perf(scene): draw preview/grip overlay from a separate GPU buffer

Command-preview / interim / grip-drag wires were appended into the main
wire buffer (all_wires), so every drag frame re-packed and re-uploaded the
entire batched base set — O(N) per move even though only the small overlay
changed.

Split them out: build_primitive keeps `wires` as the cached base only and
passes the overlay as a separate `preview_wires` Arc. The base keeps the
stable tile content id (no per-frame nonce), so its buffer stays resident
and is not re-uploaded during a drag. The overlay uploads to its own small
gpu_preview_wires buffer each frame and draws on top in the wire pass
(same pipeline, same order/depth). The dragged entity still renders
highlighted via its selected-path tessellation.

With the no-re-tess changes, a grip / command drag now costs a single
small overlay upload per move and nothing else.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

ROADMAP.md

@@ -266,6 +266,14 @@ the batched hatch pipeline (`hatch_batched_gpu.rs`) to wires:
 At 100 k wires that collapses thousands of draw calls into one. If iced
 0.14's widget-pipeline limits allow, immediate win.
 
+**Follow-up (landed): separate overlay buffer.** Command-preview / interim /
+grip-drag wires were appended into the main wire buffer, so every drag frame
+re-uploaded the entire (batched) base set. They now ride in their own small
+per-frame `gpu_preview_wires` buffer drawn on top in the wire pass, so the
+resident base buffer stays untouched during a drag — only the tiny overlay
+re-uploads. Combined with the no-re-tess work, a grip/command drag now costs
+one small overlay upload per move, nothing else.
+
 ### 3.4 Hardware instancing for repeated block inserts
 
 When the same block defn is `INSERT`-ed N times (every door / window in

src/scene/pipeline/mod.rs

@@ -81,6 +81,10 @@ pub struct Pipeline {
     wire_pixel_scissors: Vec<Option<[u32; 4]>>,
     /// Ghost copies (25% alpha) of selected wires for the X-ray depth pass.
     gpu_selected_wires: Vec<WireGpu>,
+    /// Command-preview / interim / grip-drag overlay wires. Re-uploaded every
+    /// frame they are present (small), drawn on top of the base wire pass — so
+    /// a live drag never re-uploads the resident base buffer.
+    gpu_preview_wires: Vec<WireGpu>,
     /// Phase 4-B — single batched-hatch GPU resource. Drawn in one
     /// indexed call with per-instance visibility masking the rest.
     /// Legacy per-hatch `Vec<HatchGpu>` + scissor / skip-flag plumbing
@@ -882,6 +886,7 @@ impl Pipeline {
             gpu_wires: vec![],
             wire_pixel_scissors: vec![],
             gpu_selected_wires: vec![],
+            gpu_preview_wires: vec![],
             gpu_hatch_batched: None,
             gpu_wipeouts: vec![],
             wipeout_skip_flags: vec![],
@@ -974,6 +979,22 @@ impl Pipeline {
         self.gpu_selected_wires = WireGpu::from_run(device, &selected, depth_map, None, false);
     }
 
+    /// Upload the live overlay (command preview / interim / grip-drag) wires.
+    /// Small and refreshed each frame they're present; kept separate from the
+    /// base wire buffer so a drag never re-uploads the resident base set.
+    pub fn upload_preview_wires(
+        &mut self,
+        device: &wgpu::Device,
+        wires: &[WireModel],
+        depth_map: &rustc_hash::FxHashMap<u64, f32>,
+    ) {
+        self.gpu_preview_wires = if wires.is_empty() {
+            vec![]
+        } else {
+            WireGpu::from_run(device, wires, depth_map, None, false)
+        };
+    }
+
     /// Recompute pixel scissor rects for viewport-clipped wires from the current view_proj.
     /// Called every frame from prepare() because scissor pixels shift with pan/zoom.
     pub fn compute_wire_scissors(&mut self, view_proj: glam::Mat4, clip_w: u32, clip_h: u32) {
@@ -1559,6 +1580,14 @@ impl Pipeline {
             if scissor_active {
                 pass.set_scissor_rect(0, 0, vp.width, vp.height);
             }
+            // Live overlay wires (command preview / interim / grip drag) on top
+            // of the base pass, same pipeline + depth test. No scissor.
+            for pw in &self.gpu_preview_wires {
+                if pw.instance_count > 0 {
+                    pass.set_vertex_buffer(0, pw.instance_buffer.slice(..));
+                    pass.draw(0..6, 0..pw.instance_count);
+                }
+            }
         }
 
