perf(render): hoist per-wire constants to a storage buffer (native)

Every wire segment is one instance, and each carried the wire's color,
line-weight, dash pattern and draw-depth (~44 B) inline — re-fetched per
segment even though they're constant along the wire. On native, move
those into a per-wire `WireConst` storage buffer (group 1) indexed by a
`wire_id` on the instance. The instance drops from 104 B to 60 B (~42 %)
across every entity's wires, and the shared constants are fetched once
per wire (cached) instead of once per segment — less vertex-attribute
bandwidth and less GPU upload on rebuilds.

WebGL2 has no vertex-stage storage buffers, so the wasm build keeps the
original self-contained fat instance and wire.wgsl; native uses the new
slim instance, wire_indexed.wgsl, and a group-1 bind group built per
batch in WireGpu::from_run / from_batch. Visual output is unchanged
(colors are now full f32 rather than 8-bit, if anything more precise).

Co-Authored-By: Claude Opus 4.8 <[email protected]>

src/scene/pipeline/mod.rs

@@ -32,6 +32,11 @@ pub struct Pipeline {
     /// Same shader as wire_pipeline but depth_compare=Greater, depth_write_enabled=false.
     /// Used to draw ghost copies of selected wires through occluding geometry.
     wire_xray_pipeline: wgpu::RenderPipeline,
+    /// Layout for the per-wire `WireConst` storage buffer (group 1 of the wire /
+    /// xray pipelines). `Some` on native; `None` on web (fat instance, no
+    /// storage). Passed to `WireGpu::from_run` / `from_batch` so they can build
+    /// each batch's per-wire bind group.
+    wire_const_bgl: Option<wgpu::BindGroupLayout>,
     hatch_pipeline: wgpu::RenderPipeline,
     /// Phase 4-B — single-draw batched hatch pipeline. Per-instance
     /// data lives in storage buffers; one draw call covers every
@@ -181,9 +186,19 @@ impl Pipeline {
         });
 
         // ── Wire pipeline ──────────────────────────────────────────────────
+        // Native: per-wire constants live in a storage buffer (group 1), so the
+        // slim instance only carries per-segment data. WebGL2 has no
+        // vertex-stage storage buffers, so the wasm build keeps the fat instance
+        // and a single bind group.
+        #[cfg(not(target_arch = "wasm32"))]
+        let wire_const_bgl = wire_gpu::WireConst::bind_group_layout(device);
+        #[cfg(not(target_arch = "wasm32"))]
+        let wire_bgls: &[&wgpu::BindGroupLayout] = &[&frame_bgl, &wire_const_bgl];
+        #[cfg(target_arch = "wasm32")]
+        let wire_bgls: &[&wgpu::BindGroupLayout] = &[&frame_bgl];
         let wire_layout = device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
             label: Some("wire.pipeline_layout"),
-            bind_group_layouts: &[&frame_bgl],
+            bind_group_layouts: wire_bgls,
             push_constant_ranges: &[],
         });
 
@@ -192,6 +207,11 @@ impl Pipeline {
 
         let wire_shader = device.create_shader_module(wgpu::ShaderModuleDescriptor {
             label: Some("wire.shader"),
+            #[cfg(not(target_arch = "wasm32"))]
+            source: wgpu::ShaderSource::Wgsl(std::borrow::Cow::Borrowed(include_str!(
+                "../../shaders/wire_indexed.wgsl"
+            ))),
+            #[cfg(target_arch = "wasm32")]
             source: wgpu::ShaderSource::Wgsl(std::borrow::Cow::Borrowed(include_str!(
                 "../../shaders/wire.wgsl"
             ))),
@@ -909,6 +929,10 @@ impl Pipeline {
         Self {
             wire_pipeline,
             wire_xray_pipeline,
+            #[cfg(not(target_arch = "wasm32"))]
+            wire_const_bgl: Some(wire_const_bgl),
+            #[cfg(target_arch = "wasm32")]
+            wire_const_bgl: None,
             hatch_pipeline,
             hatch_batched_pipeline,
             image_pipeline,
@@ -983,7 +1007,7 @@ impl Pipeline {
             {
                 j += 1;
             }
-            batches.extend(WireGpu::from_run(device, &wires[i..j], depth_map, scissor, mesh_edge));
+            batches.extend(WireGpu::from_run(device, &wires[i..j], depth_map, scissor, mesh_edge, self.wire_const_bgl.as_ref()));
             i = j;
         }
         self.gpu_wires = batches;
@@ -1044,7 +1068,8 @@ impl Pipeline {
         if let Some(h) = hover {
             push(h.value(), WireModel::HOVER, wires, &mut out);
         }
-        self.gpu_selected_wires = WireGpu::from_run(device, &out, depth_map, None, false);
+        self.gpu_selected_wires =
+            WireGpu::from_run(device, &out, depth_map, None, false, self.wire_const_bgl.as_ref());
     }
 
