finish gradient support

piet-wgsl/src/engine.rs

@@ -17,7 +17,7 @@
 use std::{
     borrow::Cow,
     collections::{hash_map::Entry, HashMap},
-    num::NonZeroU64,
+    num::{NonZeroU32, NonZeroU64},
     sync::atomic::{AtomicU64, Ordering},
 };
 
@@ -25,7 +25,8 @@ use futures_intrusive::channel::shared::GenericOneshotReceiver;
 use parking_lot::RawMutex;
 use wgpu::{
     util::DeviceExt, BindGroup, BindGroupLayout, Buffer, BufferAsyncError, BufferSlice, BufferView,
-    ComputePipeline, Device, Queue,
+    ComputePipeline, Device, Queue, Texture, TextureAspect, TextureFormat, TextureUsages,
+    TextureView, TextureViewDimension,
 };
 
 pub type Error = Box<dyn std::error::Error>;
@@ -58,12 +59,27 @@ pub struct BufProxy {
     id: Id,
 }
 
+#[derive(Clone, Copy)]
+pub struct ImageProxy {
+    width: u32,
+    height: u32,
+    // TODO: format
+    id: Id,
+}
+
+#[derive(Clone, Copy)]
+pub enum ResourceProxy {
+    Buf(BufProxy),
+    Image(ImageProxy),
+}
+
 pub enum Command {
     Upload(BufProxy, Vec<u8>),
+    UploadImage(ImageProxy, Vec<u8>),
     // Discussion question: third argument is vec of resources?
     // Maybe use tricks to make more ergonomic?
     // Alternative: provide bufs & images as separate sequences, like piet-gpu.
-    Dispatch(ShaderId, (u32, u32, u32), Vec<BufProxy>),
+    Dispatch(ShaderId, (u32, u32, u32), Vec<ResourceProxy>),
     Download(BufProxy),
     Clear(BufProxy, u64, Option<NonZeroU64>),
 }
@@ -92,6 +108,7 @@ pub enum BindType {
 #[derive(Default)]
 struct BindMap {
     buf_map: HashMap<Id, Buffer>,
+    image_map: HashMap<Id, (Texture, TextureView)>,
 }
 
 impl Engine {
@@ -132,6 +149,16 @@ impl Engine {
                     },
                     count: None,
                 },
+                BindType::ImageRead => wgpu::BindGroupLayoutEntry {
+                    binding: i as u32,
+                    visibility: wgpu::ShaderStages::COMPUTE,
+                    ty: wgpu::BindingType::Texture {
+                        sample_type: wgpu::TextureSampleType::Float { filterable: true },
+                        view_dimension: wgpu::TextureViewDimension::D2,
+                        multisampled: false,
+                    },
+                    count: None,
+                },
                 _ => todo!(),
             })
             .collect::<Vec<_>>();
@@ -182,6 +209,58 @@ impl Engine {
                     });
                     bind_map.insert_buf(buf_proxy.id, buf);
                 }
+                Command::UploadImage(image_proxy, bytes) => {
+                    let buf = device.create_buffer_init(&wgpu::util::BufferInitDescriptor {
+                        label: None,
+                        contents: &bytes,
+                        usage: wgpu::BufferUsages::COPY_SRC,
+                    });
+                    let texture = device.create_texture(&wgpu::TextureDescriptor {
+                        label: None,
+                        size: wgpu::Extent3d {
+                            width: image_proxy.width,
+                            height: image_proxy.height,
+                            depth_or_array_layers: 1,
+                        },
+                        mip_level_count: 1,
+                        sample_count: 1,
+                        dimension: wgpu::TextureDimension::D2,
+                        usage: TextureUsages::TEXTURE_BINDING | TextureUsages::COPY_DST,
+                        format: TextureFormat::Rgba8Unorm,
+                    });
+                    let texture_view = texture.create_view(&wgpu::TextureViewDescriptor {
+                        label: None,
+                        dimension: Some(TextureViewDimension::D2),
+                        aspect: TextureAspect::All,
+                        mip_level_count: None,
+                        base_mip_level: 0,
+                        base_array_layer: 0,
+                        array_layer_count: None,
+                        format: Some(TextureFormat::Rgba8Unorm),
+                    });
+                    encoder.copy_buffer_to_texture(
+                        wgpu::ImageCopyBuffer {
+                            buffer: &buf,
+                            layout: wgpu::ImageDataLayout {
+                                offset: 0,
+                                bytes_per_row: NonZeroU32::new(image_proxy.width * 4),
+                                rows_per_image: None,
+                            },
+                        },
+                        wgpu::ImageCopyTexture {
+                            texture: &texture,
+                            mip_level: 0,
+                            origin: wgpu::Origin3d { x: 0, y: 0, z: 0 },
+                            aspect: TextureAspect::All,
+                        },
+                        wgpu::Extent3d {
+                            width: image_proxy.width,
+                            height: image_proxy.height,
+                            depth_or_array_layers: 1,
+                        },
+                    );
+                    bind_map.insert_image(image_proxy.id, texture, texture_view)
+                }
                 Command::Dispatch(shader_id, wg_size, bindings) => {
                     println!("dispatching {:?} with {} bindings", wg_size, bindings.len());
                     let shader = &self.shaders[shader_id.0];
@@ -226,13 +305,23 @@ impl Recording {
         buf_proxy
     }
 
