code-atlas 0.2.0

Render local code repositories as architectural treemaps with line-addressed, hierarchically bundled call graphs
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
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
use std::{fs, path::Path, sync::mpsc, time::Instant};

use anyhow::{Context, Result, anyhow, ensure};
use bytemuck::{Pod, Zeroable};
use tiny_skia::{IntSize, Pixmap};
use wgpu::util::DeviceExt;

use crate::{
    layout::{LayoutOptions, call_control_points},
    model::Atlas,
    render::{
        CallTile, CallTileViewport, Palette, RenderOptions, RenderStats, draw_labels,
        effective_call_opacity, optical_density_description, plan_call_tiles,
        render_software_base_png,
    },
};

const DENSITY_FORMAT: wgpu::TextureFormat = wgpu::TextureFormat::Rgba16Float;
const OUTPUT_FORMAT: wgpu::TextureFormat = wgpu::TextureFormat::Rgba8Unorm;
const SAMPLE_COUNT: u32 = 4;
const SPLINE_COMPUTE_SHADER: &str = include_str!("spline_density.wgsl");
const SPLINE_RENDER_SHADER: &str = include_str!("spline_render.wgsl");

const RESOLVE_SHADER: &str = r#"
@group(0) @binding(0)
var density_texture: texture_2d<f32>;

struct VertexOutput {
    @builtin(position) position: vec4<f32>,
};

@vertex
fn vs_main(@builtin(vertex_index) index: u32) -> VertexOutput {
    var positions = array<vec2<f32>, 3>(
        vec2<f32>(-1.0, -1.0),
        vec2<f32>(3.0, -1.0),
        vec2<f32>(-1.0, 3.0),
    );
    var output: VertexOutput;
    output.position = vec4<f32>(positions[index], 0.0, 1.0);
    return output;
}

fn linear_to_srgb(value: f32) -> f32 {
    if value <= 0.0031308 {
        return value * 12.92;
    }
    return 1.055 * pow(value, 1.0 / 2.4) - 0.055;
}

@fragment
fn fs_main(input: VertexOutput) -> @location(0) vec4<f32> {
    let pixel = vec2<i32>(input.position.xy);
    let deposited = textureLoad(density_texture, pixel, 0);
    if deposited.a <= 0.000001 {
        return vec4<f32>(0.0);
    }
    let alpha = 1.0 - exp(-deposited.a);
    let linear = clamp(deposited.rgb / deposited.a, vec3<f32>(0.0), vec3<f32>(1.0));
    let srgb = vec3<f32>(
        linear_to_srgb(linear.r),
        linear_to_srgb(linear.g),
        linear_to_srgb(linear.b),
    );
    return vec4<f32>(srgb * alpha, alpha);
}
"#;

#[repr(C)]
#[derive(Clone, Copy, Pod, Zeroable)]
struct GpuControlPoint {
    position: [f32; 2],
}

#[repr(C)]
#[derive(Clone, Copy, Pod, Zeroable)]
struct GpuCallRecord {
    control_offset: u32,
    control_count: u32,
    segment_count: u32,
    seed: u32,
}

#[repr(C)]
#[derive(Clone, Copy, Pod, Zeroable)]
struct GpuTileInstance {
    call_index: u32,
    pass_index: u32,
}

#[repr(C)]
#[derive(Clone, Copy, Pod, Zeroable)]
struct GpuSplineParams {
    viewport: [f32; 4],
    source: [f32; 4],
    target: [f32; 4],
    pass0: [f32; 4],
    pass1: [f32; 4],
    settings: [f32; 4],
    counts: [u32; 4],
}

pub(crate) struct GpuSplineScene {
    controls: wgpu::Buffer,
    calls: wgpu::Buffer,
    call_ids: Vec<u64>,
    segment_counts: Vec<u32>,
    max_segments: u32,
}