         // ── Pass 6: wipeout fills (drawn after wires to mask them) ────────

src/scene/render.rs

@@ -32,6 +32,10 @@ pub struct CameraState {
 #[derive(Debug)]
 pub struct ViewportData {
     pub(super) wires: Arc<Vec<WireModel>>,
+    /// Live command-preview / interim / grip-drag overlay wires. Kept out of
+    /// the main `wires` buffer so a drag re-uploads only this small set each
+    /// frame, never the resident base buffer. Drawn on top in the wire pass.
+    pub(super) preview_wires: Arc<Vec<WireModel>>,
     /// 3DFACE entity wires — separated so they are uploaded to the dedicated
     /// face3d pipeline (fill + batched edges) instead of N individual WireGpu.
     pub(super) face3d_wires: Arc<Vec<WireModel>>,
@@ -272,6 +276,10 @@ impl shader::Primitive for Primitive {
                 );
                 inner.cached_selection = sel_key;
             }
+            // Live overlay (command preview / interim / grip drag) — small and
+            // refreshed every frame it's present, so a drag never re-uploads
+            // the resident base wire buffer.
+            inner.upload_preview_wires(device, &vp.preview_wires[..], &vp.draw_depths);
             let vproj = vp.uniforms.view_proj;
             inner.compute_wire_scissors(vproj, clip_size.width, clip_size.height);
             inner.compute_wipeout_scissors(vproj, clip_size.width, clip_size.height);
@@ -642,13 +650,14 @@ impl Scene {
         } else {
             (self.model_wires_for_viewport_arc(inst.handle, full.height), None)
         };
-        // Wire-buffer content id for the upload gate. Live preview / interim
+        // Wire-buffer content id for the upload gate. Preview / interim wires
-        // wires are appended below and vary frame-to-frame, so any frame
+        // are NOT part of this buffer anymore (they go in a separate per-frame
-        // carrying them forces a fresh upload, as do the non-tile paths.
+        // overlay buffer below), so the base id is the stable tile content gen
-        let has_overlay = self.interim_wire.is_some() || !self.preview_wires.is_empty();
+        // — a drag no longer re-uploads the whole base wire set every move.
+        // Non-tile paths have no stable id and force a re-upload.
         let wire_content_id = match tile_wire_gen {
-            Some(g) if !has_overlay => g,
+            Some(g) => g,
-            _ => {
+            None => {
                 let n = self.wire_force_nonce.get().wrapping_add(1);
                 self.wire_force_nonce.set(n);
                 n | (1u64 << 63)
@@ -676,12 +685,15 @@ impl Scene {
                 (Arc::new(f), Arc::new(o))
             }
         };
-        // Reuse the cached `other` Arc directly when no live preview/interim
+        // Base wire set — the cached `other` Arc directly, never cloned to
-        // wires need appending; otherwise clone once to extend.
+        // append overlays. Preview / interim wires ride in their own small
-        let all_wires = if !has_overlay {
+        // per-frame buffer so the (potentially huge) base buffer stays resident
-            other_arc
+        // and unchanged while a command preview or grip drag is live.
+        let all_wires = other_arc;
+        let preview_wires = if self.interim_wire.is_none() && self.preview_wires.is_empty() {
+            Arc::new(Vec::new())
         } else {
-            let mut v = (*other_arc).clone();
+            let mut v: Vec<WireModel> = Vec::with_capacity(self.preview_wires.len() + 1);
             if let Some(iw) = &self.interim_wire {
                 v.push(iw.clone());
             }
@@ -749,6 +761,7 @@ impl Scene {
 
         Some(ViewportData {
             wires: all_wires,
+            preview_wires,
             face3d_wires,
             draw_depths: self.draw_depth_map(),
             hatches,