-    pub fn dispatch(
-        &mut self,
-        shader: ShaderId,
-        wg_size: (u32, u32, u32),
-        resources: impl Into<Vec<BufProxy>>,
-    ) {
-        self.push(Command::Dispatch(shader, wg_size, resources.into()));
+    pub fn upload_image(&mut self, width: u32, height: u32, data: impl Into<Vec<u8>>) -> ImageProxy {
+        let data = data.into();
+        let image_proxy = ImageProxy::new(width, height);
+        self.push(Command::UploadImage(image_proxy, data));
+        image_proxy
+    }
+
+    pub fn dispatch<R>(&mut self, shader: ShaderId, wg_size: (u32, u32, u32), resources: R)
+    where
+        R: IntoIterator,
+        R::Item: Into<ResourceProxy>,
+    {
+        self.push(Command::Dispatch(
+            shader,
+            wg_size,
+            resources.into_iter().map(|r| r.into()).collect(),
+        ));
     }
 
     pub fn download(&mut self, buf: BufProxy) {
@@ -251,6 +340,35 @@ impl BufProxy {
     }
 }
 
+impl ImageProxy {
+    pub fn new(width: u32, height: u32) -> Self {
+        let id = Id::next();
+        ImageProxy { width, height, id }
+    }
+}
+
+impl ResourceProxy {
+    pub fn new_buf(size: u64) -> Self {
+        Self::Buf(BufProxy::new(size))
+    }
+
+    pub fn new_image(width: u32, height: u32) -> Self {
+        Self::Image(ImageProxy::new(width, height))
+    }
+}
+
+impl From<BufProxy> for ResourceProxy {
+    fn from(value: BufProxy) -> Self {
+        Self::Buf(value)
+    }
+}
+
+impl From<ImageProxy> for ResourceProxy {
+    fn from(value: ImageProxy) -> Self {
+        Self::Image(value)
+    }
+}
+
 impl Id {
     pub fn next() -> Id {
         let val = ID_COUNTER.fetch_add(1, Ordering::Relaxed);
@@ -264,13 +382,19 @@ impl BindMap {
         self.buf_map.insert(id, buf);
     }
 