#[derive(Debug, Clone, Copy, Default)]
pub(crate) struct GpuPassTimings {
    pub spline_evaluation_ms: Option<f64>,
    pub rasterization_ms: Option<f64>,
}

pub(crate) struct WgpuDensityRenderer {
    device: wgpu::Device,
    queue: wgpu::Queue,
    adapter_name: String,
    spline_compute_pipeline: wgpu::ComputePipeline,
    accumulate_pipeline: wgpu::RenderPipeline,
    compute_bind_group_layout: wgpu::BindGroupLayout,
    accumulate_bind_group_layout: wgpu::BindGroupLayout,
    resolve_pipeline: wgpu::RenderPipeline,
    resolve_bind_group_layout: wgpu::BindGroupLayout,
    timestamp_queries: bool,
}

fn create_spline_bind_group_layout(
    device: &wgpu::Device,
    sampled_points_read_only: bool,
    visibility: wgpu::ShaderStages,
) -> wgpu::BindGroupLayout {
    let storage = |binding, read_only| wgpu::BindGroupLayoutEntry {
        binding,
        visibility,
        ty: wgpu::BindingType::Buffer {
            ty: wgpu::BufferBindingType::Storage { read_only },
            has_dynamic_offset: false,
            min_binding_size: None,
        },
        count: None,
    };
    device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
        label: Some(if sampled_points_read_only {
            "code-atlas GPU spline render bind group layout"
        } else {
            "code-atlas GPU spline compute bind group layout"
        }),
        entries: &[
            storage(0, true),
            storage(1, true),
            storage(2, true),
            storage(3, sampled_points_read_only),
            wgpu::BindGroupLayoutEntry {
                binding: 4,
                visibility,
                ty: wgpu::BindingType::Buffer {
                    ty: wgpu::BufferBindingType::Uniform,
                    has_dynamic_offset: false,
                    min_binding_size: None,
                },
                count: None,
            },
        ],
    })
}

impl WgpuDensityRenderer {
    pub(crate) fn new() -> Result<Self> {
        let instance = wgpu::Instance::default();
        let adapter = pollster::block_on(instance.request_adapter(&wgpu::RequestAdapterOptions {
            power_preference: wgpu::PowerPreference::HighPerformance,
            compatible_surface: None,
            force_fallback_adapter: false,
            apply_limit_buckets: false,
        }))
        .context("no compatible GPU adapter is available for the wgpu backend")?;
        let density_features = adapter.get_texture_format_features(DENSITY_FORMAT);
        ensure!(
            density_features
                .flags
                .contains(wgpu::TextureFormatFeatureFlags::BLENDABLE),
            "the GPU adapter does not support blendable RGBA16Float density textures"
        );
        ensure!(
            density_features
                .flags
                .contains(wgpu::TextureFormatFeatureFlags::MULTISAMPLE_X4)
                && density_features
                    .flags
                    .contains(wgpu::TextureFormatFeatureFlags::MULTISAMPLE_RESOLVE),
            "the GPU adapter does not support 4x multisampled RGBA16Float density textures"
        );
        let adapter_info = adapter.get_info();
        let timestamp_queries = adapter.features().contains(wgpu::Features::TIMESTAMP_QUERY);
        let required_features = if timestamp_queries {
            wgpu::Features::TIMESTAMP_QUERY
        } else {
            wgpu::Features::empty()
        };
        let (device, queue) = pollster::block_on(adapter.request_device(&wgpu::DeviceDescriptor {
            label: Some("code-atlas optical-density device"),
            required_features,
            ..Default::default()
        }))
        .context("cannot create the wgpu optical-density device")?;

