roxlap-gpu 0.8.0

GPU compute-shader renderer for the roxlap voxel engine (WGPU + WGSL DDA marcher). Sibling to roxlap-core's CPU opticast.
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
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
//! GPU.11.0 gate — headless scene-DDA render.
//!
//! Stands up a real GPU device, uploads a one-grid scene whose every
//! column has a textured floor voxel, and renders it through the
//! `scene_dda.wgsl` compute pipeline that now carries the full mip
//! ladder per slot (GPU.11.0). The shader still marches mip-0, so a
//! correct render proves:
//!   1. `scene_dda.wgsl` compiles with the grown `GridStaticMeta`.
//!   2. The 112-byte std430 struct layout matches the Rust upload.
//!   3. The new per-slot occupancy / color_offsets *strides* still
//!      address mip-0 byte-identically (a floor voxel reads its
//!      colour back through the strided layout).
//!
//! Skips silently if no Vulkan/Metal/DX12 adapter is reachable.

#![allow(
    clippy::cast_precision_loss,
    clippy::cast_possible_truncation,
    clippy::doc_markdown,
    clippy::many_single_char_names,
    clippy::redundant_closure_for_method_calls
)]

use std::sync::Mutex;

use roxlap_formats::vxl::Vxl;
use roxlap_gpu::{
    decompress_chunk, Camera, GpuInitError, GpuRendererSettings, GpuSceneResident, GridUpload,
    HeadlessGpu, HeadlessSceneRenderer, SceneUpload,
};

static GPU_TEST_LOCK: Mutex<()> = Mutex::new(());

fn try_init() -> Option<(HeadlessGpu, std::sync::MutexGuard<'static, ()>)> {
    let guard = GPU_TEST_LOCK.lock().unwrap_or_else(|e| e.into_inner());
    match HeadlessGpu::new_blocking(GpuRendererSettings::default()) {
        Ok(gpu) => Some((gpu, guard)),
        Err(GpuInitError::NoAdapter) => {
            eprintln!("[skip] no GPU adapter reachable");
            None
        }
        Err(e) => {
            eprintln!("[skip] GPU init failed ({e})");
            None
        }
    }
}

/// `vsid × vsid` chunk: one textured floor voxel per column at
/// `z = 100`, colour `0x80ff_8000` (A=0x80 → brightness 1.0,
/// R=0xff, G=0x80, B=0x00). decompress_chunk builds its mip ladder.
fn floor_chunk(vsid: u32) -> Vxl {
    let n_cols = (vsid as usize) * (vsid as usize);
    let mut data: Vec<u8> = Vec::with_capacity(n_cols * 8);
    let mut column_offset: Vec<u32> = Vec::with_capacity(n_cols + 1);
    let bgra = [0x00u8, 0x80, 0xff, 0x80];
    for _ in 0..n_cols {
        column_offset.push(u32::try_from(data.len()).expect("offset fits"));
        data.extend_from_slice(&[0, 100, 100, 0]); // nextptr=0, z1=100, z1c=100, z0=0
        data.extend_from_slice(&bgra);
    }
    column_offset.push(u32::try_from(data.len()).expect("offset fits"));
    Vxl {
        vsid,
        ipo: [0.0; 3],
        ist: [1.0, 0.0, 0.0],
        ihe: [0.0, 0.0, 1.0],
        ifo: [0.0, 1.0, 0.0],
        data: data.into_boxed_slice(),
        column_offset: column_offset.into_boxed_slice(),
        mip_base_offsets: Box::new([0, n_cols + 1]),
        vbit: Box::new([]),
        vbiti: 0,
    }
}