+    fn insert_image(&mut self, id: Id, image: Texture, image_view: TextureView) {
+        self.image_map.insert(id, (image, image_view));
+    }
+
     fn create_bind_group(
         &mut self,
         device: &Device,
         layout: &BindGroupLayout,
-        bindings: &[BufProxy],
+        bindings: &[ResourceProxy],
     ) -> Result<BindGroup, Error> {
         for proxy in bindings {
+            match proxy {
+                ResourceProxy::Buf(proxy) => {
                     if let Entry::Vacant(v) = self.buf_map.entry(proxy.id) {
                         let buf = device.create_buffer(&wgpu::BufferDescriptor {
                             label: None,
@@ -283,15 +407,54 @@ impl BindMap {
                         v.insert(buf);
                     }
                 }
+                ResourceProxy::Image(proxy) => {
+                    if let Entry::Vacant(v) = self.image_map.entry(proxy.id) {
+                        let texture = device.create_texture(&wgpu::TextureDescriptor {
+                            label: None,
+                            size: wgpu::Extent3d {
+                                width: proxy.width,
+                                height: proxy.height,
+                                depth_or_array_layers: 1,
+                            },
+                            mip_level_count: 1,
+                            sample_count: 1,
+                            dimension: wgpu::TextureDimension::D2,
+                            usage: TextureUsages::TEXTURE_BINDING | TextureUsages::COPY_DST,
+                            format: TextureFormat::Rgba8Unorm,
+                        });
+                        let texture_view = texture.create_view(&wgpu::TextureViewDescriptor {
+                            label: None,
+                            dimension: Some(TextureViewDimension::D2),
+                            aspect: TextureAspect::All,
+                            mip_level_count: None,
+                            base_mip_level: 0,
+                            base_array_layer: 0,
+                            array_layer_count: None,
+                            format: Some(TextureFormat::Rgba8Unorm),
+                        });
+                        v.insert((texture, texture_view));
+                    }
+                }
+            }
+        }
         let entries = bindings
             .iter()
             .enumerate()
-            .map(|(i, proxy)| {
+            .map(|(i, proxy)| match proxy {
+                ResourceProxy::Buf(proxy) => {
                     let buf = self.buf_map.get(&proxy.id).unwrap();
                     Ok(wgpu::BindGroupEntry {
                         binding: i as u32,
                         resource: buf.as_entire_binding(),
                     })
+                }
+                ResourceProxy::Image(proxy) => {
+                    let texture = self.image_map.get(&proxy.id).unwrap();
+                    Ok(wgpu::BindGroupEntry {
+                        binding: i as u32,
+                        resource: wgpu::BindingResource::TextureView(&texture.1),
+                    })
+                }
             })
             .collect::<Result<Vec<_>, Error>>()?;
         let bind_group = device.create_bind_group(&wgpu::BindGroupDescriptor {

piet-wgsl/src/main.rs

@@ -24,6 +24,7 @@ use wgpu::{Device, Limits, Queue};
 
 mod engine;
 mod pico_svg;
+mod ramp;
 mod render;
 mod shaders;
 mod test_scene;
@@ -58,7 +59,6 @@ fn dump_buf(buf: &[u32]) {
                 println!("{}: {:x} {}", i, val, f32::from_bits(*val));
             } else {
                 println!("{}: {:x}", i, val);
-
             }
         }
     }