     /// Upload the live overlay (command preview / interim / grip-drag) wires.
@@ -1059,7 +1084,7 @@ impl Pipeline {
         self.gpu_preview_wires = if wires.is_empty() {
             vec![]
         } else {
-            WireGpu::from_run(device, wires, depth_map, None, false)
+            WireGpu::from_run(device, wires, depth_map, None, false, self.wire_const_bgl.as_ref())
         };
     }
 
@@ -1141,7 +1166,8 @@ impl Pipeline {
     ) {
         // Edge buffer is always built from `face3d_wires`, so 3DFACE
         // outlines stay on the screen regardless of mode.
-        self.gpu_face3d_edges = WireGpu::from_batch(device, face3d_wires, depth_map);
+        self.gpu_face3d_edges =
+            WireGpu::from_batch(device, face3d_wires, depth_map, self.wire_const_bgl.as_ref());
         // Fill buffer split: 3D quads + PolyfaceMesh / PolygonMesh face
         // tris go to `vertex_buffer_3d` (gated by `keep_3d_mesh_fills`);
         // 2D fills (text-LOD greek, MultiLeader background) go to
@@ -1612,6 +1638,9 @@ impl Pipeline {
             pass.set_bind_group(0, &self.uniform_bind_group, &[]);
             for edges in &self.gpu_face3d_edges {
                 if edges.instance_count > 0 {
+                    if let Some(bg) = &edges.const_bind_group {
+                        pass.set_bind_group(1, bg.as_ref(), &[]);
+                    }
                     pass.set_vertex_buffer(0, edges.instance_buffer.slice(..));
                     pass.draw(0..6, 0..edges.instance_count);
                 }
@@ -1670,6 +1699,9 @@ impl Pipeline {
                     }
                     _ => {}
                 }
+                if let Some(bg) = &wire.const_bind_group {
+                    pass.set_bind_group(1, bg.as_ref(), &[]);
+                }
                 pass.set_vertex_buffer(0, wire.instance_buffer.slice(..));
                 pass.draw(0..6, 0..wire.instance_count);
             }
@@ -1685,6 +1717,9 @@ impl Pipeline {
                 pass.set_pipeline(&self.wire_xray_pipeline);
                 for pw in &self.gpu_preview_wires {
                     if pw.instance_count > 0 {
+                        if let Some(bg) = &pw.const_bind_group {
+                            pass.set_bind_group(1, bg.as_ref(), &[]);
+                        }
                         pass.set_vertex_buffer(0, pw.instance_buffer.slice(..));
                         pass.draw(0..6, 0..pw.instance_count);
                     }
@@ -1775,6 +1810,9 @@ impl Pipeline {
             pass.set_bind_group(0, &self.uniform_bind_group, &[]);
             for wire in &self.gpu_selected_wires {
                 if wire.instance_count > 0 {
+                    if let Some(bg) = &wire.const_bind_group {
+                        pass.set_bind_group(1, bg.as_ref(), &[]);
+                    }
                     pass.set_vertex_buffer(0, wire.instance_buffer.slice(..));
                     pass.draw(0..6, 0..wire.instance_count);
                 }

