feat!: replace CPU rendering with GPU via wgpu/WGSL

Drop tiny-skia and tempfile. All rendering now runs on the GPU using a
WGSL fragment shader with rounded-rect SDF, gaussian glow falloff, and
cursor avoidance. Add color temperature support with presets (warm, cool,
studio, daylight, golden) and custom Kelvin values. Suppress MangoHUD
via environment variable.

Cargo.toml

@@ -8,8 +8,11 @@ rust-version = "1.85"
 
 [dependencies]
 anyhow = "1.0"
+bytemuck = { version = "1", features = ["derive"] }
 nix = { version = "0.29", features = ["signal", "process", "poll"] }
-tempfile = "3"
-tiny-skia = "0.12.0"
+pollster = "0.4"
+raw-window-handle = "0.6"
+wayland-backend = { version = "0.3", features = ["client_system"] }
 wayland-client = "0.31"
 wayland-protocols-wlr = { version = "0.3", features = ["client"] }
+wgpu = "29"

src/main.rs

@@ -4,71 +4,120 @@ mod wayland;
 
 use std::time::{Duration, Instant};
 
-use anyhow::{Context, Result};
-use tiny_skia::Pixmap;
+use anyhow::Result;
 
 use lock::ToggleResult;
+use render::GpuRenderer;
 use wayland::{OverlayEvent, WaylandBackend};
 
-const DISPLAY_DURATION: Duration = Duration::from_secs(5);
+const DISPLAY_DURATION: Duration = Duration::from_secs(10);
+const FRAME_INTERVAL: Duration = Duration::from_millis(16);
 
 fn main() -> Result<()> {
+    // Suppress MangoHUD — it hooks into Vulkan and we don't want an overlay on our overlay.
+    std::env::set_var("DISABLE_MANGOHUD", "1");
+
     match lock::try_toggle()? {
         ToggleResult::SignaledExisting => return Ok(()),
         ToggleResult::RunNew => {}
     }
 
+    let color = parse_color_arg();
+
     let _lock = lock::create_lock()?;
 
     let mut wl = WaylandBackend::new()?;
     let (w, h) = wl.screen_size();
-    let mut pixmap = Pixmap::new(w, h).context("create pixmap")?;
-    let mut argb_buf = vec![0u8; (w as usize) * (h as usize) * 4];
+
+    let mut renderer = GpuRenderer::new(
+        wl.raw_display_handle()?,
+        wl.raw_window_handle()?,
+        w,
+        h,
+        color,
+    )?;
 
     let mut cursor_x = w as f64 / 2.0;
     let mut cursor_y = h as f64 / 2.0;
 
-    draw(&mut wl, &mut pixmap, &mut argb_buf, cursor_x, cursor_y)?;
+    renderer.render(cursor_x, cursor_y)?;
 
     let start = Instant::now();
+    let mut last_draw = Instant::now();
+    let mut dirty = false;
 
     loop {
         let elapsed = start.elapsed();
         if elapsed >= DISPLAY_DURATION || lock::should_exit() {
             break;
         }
+
+        if dirty && last_draw.elapsed() >= FRAME_INTERVAL {
+            renderer.render(cursor_x, cursor_y)?;
+            last_draw = Instant::now();
+            dirty = false;
+        }
+
         let remaining = DISPLAY_DURATION - elapsed;
-        let event = match wl.next_event(remaining)? {
+        let poll_timeout = if dirty {
+            FRAME_INTERVAL
+                .saturating_sub(last_draw.elapsed())
+                .min(remaining)
+        } else {
+            remaining
+        };
+
+        let event = match wl.next_event(poll_timeout)? {
             Some(ev) => ev,
-            None => continue, // poll timeout — loop back to check elapsed time
+            None => continue,
         };
         match event {
             OverlayEvent::Close => break,
             OverlayEvent::CursorMoved(x, y) => {
                 cursor_x = x;
                 cursor_y = y;
-                draw(&mut wl, &mut pixmap, &mut argb_buf, cursor_x, cursor_y)?;
-            }
-            OverlayEvent::Configure(new_w, new_h) => {
-                if new_w != w || new_h != h {
-                    break;
-                }
+                dirty = true;
             }
+            OverlayEvent::Configure(_, _) => {}
         }
     }
 