/// `vsid × vsid` chunk: every column solid over `z ∈ [top, bot]`
/// (a wall/block facing a horizontal ray), colour `0x80ff_8000`.
fn block_chunk(vsid: u32, top: u8, bot: u8) -> Vxl {
    let n_cols = (vsid as usize) * (vsid as usize);
    let n_vox = (bot - top + 1) as usize;
    let mut data: Vec<u8> = Vec::with_capacity(n_cols * (4 + n_vox * 4));
    let mut column_offset: Vec<u32> = Vec::with_capacity(n_cols + 1);
    let bgra = [0x00u8, 0x80, 0xff, 0x80];
    for _ in 0..n_cols {
        column_offset.push(u32::try_from(data.len()).expect("offset fits"));
        data.extend_from_slice(&[0, top, bot, 0]); // nextptr=0, z1=top, z1c=bot, z0=0
        for _ in 0..n_vox {
            data.extend_from_slice(&bgra);
        }
    }
    column_offset.push(u32::try_from(data.len()).expect("offset fits"));
    Vxl {
        vsid,
        ipo: [0.0; 3],
        ist: [1.0, 0.0, 0.0],
        ihe: [0.0, 0.0, 1.0],
        ifo: [0.0, 1.0, 0.0],
        data: data.into_boxed_slice(),
        column_offset: column_offset.into_boxed_slice(),
        mip_base_offsets: Box::new([0, n_cols + 1]),
        vbit: Box::new([]),
        vbiti: 0,
    }
}

/// Recognisably the orange block (`0x80ff_8000` at brightness 1.0 →
/// ~(255,128,0)), not the bluish sky (~(120,150,220)). Loose because
/// coarse mips average the (uniform) block colour.
fn is_block_color(p: u32) -> bool {
    let (r, g, b) = (p & 0xff, (p >> 8) & 0xff, (p >> 16) & 0xff);
    r > 180 && (80..=175).contains(&g) && b < 70
}

/// `vsid × vsid` chunk: one textured floor voxel per column at `z =
/// surf`, with implicit voxlap **bedrock** solid below it to z=255.
/// Models a cliff/wall: only the top is coloured; the face below is
/// bedrock. Pre-fix the GPU treated bedrock as air → the face showed
/// sky. Slab `[nextptr=0, z1=surf, z1c=surf, z0=0]` + 1 colour.
fn wall_chunk(vsid: u32, surf: u8) -> Vxl {
    let n_cols = (vsid as usize) * (vsid as usize);
    let mut data: Vec<u8> = Vec::with_capacity(n_cols * 8);
    let mut column_offset: Vec<u32> = Vec::with_capacity(n_cols + 1);
    let bgra = [0x00u8, 0x80, 0xff, 0x80];
    for _ in 0..n_cols {
        column_offset.push(u32::try_from(data.len()).expect("offset fits"));
        data.extend_from_slice(&[0, surf, surf, 0]);
        data.extend_from_slice(&bgra);
    }
    column_offset.push(u32::try_from(data.len()).expect("offset fits"));
    Vxl {
        vsid,
        ipo: [0.0; 3],
        ist: [1.0, 0.0, 0.0],
        ihe: [0.0, 0.0, 1.0],
        ifo: [0.0, 1.0, 0.0],
        data: data.into_boxed_slice(),
        column_offset: column_offset.into_boxed_slice(),
        mip_base_offsets: Box::new([0, n_cols + 1]),
        vbit: Box::new([]),
        vbiti: 0,
    }
}

#[test]
fn scene_dda_marches_coarse_mip_for_distant_chunk() {
    let Some((gpu, _lock)) = try_init() else {
        return;
    };
    eprintln!("mip_render: adapter = {}", gpu.adapter_info);

    // One solid block chunk placed FAR along +y (chunk index 4) so a
    // horizontal ray enters it at t ≈ 128 — past several octaves of
    // mip_scan_dist, forcing a deep mip. The camera sits in the empty
    // chunk (0,0,0).
    let vsid = 32u32;
    let chunk = decompress_chunk(&block_chunk(vsid, 0, 31));
    assert!(chunk.mips.len() >= 5, "need a deep ladder for mip-4");

    let grid = GridUpload {
        vsid,
        origin_chunk: [0, 0, 0],
        chunks_dims: [1, 8, 1],
        pool_dims: [1, 8, 1],
        chunks: vec![([0, 4, 0], chunk)],
    };
    let scene = GpuSceneResident::upload(&gpu.device, &SceneUpload { grids: vec![grid] });

    let (w, h) = (64u32, 64u32);
    let renderer = HeadlessSceneRenderer::new(&gpu.device, &gpu.queue, w, h);
    // Camera in the empty near chunk, looking +y at the block; z=16
    // lands inside the block's z=0..31 band. right × down == forward.
    let cam = Camera {
        position: [vsid as f32 * 0.5, 0.0, 16.0],
        right: [1.0, 0.0, 0.0],
        down: [0.0, 0.0, 1.0],
        forward: [0.0, 1.0, 0.0],
        fov_y_rad: 30f32.to_radians(),
    };
    let centre = (h / 2 * w + w / 2) as usize;