src/scene/pipeline/wire_gpu.rs

@@ -60,6 +60,90 @@ fn instance_buffer_mapped(
 
 // ── Instance layout ───────────────────────────────────────────────────────
 
+// ── Native: slim per-segment instance + shared per-wire constants ───────────
+//
+// Every segment of a wire used to carry the wire's color / line-weight / dash
+// pattern / draw-depth (~44 B) on each instance — re-fetched once per segment
+// even though it's constant along the wire. On native we hoist those into a
+// per-wire `WireConst` storage buffer indexed by `wire_id`, so the instance
+// keeps only the per-segment data (endpoints + arc-length distances). Cuts the
+// instance from 104 B to 60 B (~42 %) and removes the redundant per-segment
+// re-fetch of the shared constants. WebGL2 has no vertex-stage storage buffers,
+// so the wasm build below keeps the original self-contained fat instance.
+#[cfg(not(target_arch = "wasm32"))]
+#[repr(C)]
+#[derive(Copy, Clone, bytemuck::Pod, bytemuck::Zeroable)]
+pub struct WireInstance {
+    pub pos_a: [f32; 3],
+    pub pos_a_low: [f32; 3],
+    pub pos_b: [f32; 3],
+    pub pos_b_low: [f32; 3],
+    pub distance_a: f32,
+    pub distance_b: f32,
+    /// Index into the per-wire `WireConst` storage buffer (group 1).
+    pub wire_id: u32,
+}
+
+#[cfg(not(target_arch = "wasm32"))]
+impl WireInstance {
+    pub fn layout<'a>() -> wgpu::VertexBufferLayout<'a> {
+        // Must match `InstanceIn` in wire_indexed.wgsl.
+        const ATTRS: &[wgpu::VertexAttribute] = &[
+            wgpu::VertexAttribute { offset: std::mem::offset_of!(WireInstance, pos_a) as u64,      shader_location: 0, format: wgpu::VertexFormat::Float32x3 },
+            wgpu::VertexAttribute { offset: std::mem::offset_of!(WireInstance, pos_b) as u64,      shader_location: 1, format: wgpu::VertexFormat::Float32x3 },
+            wgpu::VertexAttribute { offset: std::mem::offset_of!(WireInstance, pos_a_low) as u64,  shader_location: 2, format: wgpu::VertexFormat::Float32x3 },
+            wgpu::VertexAttribute { offset: std::mem::offset_of!(WireInstance, pos_b_low) as u64,  shader_location: 3, format: wgpu::VertexFormat::Float32x3 },
+            wgpu::VertexAttribute { offset: std::mem::offset_of!(WireInstance, distance_a) as u64, shader_location: 4, format: wgpu::VertexFormat::Float32   },
+            wgpu::VertexAttribute { offset: std::mem::offset_of!(WireInstance, distance_b) as u64, shader_location: 5, format: wgpu::VertexFormat::Float32   },
+            wgpu::VertexAttribute { offset: std::mem::offset_of!(WireInstance, wire_id) as u64,    shader_location: 6, format: wgpu::VertexFormat::Uint32    },
+        ];
+        wgpu::VertexBufferLayout {
+            array_stride: std::mem::size_of::<WireInstance>() as u64,
+            step_mode: wgpu::VertexStepMode::Instance,
+            attributes: ATTRS,
+        }
+    }
+}
+
+/// Per-wire constants shared by every segment of a wire (native only). std430
+/// layout: three vec4 then four scalars = 64 B, matching `WireConst` in
+/// wire_indexed.wgsl.
+#[cfg(not(target_arch = "wasm32"))]
+#[repr(C)]
+#[derive(Copy, Clone, bytemuck::Pod, bytemuck::Zeroable)]
+pub struct WireConst {
+    pub color: [f32; 4],
+    pub pat0: [f32; 4],
+    pub pat1: [f32; 4],
+    pub half_width: f32,
+    pub pattern_length: f32,
+    pub draw_depth: f32,
+    pub _pad: f32,
+}
+
+#[cfg(not(target_arch = "wasm32"))]
+impl WireConst {
+    /// Bind-group layout for the per-wire storage buffer (group 1 of the wire /
+    /// xray pipelines). Read-only storage, visible to the vertex stage.
+    pub fn bind_group_layout(device: &wgpu::Device) -> wgpu::BindGroupLayout {
+        device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
+            label: Some("wire_const.bgl"),
+            entries: &[wgpu::BindGroupLayoutEntry {
+                binding: 0,
+                visibility: wgpu::ShaderStages::VERTEX,
+                ty: wgpu::BindingType::Buffer {
+                    ty: wgpu::BufferBindingType::Storage { read_only: true },
+                    has_dynamic_offset: false,
+                    min_binding_size: None,
+                },
+                count: None,
+            }],
+        })
+    }
+}
+
+// ── Web (WebGL2): self-contained fat instance (no vertex-stage storage) ─────
+#[cfg(target_arch = "wasm32")]
 #[repr(C)]
 #[derive(Copy, Clone, bytemuck::Pod, bytemuck::Zeroable)]
 pub struct WireInstance {
@@ -82,6 +166,7 @@ pub struct WireInstance {
     pub draw_depth: f32,
 }
 
+#[cfg(target_arch = "wasm32")]
 impl WireInstance {
     pub fn layout<'a>() -> wgpu::VertexBufferLayout<'a> {
         // Offsets come from the struct layout (must match the shader location
@@ -123,12 +208,17 @@ pub struct WireGpu {
     /// in shaded modes so the surface reads as a clean solid; pure
     /// wireframe / HiddenLine / *WithEdges modes draw them.
     pub is_3d_mesh_edge: bool,
+    /// Per-wire constants storage (group 1), shared across all chunks of one
+    /// build. `None` on web (the fat instance carries the constants inline) and
+    /// for empty buffers. The draw loop binds it as group 1 when present.
+    pub const_bind_group: Option<std::sync::Arc<wgpu::BindGroup>>,
 }
 