piet-wgsl/src/ramp.rs

@@ -0,0 +1,137 @@
+use piet_scene::{Color, GradientStop, GradientStops};
+
+use std::collections::HashMap;
+
+const N_SAMPLES: usize = 512;
+const RETAINED_COUNT: usize = 64;
+
+#[derive(Default)]
+pub struct RampCache {
+    epoch: u64,
+    map: HashMap<GradientStops, (u32, u64)>,
+    data: Vec<u32>,
+}
+
+impl RampCache {
+    pub fn advance(&mut self) {
+        self.epoch += 1;
+        if self.map.len() > RETAINED_COUNT {
+            self.map
+                .retain(|_key, value| value.0 < RETAINED_COUNT as u32);
+            self.data.truncate(RETAINED_COUNT * N_SAMPLES);
+        }
+    }
+
+    pub fn add(&mut self, stops: &[GradientStop]) -> u32 {
+        if let Some(entry) = self.map.get_mut(stops) {
+            entry.1 = self.epoch;
+            entry.0
+        } else if self.map.len() < RETAINED_COUNT {
+            let id = (self.data.len() / N_SAMPLES) as u32;
+            self.data.extend(make_ramp(stops));
+            self.map.insert(stops.into(), (id, self.epoch));
+            id
+        } else {
+            let mut reuse = None;
+            for (stops, (id, epoch)) in &self.map {
+                if *epoch + 2 < self.epoch {
+                    reuse = Some((stops.to_owned(), *id));
+                    break;
+                }
+            }
+            if let Some((old_stops, id)) = reuse {
+                self.map.remove(&old_stops);
+                let start = id as usize * N_SAMPLES;
+                for (dst, src) in self.data[start..start + N_SAMPLES]
+                    .iter_mut()
+                    .zip(make_ramp(stops))
+                {
+                    *dst = src;
+                }
+                self.map.insert(stops.into(), (id, self.epoch));
+                id
+            } else {
+                let id = (self.data.len() / N_SAMPLES) as u32;
+                self.data.extend(make_ramp(stops));
+                self.map.insert(stops.into(), (id, self.epoch));
+                id
+            }
+        }
+    }
+
+    pub fn data(&self) -> &[u32] {
+        &self.data
+    }
+
+    pub fn width(&self) -> u32 {
+        N_SAMPLES as u32
+    }
+
+    pub fn height(&self) -> u32 {
+        (self.data.len() / N_SAMPLES) as u32
+    }
+}
+
+fn make_ramp<'a>(stops: &'a [GradientStop]) -> impl Iterator<Item = u32> + 'a {
+    let mut last_u = 0.0;
+    let mut last_c = ColorF64::from_color(stops[0].color);
+    let mut this_u = last_u;
+    let mut this_c = last_c;
+    let mut j = 0;
+    (0..N_SAMPLES).map(move |i| {
+        let u = (i as f64) / (N_SAMPLES - 1) as f64;
+        while u > this_u {
+            last_u = this_u;
+            last_c = this_c;
+            if let Some(s) = stops.get(j + 1) {
+                this_u = s.offset as f64;
+                this_c = ColorF64::from_color(s.color);
+                j += 1;
+            } else {
+                break;
+            }
+        }
+        let du = this_u - last_u;
+        let c = if du < 1e-9 {
+            this_c
+        } else {
+            last_c.lerp(&this_c, (u - last_u) / du)
+        };
+        c.to_premul_u32()
+    })
+}
+
+#[derive(Copy, Clone, Debug)]
+struct ColorF64([f64; 4]);
+
+impl ColorF64 {
+    fn from_color(color: Color) -> Self {
+        Self([
+            color.r as f64 / 255.0,
+            color.g as f64 / 255.0,
+            color.b as f64 / 255.0,
+            color.a as f64 / 255.0,
+        ])
+    }
+
+    fn lerp(&self, other: &Self, a: f64) -> Self {
+        fn l(x: f64, y: f64, a: f64) -> f64 {
+            x * (1.0 - a) + y * a
+        }
+        Self([
+            l(self.0[0], other.0[0], a),
+            l(self.0[1], other.0[1], a),
+            l(self.0[2], other.0[2], a),
+            l(self.0[3], other.0[3], a),
+        ])
+    }
+
+    fn to_premul_u32(&self) -> u32 {
+        let a = self.0[3].min(1.0).max(0.0);
+        let r = ((self.0[0] * a).min(1.0).max(0.0) * 255.0) as u32;
+        let g = ((self.0[1] * a).min(1.0).max(0.0) * 255.0) as u32;
+        let b = ((self.0[2] * a).min(1.0).max(0.0) * 255.0) as u32;
+        let a = (a * 255.0) as u32;
+        r | (g << 8) | (b << 16) | (a << 24)
+    }
+}

piet-wgsl/src/render.rs

@@ -4,7 +4,7 @@ use bytemuck::{Pod, Zeroable};
 use piet_scene::Scene;
 
 use crate::{
-    engine::{BufProxy, Recording},
+    engine::{BufProxy, Recording, ResourceProxy},
     shaders::{self, FullShaders, Shaders},
 };
 
@@ -87,7 +87,8 @@ pub fn render(scene: &Scene, shaders: &Shaders) -> (Recording, BufProxy) {
         [config_buf, scene_buf, reduced_buf, tagmonoid_buf],
     );
 
-    let path_coarse_wgs = (n_pathtag as u32 + shaders::PATH_COARSE_WG - 1) / shaders::PATH_COARSE_WG;
+    let path_coarse_wgs =
+        (n_pathtag as u32 + shaders::PATH_COARSE_WG - 1) / shaders::PATH_COARSE_WG;
     // TODO: more principled size calc
     let tiles_buf = BufProxy::new(4097 * 8);
     let segments_buf = BufProxy::new(256 * 24);
@@ -122,7 +123,29 @@ pub fn render(scene: &Scene, shaders: &Shaders) -> (Recording, BufProxy) {
 