        let spline_compute_shader = device.create_shader_module(wgpu::ShaderModuleDescriptor {
            label: Some("code-atlas GPU spline compute shader"),
            source: wgpu::ShaderSource::Wgsl(SPLINE_COMPUTE_SHADER.into()),
        });
        let spline_render_shader = device.create_shader_module(wgpu::ShaderModuleDescriptor {
            label: Some("code-atlas GPU spline render shader"),
            source: wgpu::ShaderSource::Wgsl(SPLINE_RENDER_SHADER.into()),
        });
        let compute_bind_group_layout =
            create_spline_bind_group_layout(&device, false, wgpu::ShaderStages::COMPUTE);
        let accumulate_bind_group_layout =
            create_spline_bind_group_layout(&device, true, wgpu::ShaderStages::VERTEX);
        let compute_layout = device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
            label: Some("code-atlas GPU spline compute layout"),
            bind_group_layouts: &[Some(&compute_bind_group_layout)],
            immediate_size: 0,
        });
        let spline_compute_pipeline =
            device.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
                label: Some("code-atlas GPU spline evaluation pipeline"),
                layout: Some(&compute_layout),
                module: &spline_compute_shader,
                entry_point: Some("compute_points"),
                compilation_options: Default::default(),
                cache: None,
            });
        let accumulate_layout = device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
            label: Some("code-atlas optical-density accumulation layout"),
            bind_group_layouts: &[Some(&accumulate_bind_group_layout)],
            immediate_size: 0,
        });
        let accumulate_pipeline = device.create_render_pipeline(&wgpu::RenderPipelineDescriptor {
            label: Some("code-atlas optical-density accumulation pipeline"),
            layout: Some(&accumulate_layout),
            vertex: wgpu::VertexState {
                module: &spline_render_shader,
                entry_point: Some("vs_main"),
                compilation_options: Default::default(),
                buffers: &[],
            },
            primitive: wgpu::PrimitiveState {
                topology: wgpu::PrimitiveTopology::TriangleList,
                ..Default::default()
            },
            depth_stencil: None,
            multisample: wgpu::MultisampleState {
                count: SAMPLE_COUNT,
                ..Default::default()
            },
            fragment: Some(wgpu::FragmentState {
                module: &spline_render_shader,
                entry_point: Some("fs_main"),
                compilation_options: Default::default(),
                targets: &[Some(wgpu::ColorTargetState {
                    format: DENSITY_FORMAT,
                    blend: Some(wgpu::BlendState::ADDITIVE),
                    write_mask: wgpu::ColorWrites::ALL,
                })],
            }),
            multiview_mask: None,
            cache: None,
        });

        let resolve_bind_group_layout =
            device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
                label: Some("code-atlas optical-density resolve bind group layout"),
                entries: &[wgpu::BindGroupLayoutEntry {
                    binding: 0,
                    visibility: wgpu::ShaderStages::FRAGMENT,
                    ty: wgpu::BindingType::Texture {
                        sample_type: wgpu::TextureSampleType::Float { filterable: false },
                        view_dimension: wgpu::TextureViewDimension::D2,
                        multisampled: false,
                    },
                    count: None,
                }],
            });
        let resolve_shader = device.create_shader_module(wgpu::ShaderModuleDescriptor {
            label: Some("code-atlas optical-density resolve shader"),
            source: wgpu::ShaderSource::Wgsl(RESOLVE_SHADER.into()),
        });
        let resolve_layout = device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
            label: Some("code-atlas optical-density resolve layout"),
            bind_group_layouts: &[Some(&resolve_bind_group_layout)],
            immediate_size: 0,
        });
        let resolve_pipeline = device.create_render_pipeline(&wgpu::RenderPipelineDescriptor {
            label: Some("code-atlas optical-density resolve pipeline"),
            layout: Some(&resolve_layout),
            vertex: wgpu::VertexState {
                module: &resolve_shader,
                entry_point: Some("vs_main"),
                compilation_options: Default::default(),
                buffers: &[],
            },
            primitive: wgpu::PrimitiveState::default(),
            depth_stencil: None,
            multisample: wgpu::MultisampleState::default(),
            fragment: Some(wgpu::FragmentState {
                module: &resolve_shader,
                entry_point: Some("fs_main"),
                compilation_options: Default::default(),
                targets: &[Some(wgpu::ColorTargetState {
                    format: OUTPUT_FORMAT,
                    blend: None,
                    write_mask: wgpu::ColorWrites::ALL,
                })],
            }),
            multiview_mask: None,
            cache: None,
        });

