valo 0.2.2

A 2D render engine on wgpu: display lists, planned GPU passes, and a Skia-shaped text stack
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263

//! One device driving many canvases.
//!
//! This is the claim valo's landing page rests on: a traditional 2D context
//! carries its own GPU device and browsers cap those around 16, so a dozen
//! live demos is near the ceiling before anything is drawn. Attaching them all
//! to one [`Context`] shares the glyph atlas, the image cache and the
//! render-target pool.
//!
//! Every assertion here is about RESOURCES rather than pixels, deliberately.
//! A test that only checked what was drawn would pass just as happily on
//! twelve separate devices, which is the arrangement this exists to rule out.

use std::sync::Arc;

use valo::{
    Color, Context, DisplayList, DisplayListBuilder, FontCollection, ImageDesc, Paint,
    ParagraphBuilder, PersistentCanvas, Rect, TextStyle,
};

const FORMAT: wgpu::TextureFormat = wgpu::TextureFormat::Rgba8Unorm;
const CANVASES: usize = 12;
const SIZE: [u32; 2] = [256, 256];

fn fonts() -> FontCollection {
    let dir = concat!(env!("CARGO_MANIFEST_DIR"), "/../../assets/fonts");
    let mut collection = FontCollection::new();
    collection
        .register(
            "Fira Sans",
            std::fs::read(format!("{dir}/fira_sans.ttf")).unwrap(),
        )
        .unwrap();
    collection
}

/// One card's worth of work: some text, an image, and a blurred shape. Text
/// fills the glyph atlas, the image fills the image cache, the blur takes a
/// pooled filter target — one per shared subsystem.
fn card(collection: &mut FontCollection, image: &valo::Image, seed: usize) -> Arc<DisplayList> {
    let mut builder = DisplayListBuilder::new();
    builder.draw_rect(
        Rect::new(0.0, 0.0, 256.0, 256.0),
        &Paint::from_color(Color::rgb(0.1, 0.1, 0.14)),
    );
    builder.draw_image_rect(
        image,
        Rect::new(0.0, 0.0, image.width(), image.height()),
        Rect::new(16.0, 16.0, 64.0, 64.0),
        valo::Sampling::default(),
        &Paint::from_color(Color::WHITE),
    );
    builder.draw_rect(
        Rect::new(120.0, 40.0, 90.0, 90.0),
        &Paint {
            color: Color::rgb(0.9, 0.4, 0.2),
            mask_blur: Some(valo::MaskBlur::new(6.0)),
            ..Paint::default()
        },
    );

    // The SAME text on every card: the atlas is keyed by glyph, so twelve
    // cards drawing it must not raster it twelve times.
    let mut paragraph = ParagraphBuilder::new(collection);
    let mut style = TextStyle::new("Fira Sans", 24.0, Color::WHITE);
    style.families = vec!["Fira Sans".to_owned()];
    paragraph.add_text("Shared device", &style);
    let mut paragraph = paragraph.build();
    paragraph.layout(240.0);
    use valo::DrawParagraphExt;
    builder.draw_paragraph(&paragraph, (16.0, 150.0 + (seed % 3) as f32));

    Arc::new(builder.build())
}

fn swatch(context: &mut Context) -> valo::Image {
    context.upload_image(
        ImageDesc {
            size: [32, 32],
            premultiplied: true,
            mips: false,
        },
        &vec![200u8; 32 * 32 * 4],
    )
}

/// Render `count` cards on ONE context, and report what that device holds.
fn shared(device: &wgpu::Device, queue: &wgpu::Queue, count: usize) -> (u64, u64, u64, u32) {
    let mut context = Context::new(device.clone(), queue.clone());
    let mut collection = fonts();
    let image = swatch(&mut context);
    let mut canvases: Vec<PersistentCanvas> = (0..count)
        .map(|_| PersistentCanvas::new(&mut context, SIZE, FORMAT))
        .collect();
    for (seed, canvas) in canvases.iter_mut().enumerate() {
        let list = card(&mut collection, &image, seed);
        canvas.draw(&mut context, &list, Some(Color::TRANSPARENT));
    }
    let report = context.memory_report();
    (
        report.total_bytes(),
        report.atlas.iter().map(|family| family.bytes).sum(),
        report.images.bytes,
        report.targets.count,
    )
}

/// The same work with a context — a device — PER canvas, which is what a
/// per-canvas 2D context costs.
fn per_canvas(device: &wgpu::Device, queue: &wgpu::Queue, count: usize) -> (u64, u64, u64, u32) {
    let mut totals = (0u64, 0u64, 0u64, 0u32);
    for seed in 0..count {
        let mut context = Context::new(device.clone(), queue.clone());
        let mut collection = fonts();
        let image = swatch(&mut context);
        let mut canvas = PersistentCanvas::new(&mut context, SIZE, FORMAT);
        let list = card(&mut collection, &image, seed);
        canvas.draw(&mut context, &list, Some(Color::TRANSPARENT));
        let report = context.memory_report();
        totals.0 += report.total_bytes();
        totals.1 += report.atlas.iter().map(|family| family.bytes).sum::<u64>();
        totals.2 += report.images.bytes;
        totals.3 += report.targets.count;
    }
    totals
}

/// The headline: twelve canvases on one device hold ONE atlas, ONE image
/// cache and ONE target pool, where twelve devices would hold twelve of each.
#[test]
fn one_device_holds_one_copy_of_what_twelve_would_duplicate() {
    let Some((device, queue)) = valo_harness::headless_device() else {
        eprintln!("SKIP one_device_holds_one_copy_of_what_twelve_would_duplicate");
        return;
    };

    let (one_total, one_atlas, one_images, _) = shared(&device, &queue, 1);
    let (many_total, many_atlas, many_images, many_targets) = shared(&device, &queue, CANVASES);
    let (split_total, split_atlas, split_images, split_targets) =
        per_canvas(&device, &queue, CANVASES);