 /// Expand one `WireModel` into its per-segment instance stream (1 instance per
 /// finite segment). Pulled out so both the single-wire and batched paths share
 /// the same emission logic, and so the batched path can `par_iter` across
 /// wires on cold open.
+#[cfg(target_arch = "wasm32")]
 fn pack_color(color: [f32; 4]) -> [u8; 4] {
     [
         (color[0].clamp(0.0, 1.0) * 255.0 + 0.5) as u8,
@@ -138,28 +228,10 @@ fn pack_color(color: [f32; 4]) -> [u8; 4] {
     ]
 }
 
-fn emit_wire_instances(wire: &WireModel, color: [f32; 4], draw_depth: f32) -> Vec<WireInstance> {
-    let color_u8 = pack_color(color);
-    let pat0 = [
-        wire.pattern[0],
-        wire.pattern[1],
-        wire.pattern[2],
-        wire.pattern[3],
-    ];
-    let pat1 = [
-        wire.pattern[4],
-        wire.pattern[5],
-        wire.pattern[6],
-        wire.pattern[7],
-    ];
-    let half_width = wire.line_weight_px * 0.5;
-
+/// Cumulative arc-length per point (NaN-break aware), with the dash pattern
+/// centred on the wire. Shared by the wasm and native emission paths.
+fn wire_distances(wire: &WireModel) -> Vec<f32> {
     let n = wire.points.len();
-    let seg_count = n.saturating_sub(1);
-    if seg_count == 0 {
-        return Vec::new();
-    }
-
     let mut dists = vec![0.0_f32; n];
     let mut has_break = false;
     for i in 1..n {
@@ -206,22 +278,33 @@ fn emit_wire_instances(wire: &WireModel, color: [f32; 4], draw_depth: f32) -> Ve
             }
         }
     }
+    dists
+}
 
-    // Low residual paired index-for-index with points; empty → all-zero
-    // (interactive / preview wires that don't need sub-f32 precision).
+#[inline]
+fn finite3(p: [f32; 3]) -> bool {
+    p[0].is_finite() && p[1].is_finite() && p[2].is_finite()
+}
+
+/// Web: emit fat per-segment instances (each carries the wire's constants).
+#[cfg(target_arch = "wasm32")]
+fn emit_wire_instances(wire: &WireModel, color: [f32; 4], draw_depth: f32) -> Vec<WireInstance> {
+    let color_u8 = pack_color(color);
+    let pat0 = [wire.pattern[0], wire.pattern[1], wire.pattern[2], wire.pattern[3]];
+    let pat1 = [wire.pattern[4], wire.pattern[5], wire.pattern[6], wire.pattern[7]];
+    let half_width = wire.line_weight_px * 0.5;
+    let n = wire.points.len();
+    let seg_count = n.saturating_sub(1);
+    if seg_count == 0 {
+        return Vec::new();
+    }
+    let dists = wire_distances(wire);
     let low = |i: usize| -> [f32; 3] { wire.points_low.get(i).copied().unwrap_or([0.0; 3]) };
-
     let mut instances: Vec<WireInstance> = Vec::with_capacity(seg_count);
     for i in 0..seg_count {
         let a = wire.points[i];
         let b = wire.points[i + 1];
-        if !a[0].is_finite()
-            || !a[1].is_finite()
-            || !a[2].is_finite()
-            || !b[0].is_finite()
-            || !b[1].is_finite()
-            || !b[2].is_finite()
-        {
+        if !finite3(a) || !finite3(b) {
             continue;
         }
         instances.push(WireInstance {
@@ -242,6 +325,51 @@ fn emit_wire_instances(wire: &WireModel, color: [f32; 4], draw_depth: f32) -> Ve
     instances
 }
 