        Ok(Self {
            device,
            queue,
            adapter_name: format!("{} ({:?})", adapter_info.name, adapter_info.backend),
            spline_compute_pipeline,
            accumulate_pipeline,
            compute_bind_group_layout,
            accumulate_bind_group_layout,
            resolve_pipeline,
            resolve_bind_group_layout,
            timestamp_queries,
        })
    }

    pub(crate) fn adapter_name(&self) -> &str {
        &self.adapter_name
    }

    pub(crate) fn max_texture_dimension(&self) -> u32 {
        self.device.limits().max_texture_dimension_2d
    }

    pub(crate) fn prepare_scene(&self, atlas: &Atlas) -> Result<GpuSplineScene> {
        let mut controls = Vec::new();
        let mut records = Vec::with_capacity(atlas.calls.len());
        let mut call_ids = Vec::with_capacity(atlas.calls.len());
        let mut segment_counts = Vec::with_capacity(atlas.calls.len());
        let mut max_segments = 64_u32;
        for call in &atlas.calls {
            let control_points = call_control_points(atlas, call);
            ensure!(
                control_points.len() >= 2,
                "call {} has fewer than two hierarchy control points",
                call.id
            );
            let control_offset = u32::try_from(controls.len())
                .context("GPU spline control-point offset exceeds u32")?;
            let control_count = u32::try_from(control_points.len())
                .context("GPU spline control-point count exceeds u32")?;
            controls.extend(control_points.into_iter().map(|point| GpuControlPoint {
                position: [point.x as f32, point.y as f32],
            }));
            let degree = control_count.saturating_sub(1).min(3);
            let segment_count = control_count
                .saturating_sub(degree)
                .saturating_mul(16)
                .max(64);
            max_segments = max_segments.max(segment_count);
            let seed = (call.id as u32) ^ ((call.id >> 32) as u32).rotate_left(13) ^ 0xca11_517e;
            records.push(GpuCallRecord {
                control_offset,
                control_count,
                segment_count,
                seed,
            });
            call_ids.push(call.id);
            segment_counts.push(segment_count);
        }
        if controls.is_empty() {
            controls.push(GpuControlPoint {
                position: [0.0, 0.0],
            });
        }
        if records.is_empty() {
            records.push(GpuCallRecord {
                control_offset: 0,
                control_count: 1,
                segment_count: 0,
                seed: 0,
            });
        }
        let limits = self.device.limits();
        let control_bytes = std::mem::size_of_val(controls.as_slice()) as u64;
        let call_bytes = std::mem::size_of_val(records.as_slice()) as u64;
        ensure!(
            control_bytes <= limits.max_storage_buffer_binding_size
                && call_bytes <= limits.max_storage_buffer_binding_size,
            "GPU spline scene exceeds the adapter's storage-buffer binding limit"
        );
        let controls = self
            .device
            .create_buffer_init(&wgpu::util::BufferInitDescriptor {
                label: Some("code-atlas persistent hierarchy control points"),
                contents: bytemuck::cast_slice(&controls),
                usage: wgpu::BufferUsages::STORAGE,
            });
        let calls = self
            .device
            .create_buffer_init(&wgpu::util::BufferInitDescriptor {
                label: Some("code-atlas persistent compact call records"),
                contents: bytemuck::cast_slice(&records),
                usage: wgpu::BufferUsages::STORAGE,
            });
        Ok(GpuSplineScene {
            controls,
            calls,
            call_ids,
            segment_counts,
            max_segments,
        })
    }

    pub(crate) fn render_call_tile(
        &self,
        scene: &GpuSplineScene,
        layout: &LayoutOptions,
        render: &RenderOptions,
        call_indices: &[usize],
        viewport: CallTileViewport,
    ) -> Result<(Pixmap, usize)> {
        let (pixmap, calls, _) =
            self.render_call_tile_profiled(scene, layout, render, call_indices, viewport)?;
        Ok((pixmap, calls))
    }