     Ok(())
 }
 
-fn draw(
-    wl: &mut WaylandBackend,
-    pixmap: &mut Pixmap,
-    argb_buf: &mut [u8],
-    cursor_x: f64,
-    cursor_y: f64,
-) -> Result<()> {
-    render::render_ring_light(pixmap, cursor_x, cursor_y);
-    render::pixmap_to_argb8888(pixmap, argb_buf);
-    let (w, h) = wl.screen_size();
-    wl.present(argb_buf, w, h)
+/// Parse color temperature from CLI args.
+///
+/// Usage:
+///   ringa              → pure white (default)
+///   ringa warm         → 3200K (warm tungsten)
+///   ringa cool         → 6500K (cool daylight)
+///   ringa studio       → 5000K (neutral studio)
+///   ringa daylight     → 5600K (daylight balanced)
+///   ringa golden       → 2700K (golden hour)
+///   ringa 4200         → custom Kelvin value
+fn parse_color_arg() -> [f32; 3] {
+    let arg = match std::env::args().nth(1) {
+        Some(a) => a,
+        None => return [1.0, 1.0, 1.0], // pure white
+    };
+
+    let kelvin = match arg.to_lowercase().as_str() {
+        "warm" => 3200,
+        "golden" => 2700,
+        "studio" => 5000,
+        "daylight" => 5600,
+        "cool" => 6500,
+        "white" => return [1.0, 1.0, 1.0],
+        other => match other.parse::<u32>() {
+            Ok(k) if (1000..=40000).contains(&k) => k,
+            _ => {
+                eprintln!(
+                    "usage: ringa [white|warm|golden|studio|daylight|cool|<kelvin>]"
+                );
+                std::process::exit(1);
+            }
+        },
+    };
+
+    render::kelvin_to_rgb(kelvin)
 }

src/render.rs

@@ -1,183 +1,287 @@
-//! Ring light glow rendering.
-//!
-//! The ring is a thick stroke centered on a single rounded rectangle SDF.
-//! This guarantees uniform width everywhere, including at corners.
-//! Glow bleeds outward and inward from the stroke edges.
+//! GPU renderer using wgpu + WGSL fragment shader.
 
-use tiny_skia::Pixmap;
+use anyhow::{Context, Result};
+use raw_window_handle::{RawDisplayHandle, RawWindowHandle};
+use wgpu::util::DeviceExt;
 
-// The rounded rect centerline is inset this far from the screen edge.
-// The visible ring extends RING_HALF_WIDTH on either side of this line.
-const CENTERLINE_INSET: f32 = 80.0;
-// Half the stroke width. Total ring thickness = 2 * RING_HALF_WIDTH.
-const RING_HALF_WIDTH: f32 = 30.0;
-// Corner radius of the centerline rounded rect.
-const CORNER_RADIUS: f32 = 30.0;
+#[repr(C)]
+#[derive(Clone, Copy, bytemuck::Pod, bytemuck::Zeroable)]
+struct Uniforms {
+    screen_size: [f32; 2],
+    cursor_pos: [f32; 2],
+    outer_inset: f32,
+    outer_corner_radius: f32,
+    ring_width: f32,
+    inner_corner_radius: f32,
+    glow_spread: f32,
+    cursor_avoid_radius: f32,
+    cursor_avoid_falloff: f32,
+    _pad: f32,
+    color: [f32; 3],
+    _pad2: f32,
+}
 
-// Soft glow spread outward and inward from the ring edges.
-const GLOW_SPREAD: f32 = 50.0;
-const RING_ALPHA: f32 = 1.00;
+pub struct GpuRenderer {
+    surface: wgpu::Surface<'static>,
+    device: wgpu::Device,
+    queue: wgpu::Queue,
+    pipeline: wgpu::RenderPipeline,
+    uniform_buffer: wgpu::Buffer,
+    bind_group: wgpu::BindGroup,
+    uniforms: Uniforms,
+}
 
-// Cursor avoidance.
-const CURSOR_AVOID_RADIUS: f32 = 120.0;
-const CURSOR_AVOID_FALLOFF: f32 = 60.0;
+impl GpuRenderer {
+    pub fn new(
+        display_handle: RawDisplayHandle,
+        window_handle: RawWindowHandle,
+        width: u32,
+        height: u32,
+        color: [f32; 3],
+    ) -> Result<Self> {
+        let instance = wgpu::Instance::new(wgpu::InstanceDescriptor {
+            backends: wgpu::Backends::VULKAN,
+            flags: wgpu::InstanceFlags::empty(),
+            backend_options: Default::default(),
+            memory_budget_thresholds: Default::default(),
+            display: None,
+        });
 