+/// Native: emit slim per-segment instances (positions + distances + `wire_id`)
+/// plus the one `WireConst` record every segment of this wire shares.
+#[cfg(not(target_arch = "wasm32"))]
+fn emit_wire_native(
+    wire: &WireModel,
+    wire_id: u32,
+    color: [f32; 4],
+    draw_depth: f32,
+) -> (Vec<WireInstance>, WireConst) {
+    let cst = WireConst {
+        color,
+        pat0: [wire.pattern[0], wire.pattern[1], wire.pattern[2], wire.pattern[3]],
+        pat1: [wire.pattern[4], wire.pattern[5], wire.pattern[6], wire.pattern[7]],
+        half_width: wire.line_weight_px * 0.5,
+        pattern_length: wire.pattern_length,
+        draw_depth,
+        _pad: 0.0,
+    };
+    let n = wire.points.len();
+    let seg_count = n.saturating_sub(1);
+    if seg_count == 0 {
+        return (Vec::new(), cst);
+    }
+    let dists = wire_distances(wire);
+    let low = |i: usize| -> [f32; 3] { wire.points_low.get(i).copied().unwrap_or([0.0; 3]) };
+    let mut instances: Vec<WireInstance> = Vec::with_capacity(seg_count);
+    for i in 0..seg_count {
+        let a = wire.points[i];
+        let b = wire.points[i + 1];
+        if !finite3(a) || !finite3(b) {
+            continue;
+        }
+        instances.push(WireInstance {
+            pos_a: a,
+            pos_a_low: low(i),
+            pos_b: b,
+            pos_b_low: low(i + 1),
+            distance_a: dists[i],
+            distance_b: dists[i + 1],
+            wire_id,
+        });
+    }
+    (instances, cst)
+}
+
 /// Looks up a wire's draw-order depth from the per-entity map using the
 /// handle encoded in its `name`. Falls back to 0.0 (transient / preview
 /// wires that carry no document handle).
@@ -254,6 +382,36 @@ fn wire_draw_depth(wire: &WireModel, depth_map: &rustc_hash::FxHashMap<u64, f32>
         .unwrap_or(0.0)
 }
 
+/// Build the shared per-wire `WireConst` storage buffer and its bind group
+/// (native only). All instance-buffer chunks from one build reference the same
+/// buffer via their global `wire_id`, so a single bind group is cloned into
+/// each chunk.
+#[cfg(not(target_arch = "wasm32"))]
+fn build_const_bind_group(
+    device: &wgpu::Device,
+    bgl: &wgpu::BindGroupLayout,
+    consts: &[WireConst],
+) -> std::sync::Arc<wgpu::BindGroup> {
+    use wgpu::util::DeviceExt;
+    // wgpu rejects zero-sized buffers; pad with one zeroed record when empty.
+    let one = [<WireConst as bytemuck::Zeroable>::zeroed()];
+    let data: &[WireConst] = if consts.is_empty() { &one } else { consts };
+    let buf = device.create_buffer_init(&wgpu::util::BufferInitDescriptor {
+        label: Some("wire_const.buf"),
+        contents: bytemuck::cast_slice(data),
+        usage: wgpu::BufferUsages::STORAGE,
+    });
+    let bg = device.create_bind_group(&wgpu::BindGroupDescriptor {
+        label: Some("wire_const.bg"),
+        layout: bgl,
+        entries: &[wgpu::BindGroupEntry {
+            binding: 0,
+            resource: buf.as_entire_binding(),
+        }],
+    });
+    std::sync::Arc::new(bg)
+}
+
 impl WireGpu {
 