 pub fn render_full(scene: &Scene, shaders: &FullShaders) -> (Recording, BufProxy) {
     let mut recording = Recording::default();
+    let mut ramps = crate::ramp::RampCache::default();
+    let mut drawdata_patches: Vec<(usize, u32)> = vec![];
     let data = scene.data();
+    let stop_data = &data.resources.stops;
+    for patch in &data.resources.patches {
+        use piet_scene::ResourcePatch;
+        match patch {
+            ResourcePatch::Ramp { offset, stops } => {
+                let ramp_id = ramps.add(&stop_data[stops.clone()]);
+                drawdata_patches.push((*offset, ramp_id));
+            }
+        }
+    }
+    let gradient_image = if drawdata_patches.is_empty() {
+        ResourceProxy::new_image(1, 1)
+    } else {
+        let data = ramps.data();
+        let width = ramps.width();
+        let height = ramps.height();
+        let data: &[u8] = bytemuck::cast_slice(data);
+        println!("gradient image: {}x{} ({} bytes)", width, height, data.len());
+        ResourceProxy::Image(recording.upload_image(width, height, data))
+    };
     let n_pathtag = data.tag_stream.len();
     let pathtag_padded = align_up(n_pathtag, 4 * shaders::PATHTAG_REDUCE_WG);
     // TODO: can compute size accurately, avoid reallocation
@@ -135,12 +158,27 @@ pub fn render_full(scene: &Scene, shaders: &FullShaders) -> (Recording, BufProxy
     let drawtag_base = size_to_words(scene.len());
     scene.extend(bytemuck::cast_slice(&data.drawtag_stream));
     let drawdata_base = size_to_words(scene.len());
+    if !drawdata_patches.is_empty() {
+        let mut pos = 0;
+        for patch in drawdata_patches {
+            let offset = patch.0;
+            let value = patch.1;
+            if pos < offset {
+                scene.extend_from_slice(&data.drawdata_stream[pos..offset]);
+            }
+            scene.extend_from_slice(bytemuck::bytes_of(&value));
+            pos = offset + 4;
+        }
+        if pos < data.drawdata_stream.len() {
+            scene.extend_from_slice(&data.drawdata_stream[pos..])
+        }
+    } else {
         scene.extend(&data.drawdata_stream);
+    }
     let transform_base = size_to_words(scene.len());
     scene.extend(bytemuck::cast_slice(&data.transform_stream));
     let linewidth_base = size_to_words(scene.len());
     scene.extend(bytemuck::cast_slice(&data.linewidth_stream));
-
     let n_path = data.n_path;
     // TODO: calculate for real when we do rectangles
     let n_drawobj = n_path;
@@ -159,11 +197,11 @@ pub fn render_full(scene: &Scene, shaders: &FullShaders) -> (Recording, BufProxy
         linewidth_base,
     };
     println!("{:?}", config);
-    let scene_buf = recording.upload(scene);
-    let config_buf = recording.upload(bytemuck::bytes_of(&config).to_owned());
+    let scene_buf = ResourceProxy::Buf(recording.upload(scene));
+    let config_buf = ResourceProxy::Buf(recording.upload(bytemuck::bytes_of(&config).to_owned()));
 
     let pathtag_wgs = pathtag_padded / (4 * shaders::PATHTAG_REDUCE_WG as usize);
-    let reduced_buf = BufProxy::new(pathtag_wgs as u64 * TAG_MONOID_FULL_SIZE);
+    let reduced_buf = ResourceProxy::new_buf(pathtag_wgs as u64 * TAG_MONOID_FULL_SIZE);
     // TODO: really only need pathtag_wgs - 1
     recording.dispatch(
         shaders.pathtag_reduce,
@@ -171,22 +209,24 @@ pub fn render_full(scene: &Scene, shaders: &FullShaders) -> (Recording, BufProxy
         [config_buf, scene_buf, reduced_buf],
     );
 
