BLI_assert(total > 0.0f);
+      const float scale = accumulate<float>(scale_varray, elem_island) / total;
+      const float3 center = accumulate<float3>(center_varray, elem_island) / total;

@@ -290,0 +371,4 @@
+
+  /* If result indices is for gathered array, map than back into global indices. */
+  if (face_mask.size() != mesh.faces_num) {
+    parallel_transform<int>(r_item_indices, 4098, [&](const int pos) { return face_mask[pos]; });

@@ -344,3 +393,1 @@
-  edge_selection.foreach_index([&](const int edge_index) {
-    const int2 &edge = edges[edge_index];
-    disjoint_set.join(edge[0], edge[1]);
+  edge_mask.foreach_index_optimized<int>(GrainSize(4098), [&](const int edge_i) {

@@ -449,0 +487,4 @@
+      const GroupedSpan<int> item_islands(item_offsets.as_span(), item_indices);
+      const GroupedSpan<int> vert_islands(vert_offsets.as_span(), vert_indices);
+
+      const VArray<float> &scale_varray = evaluator.get_evaluated<float>(0);

@@ -101,0 +211,4 @@
+          for (const int i : indices.slice(range)) {
+            value += values[i];
+          }
+          return join_accumulators({value / float(indices.size()), 1}, other);

@@ -207,2 +333,2 @@
-    for (const int island_index : range) {
-      const ElementIsland &island = islands[island_index];
+  AtomicDisjointSet disjoint_set(vert_mask.size());
+  const GroupedSpan<int> face_verts(mesh.face_offsets(), mesh.corner_verts());

@@ -394,1 +438,3 @@
-  scale_vertex_islands_on_axis(mesh, island, params, get_edge_verts);
+  /* If result indices is for gathered array, map than back into global indices. */
+  if (edge_mask.size() != mesh.edges_num) {
+    parallel_transform<int>(r_item_indices, 4096, [&](const int pos) { return edge_mask[pos]; });

@@ -112,0 +236,4 @@
+                            Mesh &mesh)
+{
+  MutableSpan<float3> positions = mesh.vert_positions_for_write();
+  threading::parallel_for(