     /// Merge a run of WireModels that share scissor + mesh-edge state into one
@@ -272,32 +430,76 @@ impl WireGpu {
         depth_map: &rustc_hash::FxHashMap<u64, f32>,
         scissor: Option<[f32; 4]>,
         mesh_edge: bool,
+        const_bgl: Option<&wgpu::BindGroupLayout>,
     ) -> Vec<Self> {
-        use crate::par::prelude::*;
         const MAX_INSTANCES: usize = 268_435_456 / std::mem::size_of::<WireInstance>();
-        let per: Vec<Vec<WireInstance>> = wires
-            .par_iter()
-            .map(|w| emit_wire_instances(w, w.color, wire_draw_depth(w, depth_map)))
-            .collect();
-        let mut instances: Vec<WireInstance> = Vec::with_capacity(per.iter().map(Vec::len).sum());
-        for mut v in per {
-            instances.append(&mut v);
+        #[cfg(not(target_arch = "wasm32"))]
+        {
+            use crate::par::prelude::*;
+            // Global `wire_id` = wire index; one shared WireConst buffer for all
+            // chunks. Indexed `collect` preserves wire order (the pass relies on
+            // submission order for depth-biased / transparent overlap).
+            let per: Vec<(Vec<WireInstance>, WireConst)> = wires
+                .par_iter()
+                .enumerate()
+                .map(|(idx, w)| {
+                    emit_wire_native(w, idx as u32, w.color, wire_draw_depth(w, depth_map))
+                })
+                .collect();
+            let mut instances: Vec<WireInstance> =
+                Vec::with_capacity(per.iter().map(|(v, _)| v.len()).sum());
+            let mut consts: Vec<WireConst> = Vec::with_capacity(per.len());
+            for (mut v, c) in per {
+                instances.append(&mut v);
+                consts.push(c);
+            }
+            if instances.is_empty() {
+                return vec![];
+            }
+            let bg = const_bgl.map(|bgl| build_const_bind_group(device, bgl, &consts));
+            instances
+                .chunks(MAX_INSTANCES)
+                .map(|chunk| {
+                    let buf = instance_buffer_mapped(device, "wire.run.ibuf", chunk);
+                    Self {
+                        instance_buffer: buf,
+                        instance_count: chunk.len() as u32,
+                        vp_scissor: scissor,
+                        is_3d_mesh_edge: mesh_edge,
+                        const_bind_group: bg.clone(),
+                    }
+                })
+                .collect()
         }
-        if instances.is_empty() {
-            return vec![];
+        #[cfg(target_arch = "wasm32")]
+        {
+            let _ = const_bgl;
+            let per: Vec<Vec<WireInstance>> = wires
+                .iter()
+                .map(|w| emit_wire_instances(w, w.color, wire_draw_depth(w, depth_map)))
+                .collect();
+            let mut instances: Vec<WireInstance> =
+                Vec::with_capacity(per.iter().map(Vec::len).sum());
+            for mut v in per {
+                instances.append(&mut v);
+            }
+            if instances.is_empty() {
+                return vec![];
+            }
+            instances
+                .chunks(MAX_INSTANCES)
+                .map(|chunk| {
+                    let buf = instance_buffer_mapped(device, "wire.run.ibuf", chunk);
+                    Self {
+                        instance_buffer: buf,
+                        instance_count: chunk.len() as u32,
+                        vp_scissor: scissor,
+                        is_3d_mesh_edge: mesh_edge,
+                        const_bind_group: None,
+                    }
+                })
+                .collect()
         }
-        instances
-            .chunks(MAX_INSTANCES)
-            .map(|chunk| {
-                let buf = instance_buffer_mapped(device, "wire.run.ibuf", chunk);
-                Self {
-                    instance_buffer: buf,
-                    instance_count: chunk.len() as u32,
-                    vp_scissor: scissor,
-                    is_3d_mesh_edge: mesh_edge,
-                }
-            })
-            .collect()
     }
 
     /// Merge multiple WireModels into GPU instance buffers, chunked to fit the
@@ -307,58 +509,80 @@ impl WireGpu {
         device: &wgpu::Device,
         wires: &[WireModel],
         depth_map: &rustc_hash::FxHashMap<u64, f32>,
+        const_bgl: Option<&wgpu::BindGroupLayout>,
     ) -> Vec<Self> {
         let total_segs: usize = wires.iter().map(|w| w.points.len().saturating_sub(1)).sum();
         if total_segs == 0 {
             return vec![];
         }
-
-        // Parallel per-wire instance emission. Each wire's stream is
-        // independent — `block_cache` groups wires by style upstream, so order
-        // within a batch does not affect correctness. `reduce` concatenates the
-        // per-wire Vec<WireInstance> chunks while letting rayon spread work
-        // across cores.
-        #[cfg(not(target_arch = "wasm32"))]
-        let instances: Vec<WireInstance> = wires
-            .par_iter()
-            .map(|wire| emit_wire_instances(wire, wire.color, wire_draw_depth(wire, depth_map)))
-            .reduce(Vec::new, |mut acc, mut chunk| {
-                if acc.is_empty() {
-                    chunk
-                } else if chunk.is_empty() {
-                    acc
-                } else {
-                    acc.append(&mut chunk);
-                    acc
-                }
-            });
-        // Web: no thread pool — concatenate the per-wire streams sequentially.
-        #[cfg(target_arch = "wasm32")]
-        let instances: Vec<WireInstance> = wires
-            .iter()
-            .flat_map(|wire| emit_wire_instances(wire, wire.color, wire_draw_depth(wire, depth_map)))
-            .collect();
-
-        if instances.is_empty() {
-            return vec![];
-        }
-
         // GPU max buffer size is 256 MB; chunk to stay within the limit.
         const MAX_INSTANCES: usize = 268_435_456 / std::mem::size_of::<WireInstance>();
 