-/// Render the ring light glow into `pixmap`, avoiding the cursor at `(cx, cy)`.
-pub fn render_ring_light(pixmap: &mut Pixmap, cx: f64, cy: f64) {
-    let w = pixmap.width() as f32;
-    let h = pixmap.height() as f32;
-    let cx = cx as f32;
-    let cy = cy as f32;
-    let wi = pixmap.width() as usize;
-    let hi = pixmap.height() as usize;
-
-    let data = pixmap.data_mut();
-    data.fill(0);
-
-    let screen_cx = w / 2.0;
-    let screen_cy = h / 2.0;
-
-    // Single rounded rect centerline.
-    let hw = w / 2.0 - CENTERLINE_INSET;
-    let hh = h / 2.0 - CENTERLINE_INSET;
-
-    // The affected band: from (CENTERLINE_INSET - RING_HALF_WIDTH - GLOW_SPREAD)
-    // to (CENTERLINE_INSET + RING_HALF_WIDTH + GLOW_SPREAD) from each screen edge.
-    let band_start = (CENTERLINE_INSET - RING_HALF_WIDTH - GLOW_SPREAD)
-        .max(0.0)
-        .floor() as usize;
-    let band_end = (CENTERLINE_INSET + RING_HALF_WIDTH + GLOW_SPREAD).ceil() as usize + 1;
-
-    // Top band
-    for py in band_start..band_end.min(hi) {
-        for px in 0..wi {
-            let a = pixel_alpha(px as f32, py as f32, screen_cx, screen_cy, hw, hh, cx, cy);
-            write_pixel(data, py * wi + px, a);
+        let surface = unsafe {
+            instance.create_surface_unsafe(wgpu::SurfaceTargetUnsafe::RawHandle {
+                raw_display_handle: Some(display_handle),
+                raw_window_handle: window_handle,
+            })
         }
+        .context("create wgpu surface")?;
+
+        let adapter = pollster::block_on(instance.request_adapter(&wgpu::RequestAdapterOptions {
+            compatible_surface: Some(&surface),
+            power_preference: wgpu::PowerPreference::LowPower,
+            ..Default::default()
+        }))
+        .context("no compatible GPU adapter")?;
+
+        let (device, queue) = pollster::block_on(adapter.request_device(
+            &wgpu::DeviceDescriptor {
+                label: Some("ringa"),
+                ..Default::default()
+            },
+        ))
+        .context("request GPU device")?;
+
+        let caps = surface.get_capabilities(&adapter);
+        let format = caps
+            .formats
+            .iter()
+            .find(|f| f.is_srgb())
+            .copied()
+            .or_else(|| caps.formats.first().copied())
+            .context("no supported surface format")?;
+
+        let alpha_mode = if caps
+            .alpha_modes
+            .contains(&wgpu::CompositeAlphaMode::PreMultiplied)
+        {
+            wgpu::CompositeAlphaMode::PreMultiplied
+        } else {
+            caps.alpha_modes
+                .first()
+                .copied()
+                .context("no supported alpha mode")?
+        };
+
+        surface.configure(
+            &device,
+            &wgpu::SurfaceConfiguration {
+                usage: wgpu::TextureUsages::RENDER_ATTACHMENT,
+                format,
+                width,
+                height,
+                present_mode: wgpu::PresentMode::Mailbox,
+                desired_maximum_frame_latency: 1,
+                alpha_mode,
+                view_formats: vec![],
+            },
+        );
+
+        let shader = device.create_shader_module(wgpu::ShaderModuleDescriptor {
+            label: Some("ring shader"),
+            source: wgpu::ShaderSource::Wgsl(include_str!("shader.wgsl").into()),
+        });
+
+        let uniforms = Uniforms {
+            screen_size: [width as f32, height as f32],
+            cursor_pos: [width as f32 / 2.0, height as f32 / 2.0],
+            outer_inset: 65.0,
+            outer_corner_radius: 36.0,
+            ring_width: 50.0,
+            inner_corner_radius: 100.0,
+            glow_spread: 50.0,
+            cursor_avoid_radius: 300.0,
+            cursor_avoid_falloff: 1500.0,
+            _pad: 0.0,
+            color,
+            _pad2: 0.0,
+        };
+
+        let uniform_buffer = device.create_buffer_init(&wgpu::util::BufferInitDescriptor {
+            label: Some("uniforms"),
+            contents: bytemuck::bytes_of(&uniforms),
+            usage: wgpu::BufferUsages::UNIFORM | wgpu::BufferUsages::COPY_DST,
+        });
+
+        let bind_group_layout = device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
+            label: Some("uniforms layout"),
+            entries: &[wgpu::BindGroupLayoutEntry {
+                binding: 0,
+                visibility: wgpu::ShaderStages::FRAGMENT,
+                ty: wgpu::BindingType::Buffer {
+                    ty: wgpu::BufferBindingType::Uniform,
+                    has_dynamic_offset: false,
+                    min_binding_size: None,
+                },
+                count: None,
+            }],
+        });
+
+        let bind_group = device.create_bind_group(&wgpu::BindGroupDescriptor {
+            label: Some("uniforms bind group"),
+            layout: &bind_group_layout,
+            entries: &[wgpu::BindGroupEntry {
+                binding: 0,
+                resource: uniform_buffer.as_entire_binding(),
+            }],
+        });
+
+        let pipeline_layout = device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
+            label: Some("ring pipeline layout"),
+            bind_group_layouts: &[Some(&bind_group_layout)],
+            immediate_size: 0,
+        });
+
+        let pipeline = device.create_render_pipeline(&wgpu::RenderPipelineDescriptor {
+            label: Some("ring pipeline"),
+            layout: Some(&pipeline_layout),
+            vertex: wgpu::VertexState {
+                module: &shader,
+                entry_point: Some("vs_main"),
+                buffers: &[],
+                compilation_options: Default::default(),
+            },
+            fragment: Some(wgpu::FragmentState {
+                module: &shader,
+                entry_point: Some("fs_main"),
+                targets: &[Some(wgpu::ColorTargetState {
+                    format,
+                    blend: Some(wgpu::BlendState::PREMULTIPLIED_ALPHA_BLENDING),
+                    write_mask: wgpu::ColorWrites::ALL,
+                })],
+                compilation_options: Default::default(),
+            }),
+            primitive: wgpu::PrimitiveState {
+                topology: wgpu::PrimitiveTopology::TriangleList,
+                ..Default::default()
+            },
+            depth_stencil: None,
+            multisample: Default::default(),
+            multiview_mask: None,
+            cache: None,
+        });
+
+        Ok(Self {
+            surface,
+            device,
+            queue,
+            pipeline,
+            uniform_buffer,
+            bind_group,
+            uniforms,
+        })
     }
 