-    let tagmonoid_buf =
-        BufProxy::new(pathtag_wgs as u64 * shaders::PATHTAG_REDUCE_WG as u64 * TAG_MONOID_FULL_SIZE);
+    let tagmonoid_buf = ResourceProxy::new_buf(
+        pathtag_wgs as u64 * shaders::PATHTAG_REDUCE_WG as u64 * TAG_MONOID_FULL_SIZE,
+    );
     recording.dispatch(
         shaders.pathtag_scan,
         (pathtag_wgs as u32, 1, 1),
         [config_buf, scene_buf, reduced_buf, tagmonoid_buf],
     );
     let drawobj_wgs = (n_drawobj + shaders::PATH_BBOX_WG - 1) / shaders::PATH_BBOX_WG;
-    let path_bbox_buf = BufProxy::new(n_path as u64 * PATH_BBOX_SIZE);
+    let path_bbox_buf = ResourceProxy::new_buf(n_path as u64 * PATH_BBOX_SIZE);
     recording.dispatch(
         shaders.bbox_clear,
         (drawobj_wgs, 1, 1),
         [config_buf, path_bbox_buf],
     );
-    let cubic_buf = BufProxy::new(n_pathtag as u64 * CUBIC_SIZE);
-    let path_coarse_wgs = (n_pathtag as u32 + shaders::PATH_COARSE_WG - 1) / shaders::PATH_COARSE_WG;
+    let cubic_buf = ResourceProxy::new_buf(n_pathtag as u64 * CUBIC_SIZE);
+    let path_coarse_wgs =
+        (n_pathtag as u32 + shaders::PATH_COARSE_WG - 1) / shaders::PATH_COARSE_WG;
     recording.dispatch(
         shaders.pathseg,
         (path_coarse_wgs, 1, 1),
@@ -198,14 +238,14 @@ pub fn render_full(scene: &Scene, shaders: &FullShaders) -> (Recording, BufProxy
             cubic_buf,
         ],
     );
-    let draw_reduced_buf = BufProxy::new(drawobj_wgs as u64 * DRAWMONOID_SIZE);
+    let draw_reduced_buf = ResourceProxy::new_buf(drawobj_wgs as u64 * DRAWMONOID_SIZE);
     recording.dispatch(
         shaders.draw_reduce,
         (drawobj_wgs, 1, 1),
         [config_buf, scene_buf, draw_reduced_buf],
     );
-    let draw_monoid_buf = BufProxy::new(n_drawobj as u64 * DRAWMONOID_SIZE);
-    let info_buf = BufProxy::new(n_drawobj as u64 * MAX_DRAWINFO_SIZE);
+    let draw_monoid_buf = ResourceProxy::new_buf(n_drawobj as u64 * DRAWMONOID_SIZE);
+    let info_buf = ResourceProxy::new_buf(n_drawobj as u64 * MAX_DRAWINFO_SIZE);
     recording.dispatch(
         shaders.draw_leaf,
         (drawobj_wgs, 1, 1),
@@ -218,16 +258,17 @@ pub fn render_full(scene: &Scene, shaders: &FullShaders) -> (Recording, BufProxy
             info_buf,
         ],
     );
-    let draw_bbox_buf = BufProxy::new(n_path as u64 * DRAW_BBOX_SIZE);
+    let draw_bbox_buf = ResourceProxy::new_buf(n_path as u64 * DRAW_BBOX_SIZE);
     let bump_buf = BufProxy::new(BUMP_SIZE);
     // Not actually used yet.
-    let clip_bbox_buf = BufProxy::new(1024);
-    let bin_data_buf = BufProxy::new(1 << 20);
+    let clip_bbox_buf = ResourceProxy::new_buf(1024);
+    let bin_data_buf = ResourceProxy::new_buf(1 << 20);
     let width_in_bins = (config.width_in_tiles + 15) / 16;
     let height_in_bins = (config.height_in_tiles + 15) / 16;
     let n_bins = width_in_bins * height_in_bins;
-    let bin_header_buf = BufProxy::new((n_bins * drawobj_wgs) as u64 * BIN_HEADER_SIZE);
+    let bin_header_buf = ResourceProxy::new_buf((n_bins * drawobj_wgs) as u64 * BIN_HEADER_SIZE);
     recording.clear_all(bump_buf);
+    let bump_buf = ResourceProxy::Buf(bump_buf);
     recording.dispatch(
         shaders.binning,
         (drawobj_wgs, 1, 1),
@@ -242,8 +283,8 @@ pub fn render_full(scene: &Scene, shaders: &FullShaders) -> (Recording, BufProxy
             bin_header_buf,
         ],
     );
-    let path_buf = BufProxy::new(n_path as u64 * PATH_SIZE);
-    let tile_buf = BufProxy::new(1 << 20);
+    let path_buf = ResourceProxy::new_buf(n_path as u64 * PATH_SIZE);
+    let tile_buf = ResourceProxy::new_buf(1 << 20);
     let path_wgs = (n_path + shaders::PATH_BBOX_WG - 1) / shaders::PATH_BBOX_WG;
     recording.dispatch(
         shaders.tile_alloc,
@@ -258,7 +299,7 @@ pub fn render_full(scene: &Scene, shaders: &FullShaders) -> (Recording, BufProxy
         ],
     );
 