    // The per-canvas backing pair is NOT in valo's own accounting (it wraps
    // host textures), and it costs the same either way — so it is added to
    // both sides rather than quietly left out of the flattering one.
    let backing_pair = 2 * u64::from(SIZE[0]) * u64::from(SIZE[1]) * 4;
    let per_canvas_cost = backing_pair * CANVASES as u64;
    let mib = |bytes: u64| bytes as f64 / (1024.0 * 1024.0);
    println!(
        "\n{CANVASES} canvases, {}x{} each\n\
         \x20 shared device : {:.2} MiB device-level + {:.2} MiB backings = {:.2} MiB\n\
         \x20 device each   : {:.2} MiB device-level + {:.2} MiB backings = {:.2} MiB\n\
         \x20 atlas         : {:.2} MiB shared vs {:.2} MiB split\n\
         \x20 pooled targets: {many_targets} shared vs {split_targets} split\n\
         \x20 one canvas    : {:.2} MiB device-level\n",
        SIZE[0],
        SIZE[1],
        mib(many_total),
        mib(per_canvas_cost),
        mib(many_total + per_canvas_cost),
        mib(split_total),
        mib(per_canvas_cost),
        mib(split_total + per_canvas_cost),
        mib(many_atlas),
        mib(split_atlas),
        mib(one_total),
    );

    // The atlas is the claim that matters: it is per-device and keyed by
    // glyph, so the same text on twelve canvases rasters once.
    assert_eq!(
        many_atlas, one_atlas,
        "twelve canvases must share one glyph atlas"
    );
    assert_eq!(
        split_atlas,
        one_atlas * CANVASES as u64,
        "twelve devices each pay for their own atlas — the baseline being beaten"
    );

    // Same for uploaded images.
    assert_eq!(many_images, one_images, "one image cache, not twelve");
    assert_eq!(split_images, one_images * CANVASES as u64);

    // Pooled targets are reused across canvases rather than accumulating.
    assert!(
        many_targets <= split_targets,
        "a shared pool must not hold more targets than separate ones: \
         {many_targets} vs {split_targets}"
    );

    assert!(
        many_total < split_total,
        "sharing must actually save: {many_total} B shared vs {split_total} B split"
    );
}

/// Per-canvas state stays per-canvas: each keeps its own pixels, and drawing
/// into one leaves the others alone.
#[test]
fn canvases_on_one_device_keep_their_own_pixels() {
    let Some((device, queue)) = valo_harness::headless_device() else {
        eprintln!("SKIP canvases_on_one_device_keep_their_own_pixels");
        return;
    };
    let mut context = Context::new(device.clone(), queue.clone());
    let colors = [
        Color::rgb(1.0, 0.0, 0.0),
        Color::rgb(0.0, 1.0, 0.0),
        Color::rgb(0.0, 0.0, 1.0),
    ];
    let mut canvases: Vec<PersistentCanvas> = colors
        .iter()
        .map(|_| PersistentCanvas::new(&mut context, [32, 32], FORMAT))
        .collect();

    for (canvas, color) in canvases.iter_mut().zip(colors) {
        let mut builder = DisplayListBuilder::new();
        builder.draw_rect(Rect::new(0.0, 0.0, 32.0, 32.0), &Paint::from_color(color));
        canvas.draw(&mut context, &Arc::new(builder.build()), None);
    }

    // Draw again into only the first: the others must not move.
    let mut builder = DisplayListBuilder::new();
    builder.draw_rect(
        Rect::new(0.0, 0.0, 32.0, 32.0),
        &Paint::from_color(Color::rgb(1.0, 1.0, 1.0)),
    );
    canvases[0].draw(&mut context, &Arc::new(builder.build()), None);

    let expected = [[255, 255, 255], [0, 255, 0], [0, 0, 255]];
    for (canvas, want) in canvases.iter().zip(expected) {
        let pixels =
            valo_harness::read_texture_rgba(&device, &queue, canvas.front().texture(), [32, 32]);
        assert_eq!(&pixels[..3], &want, "each canvas keeps its own pixels");
    }
}

/// Dropping one canvas must not disturb the device or its siblings — a card
/// scrolled out of the DOM should free its backing and nothing else.
#[test]
fn dropping_one_canvas_leaves_the_others_drawing() {
    let Some((device, queue)) = valo_harness::headless_device() else {
        eprintln!("SKIP dropping_one_canvas_leaves_the_others_drawing");
        return;
    };
    let mut context = Context::new(device.clone(), queue.clone());
    let mut keep = PersistentCanvas::new(&mut context, [32, 32], FORMAT);
    let discard = PersistentCanvas::new(&mut context, [32, 32], FORMAT);

    let mut builder = DisplayListBuilder::new();
    builder.draw_rect(
        Rect::new(0.0, 0.0, 32.0, 32.0),
        &Paint::from_color(Color::rgb(0.0, 1.0, 0.0)),
    );
    let green = Arc::new(builder.build());
    keep.draw(&mut context, &green, None);
    drop(discard);

    // The survivor still renders, and its earlier pixels are still there.
    let empty = Arc::new(DisplayListBuilder::new().build());
    keep.draw(&mut context, &empty, None);
    let pixels = valo_harness::read_texture_rgba(&device, &queue, keep.front().texture(), [32, 32]);
    assert_eq!(&pixels[..3], &[0, 255, 0]);
}