    // mip-0 baseline (LOD off): the block renders.
    let fb0 = renderer.render(
        &gpu.device,
        &gpu.queue,
        &scene,
        &[cam],
        cam.fov_y_rad,
        64,
        0.0,
    );
    assert!(
        is_block_color(fb0[centre]),
        "mip-0 centre should be the block, got {:#08x}",
        fb0[centre],
    );

    // Force a deep mip: msd=8 at t≈128 → mip-4. If mip-N occupancy /
    // colour addressing were wrong the block would vanish (sky) or
    // render a garbage colour.
    let fb4 = renderer.render(
        &gpu.device,
        &gpu.queue,
        &scene,
        &[cam],
        cam.fov_y_rad,
        64,
        8.0,
    );
    eprintln!(
        "mip_render: centre mip0={:#08x} mip4={:#08x}",
        fb0[centre], fb4[centre]
    );
    assert!(
        is_block_color(fb4[centre]),
        "coarse-mip centre should still be the block, got {:#08x}",
        fb4[centre],
    );

    // The coarse render should broadly agree with mip-0 (same block
    // fills the view) — most pixels classify the same way.
    let agree = fb0
        .iter()
        .zip(&fb4)
        .filter(|(a, b)| is_block_color(**a) == is_block_color(**b))
        .count();
    let frac = agree as f32 / fb0.len() as f32;
    eprintln!("mip_render: block/sky agreement = {frac:.3}");
    assert!(
        frac > 0.9,
        "mip-0 vs mip-4 block coverage diverged: {frac:.3}"
    );
}

#[test]
fn scene_dda_aabb_early_out_away_is_sky() {
    // GPU.13.0 — the occupied chunk-AABB early-out must terminate a
    // ray the moment it has left the box along its travel direction.
    // One block chunk at +y (chunk 4); the camera sits in the near
    // chunk (0,0,0) but looks the OTHER way (−y). Every ray starts
    // already past the AABB's near slab (p.y=0 < aabb_min.y=4 with
    // step.y<0) → instant early-out → pure sky, no block pixels.
    let Some((gpu, _lock)) = try_init() else {
        return;
    };
    let vsid = 32u32;
    let chunk = decompress_chunk(&block_chunk(vsid, 0, 31));
    let grid = GridUpload {
        vsid,
        origin_chunk: [0, 0, 0],
        chunks_dims: [1, 8, 1],
        pool_dims: [1, 8, 1],
        chunks: vec![([0, 4, 0], chunk)],
    };
    let scene = GpuSceneResident::upload(&gpu.device, &SceneUpload { grids: vec![grid] });
    // Sanity: the upload computed the occupied AABB at chunk y=4.
    assert_eq!(scene.static_meta[0].aabb_min, [0, 4, 0]);
    assert_eq!(scene.static_meta[0].aabb_max, [0, 4, 0]);

    let (w, h) = (64u32, 64u32);
    let renderer = HeadlessSceneRenderer::new(&gpu.device, &gpu.queue, w, h);
    let cam = Camera {
        position: [vsid as f32 * 0.5, 0.0, 16.0],
        right: [-1.0, 0.0, 0.0],
        down: [0.0, 0.0, 1.0],
        forward: [0.0, -1.0, 0.0], // look AWAY from the block
        fov_y_rad: 30f32.to_radians(),
    };
    let fb = renderer.render(
        &gpu.device,
        &gpu.queue,
        &scene,
        &[cam],
        cam.fov_y_rad,
        64,
        0.0,
    );
    let block_px = fb.iter().filter(|&&p| is_block_color(p)).count();
    assert_eq!(
        block_px, 0,
        "looking away from the only chunk must be all sky, got {block_px} block pixels",
    );
}