-        instances
-            .chunks(MAX_INSTANCES)
-            .enumerate()
-            .map(|(i, chunk)| {
-                let label = format!("wire.batch.ibuf.{i}");
-                let instance_buffer = instance_buffer_mapped(device, &label, chunk);
-                Self {
-                    instance_buffer,
-                    instance_count: chunk.len() as u32,
-                    vp_scissor: None,
-                    is_3d_mesh_edge: false,
-                }
-            })
-            .collect()
+        #[cfg(not(target_arch = "wasm32"))]
+        {
+            use crate::par::prelude::*;
+            // `block_cache` groups wires by style upstream; order within a batch
+            // doesn't affect correctness, but indexed `collect` gives each wire a
+            // stable `wire_id` into the shared WireConst buffer.
+            let per: Vec<(Vec<WireInstance>, WireConst)> = wires
+                .par_iter()
+                .enumerate()
+                .map(|(idx, w)| {
+                    emit_wire_native(w, idx as u32, w.color, wire_draw_depth(w, depth_map))
+                })
+                .collect();
+            let mut instances: Vec<WireInstance> =
+                Vec::with_capacity(per.iter().map(|(v, _)| v.len()).sum());
+            let mut consts: Vec<WireConst> = Vec::with_capacity(per.len());
+            for (mut v, c) in per {
+                instances.append(&mut v);
+                consts.push(c);
+            }
+            if instances.is_empty() {
+                return vec![];
+            }
+            let bg = const_bgl.map(|bgl| build_const_bind_group(device, bgl, &consts));
+            instances
+                .chunks(MAX_INSTANCES)
+                .enumerate()
+                .map(|(i, chunk)| {
+                    let label = format!("wire.batch.ibuf.{i}");
+                    let instance_buffer = instance_buffer_mapped(device, &label, chunk);
+                    Self {
+                        instance_buffer,
+                        instance_count: chunk.len() as u32,
+                        vp_scissor: None,
+                        is_3d_mesh_edge: false,
+                        const_bind_group: bg.clone(),
+                    }
+                })
+                .collect()
+        }
+        #[cfg(target_arch = "wasm32")]
+        {
+            let _ = const_bgl;
+            let instances: Vec<WireInstance> = wires
+                .iter()
+                .flat_map(|w| emit_wire_instances(w, w.color, wire_draw_depth(w, depth_map)))
+                .collect();
+            if instances.is_empty() {
+                return vec![];
+            }
+            instances
+                .chunks(MAX_INSTANCES)
+                .enumerate()
+                .map(|(i, chunk)| {
+                    let label = format!("wire.batch.ibuf.{i}");
+                    let instance_buffer = instance_buffer_mapped(device, &label, chunk);
+                    Self {
+                        instance_buffer,
+                        instance_count: chunk.len() as u32,
+                        vp_scissor: None,
+                        is_3d_mesh_edge: false,
+                        const_bind_group: None,
+                    }
+                })
+                .collect()
+        }
     }
 }