    pub(crate) fn render_call_tile_profiled(
        &self,
        scene: &GpuSplineScene,
        layout: &LayoutOptions,
        render: &RenderOptions,
        call_indices: &[usize],
        viewport: CallTileViewport,
    ) -> Result<(Pixmap, usize, GpuPassTimings)> {
        ensure!(
            viewport.width > 0 && viewport.height > 0,
            "GPU call tile dimensions must be non-zero"
        );
        let max_dimension = self.device.limits().max_texture_dimension_2d;
        ensure!(
            viewport.width <= max_dimension && viewport.height <= max_dimension,
            "GPU call tile {}x{} exceeds the adapter's {}-pixel texture limit",
            viewport.width,
            viewport.height,
            max_dimension
        );
        let extent = wgpu::Extent3d {
            width: viewport.width,
            height: viewport.height,
            depth_or_array_layers: 1,
        };
        let density_msaa = self.device.create_texture(&wgpu::TextureDescriptor {
            label: Some("code-atlas multisampled optical-density tile"),
            size: extent,
            mip_level_count: 1,
            sample_count: SAMPLE_COUNT,
            dimension: wgpu::TextureDimension::D2,
            format: DENSITY_FORMAT,
            usage: wgpu::TextureUsages::RENDER_ATTACHMENT,
            view_formats: &[],
        });
        let density_resolved = self.device.create_texture(&wgpu::TextureDescriptor {
            label: Some("code-atlas resolved optical-density tile"),
            size: extent,
            mip_level_count: 1,
            sample_count: 1,
            dimension: wgpu::TextureDimension::D2,
            format: DENSITY_FORMAT,
            usage: wgpu::TextureUsages::RENDER_ATTACHMENT | wgpu::TextureUsages::TEXTURE_BINDING,
            view_formats: &[],
        });
        let output = self.device.create_texture(&wgpu::TextureDescriptor {
            label: Some("code-atlas resolved call tile"),
            size: extent,
            mip_level_count: 1,
            sample_count: 1,
            dimension: wgpu::TextureDimension::D2,
            format: OUTPUT_FORMAT,
            usage: wgpu::TextureUsages::RENDER_ATTACHMENT | wgpu::TextureUsages::COPY_SRC,
            view_formats: &[],
        });
        let density_msaa_view = density_msaa.create_view(&wgpu::TextureViewDescriptor::default());
        let density_resolved_view =
            density_resolved.create_view(&wgpu::TextureViewDescriptor::default());
        let output_view = output.create_view(&wgpu::TextureViewDescriptor::default());

        let opacity = effective_call_opacity(scene.call_ids.len(), render);
        let mut ordered_indices = call_indices.to_vec();
        ordered_indices.sort_unstable_by_key(|index| {
            (
                scene.call_ids.get(*index).copied().unwrap_or(u64::MAX),
                *index,
            )
        });
        let mut instances = Vec::with_capacity(ordered_indices.len().saturating_mul(2));
        let mut calls_drawn = 0;
        for index in ordered_indices {
            let Some(&segment_count) = scene.segment_counts.get(index) else {
                continue;
            };
            if segment_count == 0 {
                continue;
            }
            let call_index = u32::try_from(index).context("GPU call index exceeds u32")?;
            instances.push(GpuTileInstance {
                call_index,
                pass_index: 0,
            });
            instances.push(GpuTileInstance {
                call_index,
                pass_index: 1,
            });
            calls_drawn += 1;
        }
        self.clear_density_target(&density_msaa_view, &density_resolved_view);
        let batch_capacity = self.instance_batch_capacity(scene)?;
        let mut pass_timings = GpuPassTimings::default();
        for batch in instances.chunks(batch_capacity) {
            let batch_timings = self.submit_spline_instances(
                scene,
                layout,
                render,
                opacity,
                viewport,
                batch,
                &density_msaa_view,
                &density_resolved_view,
            )?;
            pass_timings.spline_evaluation_ms = sum_optional(
                pass_timings.spline_evaluation_ms,
                batch_timings.spline_evaluation_ms,
            );
            pass_timings.rasterization_ms = sum_optional(
                pass_timings.rasterization_ms,
                batch_timings.rasterization_ms,
            );
        }