#[test]
fn scene_dda_zero_grids_renders_sky() {
    // Regression for the sprite-only / empty-scene GPU path: a scene
    // with ZERO grids must still render a valid frame — the scene pass
    // fills the flat sky everywhere (+ far depth), giving the sprite
    // pass a background to composite over. Pre-fix the render facade
    // short-circuited a grid-less scene to a bare clear and never ran
    // `render_scene`, so a sprite-only viewer (no voxel grids) showed
    // only the clear colour with the model invisible. This exercises
    // the engine half the facade fix newly relies on: grid_count == 0
    // with zero cameras renders the uniform sky without panicking.
    let Some((gpu, _lock)) = try_init() else {
        return;
    };
    let scene = GpuSceneResident::upload(&gpu.device, &SceneUpload { grids: vec![] });
    assert_eq!(
        scene.grid_count, 0,
        "empty SceneUpload → zero-grid resident"
    );

    let (w, h) = (32u32, 32u32);
    let renderer = HeadlessSceneRenderer::new(&gpu.device, &gpu.queue, w, h);
    // Zero cameras matches grid_count == 0 (render_scene asserts equal).
    // Pre-fix this path was never dispatched (the facade short-circuited
    // to a clear); the win here is it runs end-to-end without panic and
    // produces a clean, consistent background for the sprite pass.
    let fb = renderer.render(
        &gpu.device,
        &gpu.queue,
        &scene,
        &[],
        30f32.to_radians(),
        64,
        0.0,
    );
    assert_eq!(fb.len(), (w * h) as usize);

    // No grids → no per-pixel grid hits → every pixel is the flat sky
    // sample (headless default sky [120, 150, 220]). Uniform proves the
    // zero-grid path reads no garbage grid data; bluish (b > r) + not
    // black proves the sky direction is well-formed — the pre-fix
    // (0,0,1) default fed atan2(0,0) → NaN → a black sample.
    let first = fb[0];
    assert!(
        fb.iter().all(|&p| p == first),
        "zero-grid frame must be a uniform sky, got varied pixels (first={first:#08x})",
    );
    assert_ne!(
        first & 0x00ff_ffff,
        0,
        "sky must not be black, got {first:#08x}"
    );
    let (r, _g, b) = (first & 0xff, (first >> 8) & 0xff, (first >> 16) & 0xff);
    assert!(b > r, "headless sky is bluish (b>r), got {first:#08x}");
}

/// GPU side-shade (voxlap setsideshades) darkens a grid face. Camera
/// looks straight down at a floor plane; the floor is hit via a +z
/// step, so its shade comes from the `bot` lane. Rendering with
/// `bot = 64` must darken the floor ~half vs the unshaded baseline,
/// proving the scene-DDA face detection + brightness reduction work.
/// (Exact CPU parity needs visual inspection; this guards the
/// mechanism + uniform plumbing against regressions.)
#[test]
fn scene_dda_side_shades_darken_floor() {
    let Some((gpu, _lock)) = try_init() else {
        return;
    };
    let vsid = 32u32;
    let chunk = decompress_chunk(&floor_chunk(vsid)); // floor voxel at z=100
    let grid = GridUpload {
        vsid,
        origin_chunk: [0, 0, 0],
        chunks_dims: [1, 1, 1],
        pool_dims: [1, 1, 1],
        chunks: vec![([0, 0, 0], chunk)],
    };
    let scene = GpuSceneResident::upload(&gpu.device, &SceneUpload { grids: vec![grid] });

    let (w, h) = (64u32, 64u32);
    let mut renderer = HeadlessSceneRenderer::new(&gpu.device, &gpu.queue, w, h);
    // Above the floor (z=50 < 100), looking straight down (+z, voxlap
    // z-down). right × down == forward.
    let cam = Camera {
        position: [16.0, 16.0, 50.0],
        right: [1.0, 0.0, 0.0],
        down: [0.0, 1.0, 0.0],
        forward: [0.0, 0.0, 1.0],
        fov_y_rad: 60f32.to_radians(),
    };
    let centre = (h / 2 * w + w / 2) as usize;
    let lum = |p: u32| (p & 0xff) + ((p >> 8) & 0xff) + ((p >> 16) & 0xff);

    let fb0 = renderer.render(
        &gpu.device,
        &gpu.queue,
        &scene,
        &[cam],
        cam.fov_y_rad,
        64,
        0.0,
    );
    let base = fb0[centre];
    assert!(
        is_block_color(base),
        "centre should be the lit floor, got {base:#08x}",
    );