-    // Bottom band
-    let bottom_start = hi.saturating_sub(band_end);
-    let bottom_end = hi.saturating_sub(band_start);
-    for py in bottom_start..bottom_end.min(hi) {
-        if py < band_end {
-            continue;
-        }
-        for px in 0..wi {
-            let a = pixel_alpha(px as f32, py as f32, screen_cx, screen_cy, hw, hh, cx, cy);
-            write_pixel(data, py * wi + px, a);
-        }
-    }
+    pub fn render(&mut self, cx: f64, cy: f64) -> Result<()> {
+        self.uniforms.cursor_pos = [cx as f32, cy as f32];
+        self.queue
+            .write_buffer(&self.uniform_buffer, 0, bytemuck::bytes_of(&self.uniforms));
 
-    // Left band (between top and bottom)
-    let vert_start = band_end.min(hi);
-    let vert_end = hi.saturating_sub(band_end);
-    for py in vert_start..vert_end {
-        for px in band_start..band_end.min(wi) {
-            let a = pixel_alpha(px as f32, py as f32, screen_cx, screen_cy, hw, hh, cx, cy);
-            write_pixel(data, py * wi + px, a);
-        }
-    }
-
-    // Right band (between top and bottom)
-    let right_start = wi.saturating_sub(band_end);
-    let right_end = wi.saturating_sub(band_start);
-    for py in vert_start..vert_end {
-        for px in right_start..right_end.min(wi) {
-            if px < band_end {
-                continue;
+        let frame = match self.surface.get_current_texture() {
+            wgpu::CurrentSurfaceTexture::Success(tex)
+            | wgpu::CurrentSurfaceTexture::Suboptimal(tex) => tex,
+            wgpu::CurrentSurfaceTexture::Timeout | wgpu::CurrentSurfaceTexture::Occluded => {
+                return Ok(())
             }
-            let a = pixel_alpha(px as f32, py as f32, screen_cx, screen_cy, hw, hh, cx, cy);
-            write_pixel(data, py * wi + px, a);
+            wgpu::CurrentSurfaceTexture::Outdated | wgpu::CurrentSurfaceTexture::Lost => {
+                anyhow::bail!("surface lost")
+            }
+            _ => anyhow::bail!("unexpected surface texture state"),
+        };
+
+        let view = frame
+            .texture
+            .create_view(&wgpu::TextureViewDescriptor::default());
+
+        let mut encoder = self
+            .device
+            .create_command_encoder(&wgpu::CommandEncoderDescriptor {
+                label: Some("ring encoder"),
+            });
+
+        {
+            let mut pass = encoder.begin_render_pass(&wgpu::RenderPassDescriptor {
+                label: Some("ring pass"),
+                color_attachments: &[Some(wgpu::RenderPassColorAttachment {
+                    view: &view,
+                    resolve_target: None,
+                    ops: wgpu::Operations {
+                        load: wgpu::LoadOp::Clear(wgpu::Color::TRANSPARENT),
+                        store: wgpu::StoreOp::Store,
+                    },
+                    depth_slice: None,
+                })],
+                depth_stencil_attachment: None,
+                ..Default::default()
+            });
+
+            pass.set_pipeline(&self.pipeline);
+            pass.set_bind_group(0, &self.bind_group, &[]);
+            pass.draw(0..3, 0..1);
         }
+
+        self.queue.submit(std::iter::once(encoder.finish()));
+        frame.present();
+        Ok(())
     }
 }
 