-    let segments_buf = BufProxy::new(1 << 24);
+    let segments_buf = ResourceProxy::new_buf(1 << 24);
     recording.dispatch(
         shaders.path_coarse,
         (path_coarse_wgs, 1, 1),
@@ -278,7 +319,7 @@ pub fn render_full(scene: &Scene, shaders: &FullShaders) -> (Recording, BufProxy
         (path_wgs, 1, 1),
         [config_buf, path_buf, tile_buf],
     );
-    let ptcl_buf = BufProxy::new(1 << 24);
+    let ptcl_buf = ResourceProxy::new_buf(1 << 24);
     recording.dispatch(
         shaders.coarse,
         (width_in_bins, height_in_bins, 1),
@@ -300,7 +341,14 @@ pub fn render_full(scene: &Scene, shaders: &FullShaders) -> (Recording, BufProxy
     recording.dispatch(
         shaders.fine,
         (config.width_in_tiles, config.height_in_tiles, 1),
-        [config_buf, tile_buf, segments_buf, out_buf, ptcl_buf],
+        [
+            config_buf,
+            tile_buf,
+            segments_buf,
+            ResourceProxy::Buf(out_buf),
+            ptcl_buf,
+            gradient_image,
+        ],
     );
 
     let download_buf = out_buf;

piet-wgsl/src/shaders.rs

@@ -255,6 +255,7 @@ pub fn full_shaders(device: &Device, engine: &mut Engine) -> Result<FullShaders,
             BindType::BufReadOnly,
             BindType::Buffer,
             BindType::BufReadOnly,
+            BindType::ImageRead,
         ],
     )?;
     Ok(FullShaders {

piet-wgsl/src/test_scene.rs

@@ -15,14 +15,16 @@
 // Also licensed under MIT license, at your choice.
 
 use kurbo::BezPath;
-use piet_scene::{Affine, Brush, Color, Fill, PathElement, Point, Scene, SceneBuilder, Stroke};
+use piet_scene::{Affine, Brush, Color, Fill, LinearGradient, PathElement, Point, Scene, SceneBuilder, Stroke, GradientStop};
 
 use crate::pico_svg::PicoSvg;
 
 pub fn gen_test_scene() -> Scene {
     let mut scene = Scene::default();
     let mut builder = SceneBuilder::for_scene(&mut scene);
-    if false {
+    let scene_ix = 1;
+    match scene_ix {
+        0 => {
             let path = [
                 PathElement::MoveTo(Point::new(100.0, 100.0)),
                 PathElement::LineTo(Point::new(500.0, 120.0)),
@@ -40,11 +42,44 @@ pub fn gen_test_scene() -> Scene {
             let style = simple_stroke(1.0);
             let brush = Brush::Solid(Color::rgb8(0xa0, 0x00, 0x00));
             builder.stroke(&style, transform, &brush, None, &path);
-    } else {
-        let xml_str = std::str::from_utf8(include_bytes!("../../piet-gpu/Ghostscript_Tiger.svg")).unwrap();
+        }
+        1 => {
+            let path = [
+                PathElement::MoveTo(Point::new(100.0, 100.0)),
+                PathElement::LineTo(Point::new(300.0, 100.0)),
+                PathElement::LineTo(Point::new(300.0, 300.0)),
+                PathElement::LineTo(Point::new(100.0, 300.0)),
+                PathElement::Close,
+            ];
+            let gradient = Brush::LinearGradient(LinearGradient {
+                start: Point::new(100.0, 100.0),
+                end: Point::new(300.0, 300.0),
+                extend: piet_scene::ExtendMode::Pad,
+                stops: vec![
+                    GradientStop {
+                        offset: 0.0,
+                        color: Color::rgb8(255, 0, 0),
+                    },
+                    GradientStop {
+                        offset: 0.5,
+                        color: Color::rgb8(0, 255, 0),
+                    },
+                    GradientStop {
+                        offset: 1.0,
+                        color: Color::rgb8(0, 0, 255),
+                    },
+                ].into()
+            });
+            builder.fill(Fill::NonZero, Affine::scale(3.0, 3.0), &gradient, None, &path);
+        }
+        _ => {
+            let xml_str =
+                std::str::from_utf8(include_bytes!("../../piet-gpu/Ghostscript_Tiger.svg"))
+                    .unwrap();
             let svg = PicoSvg::load(xml_str, 6.0).unwrap();
             render_svg(&mut builder, &svg, false);
         }
+    }
     scene
 }
 
@@ -94,7 +129,6 @@ fn convert_bez_path<'a>(path: &'a BezPath) -> impl Iterator<Item = PathElement>
         .map(|el| PathElement::from_kurbo(*el))
 }
 
-
 fn simple_stroke(width: f32) -> Stroke<[f32; 0]> {
     Stroke {
         width,