    // Darken the floor face (bot lane = side_shades[1]).
    renderer.set_side_shades([0, 64, 0, 0, 0, 0]);
    let fb1 = renderer.render(
        &gpu.device,
        &gpu.queue,
        &scene,
        &[cam],
        cam.fov_y_rad,
        64,
        0.0,
    );
    let shaded = fb1[centre];
    assert!(
        lum(shaded) < lum(base),
        "side-shade should darken the floor: {base:#08x} -> {shaded:#08x}",
    );
    // bot=64 of 128 → ~half brightness, still textured (not black sky).
    assert_ne!(
        shaded & 0x00ff_ffff,
        0,
        "half-shaded floor must not be black"
    );
}

#[test]
fn aabb_tracks_streaming_refresh_and_evict() {
    // GPU.13.0 — the early-out box is maintained live: installing a
    // chunk at a new index must GROW the AABB (so the shader never
    // skips streamed-in terrain), and evicting it must SHRINK it back.
    let Some((gpu, _lock)) = try_init() else {
        return;
    };
    let vsid = 32u32;
    let grid = GridUpload {
        vsid,
        origin_chunk: [0, 0, 0],
        chunks_dims: [1, 1, 1],
        pool_dims: [8, 8, 1], // room for streamed indices
        chunks: vec![([0, 0, 0], decompress_chunk(&block_chunk(vsid, 0, 31)))],
    };
    let mut scene = GpuSceneResident::upload(&gpu.device, &SceneUpload { grids: vec![grid] });
    assert_eq!(scene.static_meta[0].aabb_min, [0, 0, 0]);
    assert_eq!(scene.static_meta[0].aabb_max, [0, 0, 0]);

    // Stream in a far chunk at (3, 2, 0) → AABB grows to cover it.
    let far = decompress_chunk(&block_chunk(vsid, 0, 31));
    scene.refresh_chunk(&gpu.queue, 0, [3, 2, 0], &far);
    assert_eq!(scene.static_meta[0].aabb_min, [0, 0, 0]);
    assert_eq!(scene.static_meta[0].aabb_max, [3, 2, 0]);

    // Evict it → AABB shrinks back to the lone origin chunk.
    scene.evict_chunk(&gpu.queue, 0, [3, 2, 0]);
    assert_eq!(scene.static_meta[0].aabb_min, [0, 0, 0]);
    assert_eq!(scene.static_meta[0].aabb_max, [0, 0, 0]);
}

#[test]
fn scene_dda_renders_bedrock_wall_face_solid() {
    let Some((gpu, _lock)) = try_init() else {
        return;
    };
    eprintln!("wall_render: adapter = {}", gpu.adapter_info);

    // A wall chunk: textured top at z=40, bedrock 41..255 below. Place
    // it far along +y; the camera looks at its face from BELOW the
    // textured top (z=128, deep in the bedrock region) — exactly the
    // cliff-face view that pre-fix showed sky through.
    let vsid = 32u32;
    let chunk = decompress_chunk(&wall_chunk(vsid, 40));
    let grid = GridUpload {
        vsid,
        origin_chunk: [0, 0, 0],
        chunks_dims: [1, 8, 1],
        pool_dims: [1, 8, 1],
        chunks: vec![([0, 4, 0], chunk)],
    };
    let scene = GpuSceneResident::upload(&gpu.device, &SceneUpload { grids: vec![grid] });