-/// Convert a tiny-skia `Pixmap` (premultiplied RGBA) to Wayland's ARGB8888
-/// format (premultiplied BGRA in memory on little-endian).
-pub fn pixmap_to_argb8888(pixmap: &Pixmap, out: &mut [u8]) {
-    let src = pixmap.data();
-    assert_eq!(out.len(), src.len());
-    for (dst, src) in out.chunks_exact_mut(4).zip(src.chunks_exact(4)) {
-        dst[0] = src[2]; // B
-        dst[1] = src[1]; // G
-        dst[2] = src[0]; // R
-        dst[3] = src[3]; // A
-    }
-}
+/// Convert color temperature in Kelvin to linear RGB.
+/// Based on Tanner Helland's algorithm.
+pub fn kelvin_to_rgb(kelvin: u32) -> [f32; 3] {
+    let temp = (kelvin as f64 / 100.0).clamp(10.0, 400.0);
 
-// ---- Private helpers --------------------------------------------------------
-
-/// Compute glow alpha for a single pixel using a single rounded rect SDF.
-///
-/// The ring is all pixels where |sdf| <= RING_HALF_WIDTH (a thick stroke).
-/// Glow fades from the ring edges outward/inward over GLOW_SPREAD pixels.
-#[allow(clippy::too_many_arguments)]
-fn pixel_alpha(
-    px: f32,
-    py: f32,
-    scx: f32,
-    scy: f32,
-    hw: f32,
-    hh: f32,
-    cursor_x: f32,
-    cursor_y: f32,
-) -> u8 {
-    // Signed distance to the rounded rect centerline.
-    let sd = rounded_rect_sdf(px, py, scx, scy, hw, hh, CORNER_RADIUS);
-    let abs_sd = sd.abs();
-
-    let glow = if abs_sd <= RING_HALF_WIDTH {
-        // Inside the ring stroke — full brightness.
-        RING_ALPHA
+    let r = if temp <= 66.0 {
+        1.0
     } else {
-        // Distance from the nearest ring edge.
-        let dist = abs_sd - RING_HALF_WIDTH;
-        if dist >= GLOW_SPREAD {
-            return 0;
-        }
-        let t = dist / GLOW_SPREAD;
-        smoothstep(1.0 - t) * RING_ALPHA
+        let r = 329.698727446 * (temp - 60.0).powf(-0.1332047592);
+        (r / 255.0).clamp(0.0, 1.0)
     };
 
-    // Cursor avoidance.
-    let dx = px - cursor_x;
-    let dy = py - cursor_y;
-    let cursor_dist = (dx * dx + dy * dy).sqrt();
-    let cursor_factor =
-        smoothstep(((cursor_dist - CURSOR_AVOID_RADIUS) / CURSOR_AVOID_FALLOFF).clamp(0.0, 1.0));
+    let g = if temp <= 66.0 {
+        let g = 99.4708025861 * temp.ln() - 161.1195681661;
+        (g / 255.0).clamp(0.0, 1.0)
+    } else {
+        let g = 288.1221695283 * (temp - 60.0).powf(-0.0755148492);
+        (g / 255.0).clamp(0.0, 1.0)
+    };
 
-    let alpha = glow * cursor_factor;
-    (alpha * 255.0) as u8
-}
+    let b = if temp >= 66.0 {
+        1.0
+    } else if temp <= 19.0 {
+        0.0
+    } else {
+        let b = 138.5177312231 * (temp - 10.0).ln() - 305.0447927307;
+        (b / 255.0).clamp(0.0, 1.0)
+    };
 