src/shaders/wire_indexed.wgsl

@@ -0,0 +1,151 @@
+// Wire shader (native) — same as wire.wgsl, but the per-wire constants
+// (color / half_width / dash pattern / draw_depth) live in a per-wire storage
+// buffer indexed by `wire_id` instead of being replicated on every segment
+// instance. Cuts the instance from 104 B to 60 B and removes the redundant
+// per-segment re-fetch of constants. WebGL2 has no vertex-stage storage
+// buffers, so the wasm build uses wire.wgsl (fat instance) instead.
+
+struct Uniforms {
+    viewport_size:    vec2<f32>,
+    world_per_pixel:  f32,
+    lwdisplay_enable: f32,
+    flat_shade: f32,
+    transparency_enable: f32,
+    _pad: vec2<f32>,
+    view_rot:         mat4x4<f32>,
+    eye_high:         vec3<f32>,
+    _pad_eh:          f32,
+    eye_low:          vec3<f32>,
+    _pad_el:          f32,
+}
+@group(0) @binding(0) var<uniform> u: Uniforms;
+
+// Per-wire constants (std430). Must match `WireConst` in wire_gpu.rs.
+struct WireConst {
+    color:          vec4<f32>,
+    pat0:           vec4<f32>,
+    pat1:           vec4<f32>,
+    half_width:     f32,
+    pattern_length: f32,
+    draw_depth:     f32,
+    _pad:           f32,
+}
+@group(1) @binding(0) var<storage, read> wire_consts: array<WireConst>;
+
+struct InstanceIn {
+    @location(0) pos_a:      vec3<f32>,
+    @location(1) pos_b:      vec3<f32>,
+    @location(2) pos_a_low:  vec3<f32>,
+    @location(3) pos_b_low:  vec3<f32>,
+    @location(4) distance_a: f32,
+    @location(5) distance_b: f32,
+    @location(6) wire_id:    u32,
+}
+
+const DRAW_ORDER_BIAS: f32 = 0.001;
+
+struct VertexOut {
+    @builtin(position)              clip_pos:       vec4<f32>,
+    @location(0)                    color:          vec4<f32>,
+    @location(1)                    distance:       f32,
+    @location(2)                    pattern_length: f32,
+    @location(3)                    pat0:           vec4<f32>,
+    @location(4)                    pat1:           vec4<f32>,
+    @location(5) @interpolate(flat) min_elem:       f32,
+}
+
+@vertex fn vs_main(@builtin(vertex_index) vid: u32, in: InstanceIn) -> VertexOut {
+    let c = wire_consts[in.wire_id];
+
+    let which_end_arr = array<f32, 6>(0.0, 1.0, 1.0, 0.0, 1.0, 0.0);
+    let side_arr      = array<f32, 6>(-1.0, -1.0, 1.0, -1.0, 1.0, 1.0);
+    let which_end = which_end_arr[vid];
+    let side      = side_arr[vid];
+
+    let rel_a = (in.pos_a - u.eye_high) + (in.pos_a_low - u.eye_low);
+    let rel_b = (in.pos_b - u.eye_high) + (in.pos_b_low - u.eye_low);
+    let clip_a = u.view_rot * vec4<f32>(rel_a, 1.0);
+    let clip_b = u.view_rot * vec4<f32>(rel_b, 1.0);
+
+    let ndc_a = clip_a.xy / clip_a.w;
+    let ndc_b = clip_b.xy / clip_b.w;
+
+    let screen_a = ndc_a * u.viewport_size * 0.5;
+    let screen_b = ndc_b * u.viewport_size * 0.5;
+
+    let seg = screen_b - screen_a;
+    let seg_len = length(seg);
+    var perp: vec2<f32>;
+    if seg_len > 1e-4 {
+        let dir = seg / seg_len;
+        perp = vec2<f32>(-dir.y, dir.x);
+    } else {
+        perp = vec2<f32>(0.0, 1.0);
+    }
+
+    let perp_ndc = perp / (u.viewport_size * 0.5);
+
+    let clip_pos = mix(clip_a, clip_b, which_end);
+
+    let hw = select(0.5, c.half_width, u.lwdisplay_enable > 0.5);
+
+    let ndc_offset = perp_ndc * hw * side;
+    let final_clip = clip_pos + vec4<f32>(ndc_offset * clip_pos.w, 0.0, 0.0);
+
+    var min_elem: f32 = c.pattern_length;
+    let elems = array<f32, 8>(
+        c.pat0.x, c.pat0.y, c.pat0.z, c.pat0.w,
+        c.pat1.x, c.pat1.y, c.pat1.z, c.pat1.w,
+    );
+    for (var i = 0u; i < 8u; i++) {
+        let e = abs(elems[i]);
+        if e > 0.0 && e < min_elem { min_elem = e; }
+    }
+
+    var out: VertexOut;
+    out.clip_pos       = final_clip;
+    out.clip_pos.z     = out.clip_pos.z - c.draw_depth * DRAW_ORDER_BIAS * out.clip_pos.w;
+    out.color          = c.color;
+    out.distance       = mix(in.distance_a, in.distance_b, which_end);
+    out.pattern_length = c.pattern_length;
+    out.pat0           = c.pat0;
+    out.pat1           = c.pat1;
+    out.min_elem       = min_elem;
+    return out;
+}
+
+fn in_dash(dist: f32, pat_len: f32, p0: vec4<f32>, p1: vec4<f32>) -> bool {
+    let d   = ((dist % pat_len) + pat_len) % pat_len;
+    var pos = 0.0f;
+    let dot_half = u.world_per_pixel * 0.75;
+    let elems = array<f32, 8>(p0.x, p0.y, p0.z, p0.w, p1.x, p1.y, p1.z, p1.w);
+    var count = 0u;
+    for (var i = 0u; i < 8u; i++) {
+        if elems[i] != 0.0 { count = i + 1u; }
+    }
+    for (var i = 0u; i < count; i++) {
+        let elem = elems[i];
+        if elem == 0.0 {
+            let dd = abs(d - pos);
+            if min(dd, pat_len - dd) <= dot_half { return true; }
+        } else if elem > 0.0 {
+            if d >= pos && d < pos + elem { return true; }
+            pos += elem;
+        } else {
+            pos += -elem;
+        }
+    }
+    return false;
+}
+
+@fragment fn fs_main(in: VertexOut) -> @location(0) vec4<f32> {
+    if in.pattern_length > 0.0 {
+        if in.min_elem >= u.world_per_pixel {
+            if !in_dash(in.distance, in.pattern_length, in.pat0, in.pat1) {
+                discard;
+            }
+        }
+    }
+    let alpha = select(1.0, in.color.a, u.transparency_enable > 0.5);
+    return vec4<f32>(in.color.rgb, alpha);
+}