    let (w, h) = (64u32, 64u32);
    let renderer = HeadlessSceneRenderer::new(&gpu.device, &gpu.queue, w, h);
    let cam = Camera {
        position: [vsid as f32 * 0.5, 0.0, 128.0], // z=128 = bedrock region
        right: [1.0, 0.0, 0.0],
        down: [0.0, 0.0, 1.0],
        forward: [0.0, 1.0, 0.0],
        fov_y_rad: 30f32.to_radians(),
    };
    let fb = renderer.render(
        &gpu.device,
        &gpu.queue,
        &scene,
        &[cam],
        cam.fov_y_rad,
        64,
        0.0,
    );
    let centre = fb[(h / 2 * w + w / 2) as usize];
    eprintln!("wall_render: centre pixel = {centre:#08x}");
    // The bedrock face must be SOLID and inherit the surface colour
    // (was sky before the bedrock-as-solid fix).
    assert!(
        is_block_color(centre),
        "bedrock wall face should be solid surface colour, got {centre:#08x} (sky = regression)",
    );
}

#[test]
fn scene_dda_renders_floor_through_mip_layout() {
    let Some((gpu, _lock)) = try_init() else {
        return;
    };
    eprintln!("scene_render: adapter = {}", gpu.adapter_info);

    let vsid = 64u32;
    let chunk = decompress_chunk(&floor_chunk(vsid));
    // Sanity: the mip ladder was built (GPU.11.0 plumbing).
    assert!(chunk.mips.len() >= 2, "expected a mip ladder");

    let grid = GridUpload {
        vsid,
        origin_chunk: [0, 0, 0],
        chunks_dims: [1, 1, 1],
        pool_dims: [1, 1, 1],
        chunks: vec![([0, 0, 0], chunk)],
    };
    let scene = GpuSceneResident::upload(&gpu.device, &SceneUpload { grids: vec![grid] });
    eprintln!("scene_render: resident {} bytes", scene.resident_bytes());

    let (w, h) = (64u32, 64u32);
    let renderer = HeadlessSceneRenderer::new(&gpu.device, &gpu.queue, w, h);

    // Camera at the chunk's XY centre, above the floor (small z),
    // looking straight down (+z). right × down == forward (RH).
    let cam = Camera {
        position: [vsid as f32 * 0.5, vsid as f32 * 0.5, 20.0],
        right: [1.0, 0.0, 0.0],
        down: [0.0, 1.0, 0.0],
        forward: [0.0, 0.0, 1.0],
        fov_y_rad: 30f32.to_radians(),
    };
    let fb = renderer.render(
        &gpu.device,
        &gpu.queue,
        &scene,
        &[cam],
        30f32.to_radians(),
        64,
        0.0, // mip_scan_dist=0 → always mip-0
    );
    assert_eq!(fb.len(), (w * h) as usize);

    // Centre pixel: the down-ray hits the z=100 floor voxel of the
    // centre column. Expected colour = (R,G,B) * (alpha/128) / 255
    // → (255,128,0) at brightness 1.0 → rgba8 ≈ (255,128,0).
    let centre = fb[(h / 2 * w + w / 2) as usize];
    let (r, g, b) = (centre & 0xff, (centre >> 8) & 0xff, (centre >> 16) & 0xff);
    eprintln!("scene_render: centre pixel = ({r}, {g}, {b})");
    assert!(r > 200, "floor R should be ~255, got {r}");
    assert!((100..=160).contains(&g), "floor G should be ~128, got {g}");
    assert!(b < 40, "floor B should be ~0, got {b}");

    // The floor fills the frame at this near-vertical view; assert a
    // solid majority of pixels are floor-coloured (not sky / clear),
    // proving the strided mip-0 lookup works across the whole image.
    let floor_px = fb
        .iter()
        .filter(|&&p| {
            let (r, g, b) = (p & 0xff, (p >> 8) & 0xff, (p >> 16) & 0xff);
            r > 200 && (90..=170).contains(&g) && b < 50
        })
        .count();
    let frac = floor_px as f32 / fb.len() as f32;
    eprintln!("scene_render: floor fraction = {frac:.3}");
    assert!(frac > 0.6, "expected floor to fill the view, got {frac:.3}");
}