-/// Signed distance field for a rounded rectangle.
-///
-/// Returns negative inside, positive outside, zero on the boundary.
-fn rounded_rect_sdf(px: f32, py: f32, cx: f32, cy: f32, hw: f32, hh: f32, r: f32) -> f32 {
-    let dx = (px - cx).abs() - (hw - r);
-    let dy = (py - cy).abs() - (hh - r);
-    let outside = dx.max(0.0);
-    let outside_y = dy.max(0.0);
-    let outside_len = (outside * outside + outside_y * outside_y).sqrt();
-    let inside = dx.max(dy).min(0.0);
-    outside_len + inside - r
-}
-
-/// Write a premultiplied white pixel with the given alpha.
-fn write_pixel(data: &mut [u8], idx: usize, alpha: u8) {
-    if alpha == 0 {
-        return;
-    }
-    let offset = idx * 4;
-    data[offset] = alpha;
-    data[offset + 1] = alpha;
-    data[offset + 2] = alpha;
-    data[offset + 3] = alpha;
-}
-
-/// Hermite smoothstep: 3t^2 - 2t^3
-fn smoothstep(t: f32) -> f32 {
-    let t = t.clamp(0.0, 1.0);
-    t * t * (3.0 - 2.0 * t)
+    [r as f32, g as f32, b as f32]
 }

src/shader.wgsl

@@ -0,0 +1,75 @@
+struct Uniforms {
+    screen_size: vec2f,
+    cursor_pos: vec2f,
+    // Outer rounded rect: inset from screen edge
+    outer_inset: f32,
+    outer_corner_radius: f32,
+    // Inner rounded rect: inset further by ring_width
+    ring_width: f32,
+    inner_corner_radius: f32,
+    // Glow
+    glow_spread: f32,
+    // Cursor avoidance
+    cursor_avoid_radius: f32,
+    cursor_avoid_falloff: f32,
+    // Color temperature (linear RGB)
+    _pad: f32,
+    color: vec3f,
+    _pad2: f32,
+}
+
+@group(0) @binding(0) var<uniform> u: Uniforms;
+
+@vertex
+fn vs_main(@builtin(vertex_index) i: u32) -> @builtin(position) vec4f {
+    let x = f32(i32(i) / 2) * 4.0 - 1.0;
+    let y = f32(i32(i) % 2) * 4.0 - 1.0;
+    return vec4f(x, y, 0.0, 1.0);
+}
+
+// Rounded rectangle SDF (Inigo Quilez).
+fn rounded_rect_sdf(p: vec2f, h: vec2f, r: f32) -> f32 {
+    let q = abs(p) - h + vec2f(r);
+    return length(max(q, vec2f(0.0))) + min(max(q.x, q.y), 0.0) - r;
+}
+
+@fragment
+fn fs_main(@builtin(position) pos: vec4f) -> @location(0) vec4f {
+    let center = u.screen_size / 2.0;
+    let p = pos.xy - center;
+
+    // Outer rounded rect
+    let outer_half = center - vec2f(u.outer_inset);
+    let sd_outer = rounded_rect_sdf(p, outer_half, u.outer_corner_radius);
+
+    // Inner rounded rect
+    let inner_half = outer_half - vec2f(u.ring_width);
+    let sd_inner = rounded_rect_sdf(p, inner_half, u.inner_corner_radius);
+
+    // Ring: inside outer AND outside inner = solid at full brightness
+    var alpha: f32;
+    if sd_outer <= 0.0 && sd_inner >= 0.0 {
+        alpha = 1.0;
+    } else if sd_outer > 0.0 {
+        // Gaussian glow fading outward
+        let t = sd_outer / u.glow_spread;
+        alpha = exp(-3.0 * t * t);
+    } else {
+        // Gaussian glow fading inward
+        let t = -sd_inner / u.glow_spread;
+        alpha = exp(-3.0 * t * t);
+    }
+
+    // Cursor avoidance
+    let cursor_dist = distance(pos.xy, u.cursor_pos);
+    let avoid_start = u.cursor_avoid_radius;
+    let avoid_end = u.cursor_avoid_radius + u.cursor_avoid_falloff;
+    if cursor_dist < avoid_end {
+        let t_avoid = clamp((cursor_dist - avoid_start) / (avoid_end - avoid_start), 0.0, 1.0);
+        alpha *= t_avoid * t_avoid * (3.0 - 2.0 * t_avoid);
+    }
+
+    // Premultiplied color output
+    let rgb = u.color * alpha;
+    return vec4f(rgb, alpha);
+}

src/wayland.rs

@@ -1,12 +1,10 @@
 //! Wayland backend: layer-shell overlay with keyboard and pointer tracking.
 //!
-//! Creates a full-screen layer-shell surface for the ring light overlay.
-//! Tracks pointer position for cursor avoidance and captures keyboard
-//! for Escape to exit.
+//! Creates a full-screen layer-shell surface. Exposes raw Wayland handles
+//! for wgpu surface creation. No CPU pixel buffers — GPU renders directly.
 
 use std::collections::VecDeque;
-use std::io::Write;
-use std::os::fd::AsFd;
+use std::ptr::NonNull;
 use std::time::Duration;
 
 use wayland_client::{
@@ -15,32 +13,28 @@ use wayland_client::{
         wl_buffer, wl_compositor, wl_keyboard, wl_pointer, wl_region, wl_registry, wl_seat,
         wl_shm, wl_shm_pool, wl_surface,
     },
-    Connection, Dispatch, EventQueue, QueueHandle, WEnum,
+    Connection, Dispatch, EventQueue, Proxy, QueueHandle, WEnum,
 };
 use wayland_protocols_wlr::layer_shell::v1::client::{zwlr_layer_shell_v1, zwlr_layer_surface_v1};
 
 use anyhow::{Context, Result};
+use raw_window_handle::{
+    RawDisplayHandle, RawWindowHandle, WaylandDisplayHandle, WaylandWindowHandle,
+};
 
-// Linux evdev key code for Escape.
 const KEY_ESCAPE: u32 = 1;
 
-/// Events emitted by the Wayland backend to the main loop.
+#[allow(dead_code)]
 pub enum OverlayEvent {
-    /// Escape pressed or surface closed by compositor.
     Close,
-    /// Pointer moved to (x, y) in surface-local coordinates.
     CursorMoved(f64, f64),
-    /// Surface (re-)configured with new dimensions.
     Configure(u32, u32),
 }
 
-// ===========================================================================
-// WaylandBackend — public API.
-// ===========================================================================
-
 pub struct WaylandBackend {
     state: WaylandState,
     event_queue: EventQueue<WaylandState>,
+    #[allow(dead_code)]
     queue_handle: QueueHandle<WaylandState>,
 }
 
@@ -73,9 +67,6 @@ impl WaylandBackend {
         if state.compositor.is_none() {
             anyhow::bail!("compositor does not advertise wl_compositor");
         }
-        if state.shm.is_none() {
-            anyhow::bail!("compositor does not advertise wl_shm");
-        }
         if state.layer_shell.is_none() {
             anyhow::bail!(
                 "compositor does not support wlr-layer-shell-unstable-v1. \
@@ -105,10 +96,23 @@ impl WaylandBackend {
         (self.state.screen_w, self.state.screen_h)
     }
 
-    /// Wait up to `timeout` for the next event. Returns `None` on timeout
-    /// or if the surface has been torn down.
+    /// Raw Wayland display handle for wgpu surface creation.
+    pub fn raw_display_handle(&self) -> Result<RawDisplayHandle> {
+        let ptr = self.state.conn.backend().display_ptr() as *mut std::ffi::c_void;
+        let nn = NonNull::new(ptr).context("wl_display pointer is null")?;
+        Ok(RawDisplayHandle::Wayland(WaylandDisplayHandle::new(nn)))
+    }
+
+    /// Raw Wayland surface handle for wgpu surface creation.
+    pub fn raw_window_handle(&self) -> Result<RawWindowHandle> {
+        let surface = self.state.surface.as_ref().context("no wl_surface")?;
+        let ptr = surface.id().as_ptr() as *mut std::ffi::c_void;
+        let nn = NonNull::new(ptr).context("wl_surface pointer is null")?;
+        Ok(RawWindowHandle::Wayland(WaylandWindowHandle::new(nn)))
+    }
+
+    /// Wait up to `timeout` for the next event. Returns `None` on timeout.
     pub fn next_event(&mut self, timeout: Duration) -> Result<Option<OverlayEvent>> {
-        // First drain any already-buffered events.
         if let Some(ev) = self.state.pending_events.pop_front() {
             return Ok(Some(ev));
         }
@@ -116,62 +120,71 @@ impl WaylandBackend {
             return Ok(None);
         }
 
-        // Flush outgoing requests, then poll the fd with a timeout.
         self.event_queue
             .flush()
             .context("flush before poll")?;
 
-        let read_guard = self.event_queue.prepare_read().unwrap();
-        let fd = read_guard.connection_fd();
-        let mut pollfd = [nix::poll::PollFd::new(fd, nix::poll::PollFlags::POLLIN)];
-        let millis = timeout.as_millis().min(u16::MAX as u128) as u16;
-        let n = nix::poll::poll(&mut pollfd, millis).context("poll")?;
-
-        if n > 0 {
-            // Data available — read and dispatch.
-            let _ = read_guard.read();
-        } else {
-            // Timeout — drop the read guard (cancels the read).
-            drop(read_guard);
-            return Ok(None);
-        }
-
+        // Try to dispatch any already-buffered events first.
         self.event_queue
             .dispatch_pending(&mut self.state)
-            .context("dispatch_pending")?;
+            .context("dispatch_pending (pre-poll)")?;
+        if let Some(ev) = self.state.pending_events.pop_front() {
+            // Re-run coalescing below with this event re-queued.
+            self.state.pending_events.push_front(ev);
+        } else {
+            // No pending events — try to read from the fd with a timeout.
+            match self.event_queue.prepare_read() {
+                Some(read_guard) => {
+                    let fd = read_guard.connection_fd();
+                    let mut pollfd =
+                        [nix::poll::PollFd::new(fd, nix::poll::PollFlags::POLLIN)];
+                    let millis = timeout.as_millis().min(u16::MAX as u128) as u16;
+                    let n = nix::poll::poll(&mut pollfd, millis).context("poll")?;
 
-        Ok(self.state.pending_events.pop_front())
-    }
+                    if n > 0 {
+                        let _ = read_guard.read();
+                    } else {
+                        drop(read_guard);
+                        return Ok(None);
+                    }
 
-    /// Copy an ARGB8888 pixel buffer to the overlay surface via wl_shm.
-    pub fn present(&mut self, pixels: &[u8], width: u32, height: u32) -> Result<()> {
-        let shm = self.state.shm.as_ref().context("wl_shm not available")?;
-        let surface = self
-            .state
-            .surface
-            .as_ref()
-            .context("wl_surface not available")?;
+                    self.event_queue
+                        .dispatch_pending(&mut self.state)
+                        .context("dispatch_pending (post-read)")?;
+                }
+                None => {
+                    // Another queue has events buffered — dispatch ours.
+                    self.event_queue
+                        .dispatch_pending(&mut self.state)
+                        .context("dispatch_pending (no-read)")?;
+                }
+            }
+        }
 
-        let stride = width * 4;
-        let mut file = tempfile::tempfile().context("create shm tempfile")?;
-        file.write_all(pixels).context("write pixel buffer")?;
-
-        let pool = shm.create_pool(file.as_fd(), pixels.len() as i32, &self.queue_handle, ());
-        let buffer = pool.create_buffer(
-            0,
-            width as i32,
-            height as i32,
-            stride as i32,
-            wl_shm::Format::Argb8888,
-            &self.queue_handle,
-            (),
-        );
-
-        surface.attach(Some(&buffer), 0, 0);
-        surface.damage_buffer(0, 0, width as i32, height as i32);
-        surface.commit();
-        pool.destroy();
-        Ok(())
+        // Coalesce CursorMoved events.
+        let mut last_cursor: Option<(f64, f64)> = None;
+        let mut first_non_cursor: Option<OverlayEvent> = None;
+        while let Some(ev) = self.state.pending_events.pop_front() {
+            match ev {
+                OverlayEvent::CursorMoved(x, y) => last_cursor = Some((x, y)),
+                other => {
+                    first_non_cursor = Some(other);
+                    break;
+                }
+            }
+        }
+        if let Some(ev) = first_non_cursor {
+            if let Some((x, y)) = last_cursor {
+                self.state
+                    .pending_events
+                    .push_front(OverlayEvent::CursorMoved(x, y));
+            }
+            return Ok(Some(ev));
+        }
+        if let Some((x, y)) = last_cursor {
+            return Ok(Some(OverlayEvent::CursorMoved(x, y)));
+        }
+        Ok(None)
     }
 }
 
@@ -195,6 +208,7 @@ struct WaylandState {
     conn: Connection,
 
     compositor: Option<wl_compositor::WlCompositor>,
+    #[allow(dead_code)]
     shm: Option<wl_shm::WlShm>,
     seat: Option<wl_seat::WlSeat>,
     layer_shell: Option<zwlr_layer_shell_v1::ZwlrLayerShellV1>,
@@ -221,6 +235,11 @@ impl WaylandState {
             .context("zwlr_layer_shell_v1 not available")?;
 
         let surface = compositor.create_surface(qh, ());
+
+        let input_region = compositor.create_region(qh, ());
+        input_region.add(0, 0, i32::MAX, i32::MAX);
+        surface.set_input_region(Some(&input_region));
+
         let layer_surface = layer_shell.get_layer_surface(
             &surface,
             None,
@@ -347,12 +366,8 @@ impl Dispatch<wl_pointer::WlPointer, ()> for WaylandState {
                 surface_x,
                 surface_y,
                 ..
-            } => {
-                state
-                    .pending_events
-                    .push_back(OverlayEvent::CursorMoved(surface_x, surface_y));
             }
-            wl_pointer::Event::Motion {
+            | wl_pointer::Event::Motion {
                 surface_x,
                 surface_y,
                 ..