        let resolve_bind_group = self.device.create_bind_group(&wgpu::BindGroupDescriptor {
            label: Some("code-atlas optical-density resolve bind group"),
            layout: &self.resolve_bind_group_layout,
            entries: &[wgpu::BindGroupEntry {
                binding: 0,
                resource: wgpu::BindingResource::TextureView(&density_resolved_view),
            }],
        });
        let mut encoder = self
            .device
            .create_command_encoder(&wgpu::CommandEncoderDescriptor {
                label: Some("code-atlas optical-density resolve encoder"),
            });
        {
            let mut pass = encoder.begin_render_pass(&wgpu::RenderPassDescriptor {
                label: Some("code-atlas optical-density resolve pass"),
                color_attachments: &[Some(wgpu::RenderPassColorAttachment {
                    view: &output_view,
                    depth_slice: None,
                    resolve_target: None,
                    ops: wgpu::Operations {
                        load: wgpu::LoadOp::Clear(wgpu::Color::TRANSPARENT),
                        store: wgpu::StoreOp::Store,
                    },
                })],
                depth_stencil_attachment: None,
                timestamp_writes: None,
                occlusion_query_set: None,
                multiview_mask: None,
            });
            pass.set_pipeline(&self.resolve_pipeline);
            pass.set_bind_group(0, &resolve_bind_group, &[]);
            pass.draw(0..3, 0..1);
        }
        self.queue.submit(Some(encoder.finish()));

        let pixels = read_texture(&self.device, &self.queue, &output, extent)?;
        let size = IntSize::from_wh(viewport.width, viewport.height)
            .context("invalid GPU call tile dimensions")?;
        let pixmap = Pixmap::from_vec(pixels, size).context("invalid GPU call tile pixels")?;
        Ok((pixmap, calls_drawn, pass_timings))
    }

    fn instance_batch_capacity(&self, scene: &GpuSplineScene) -> Result<usize> {
        let point_stride = u64::from(scene.max_segments + 1);
        let bytes_per_instance = point_stride
            .checked_mul(std::mem::size_of::<GpuControlPoint>() as u64)
            .context("GPU sampled-point stride overflows")?;
        let limits = self.device.limits();
        let storage_capacity = limits
            .max_storage_buffer_binding_size
            .min(limits.max_buffer_size)
            / bytes_per_instance.max(1);
        let dispatch_capacity = u64::from(limits.max_compute_workgroups_per_dimension) * 8;
        let capacity = storage_capacity
            .min(dispatch_capacity)
            .min(usize::MAX as u64) as usize;
        let capacity = capacity & !1;
        ensure!(
            capacity >= 2,
            "GPU storage limits cannot hold one two-pass spline instance"
        );
        Ok(capacity)
    }

    fn clear_density_target(
        &self,
        multisampled_view: &wgpu::TextureView,
        resolved_view: &wgpu::TextureView,
    ) {
        let mut encoder = self
            .device
            .create_command_encoder(&wgpu::CommandEncoderDescriptor {
                label: Some("code-atlas optical-density clear encoder"),
            });
        {
            let _pass = encoder.begin_render_pass(&wgpu::RenderPassDescriptor {
                label: Some("code-atlas optical-density clear pass"),
                color_attachments: &[Some(wgpu::RenderPassColorAttachment {
                    view: multisampled_view,
                    depth_slice: None,
                    resolve_target: Some(resolved_view),
                    ops: wgpu::Operations {
                        load: wgpu::LoadOp::Clear(wgpu::Color::TRANSPARENT),
                        store: wgpu::StoreOp::Store,
                    },
                })],
                depth_stencil_attachment: None,
                timestamp_writes: None,
                occlusion_query_set: None,
                multiview_mask: None,
            });
        }
        self.queue.submit(Some(encoder.finish()));
    }

    #[allow(clippy::too_many_arguments)]
    fn submit_spline_instances(
        &self,
        scene: &GpuSplineScene,
        layout: &LayoutOptions,
        render: &RenderOptions,
        opacity: u8,
        viewport: CallTileViewport,
        instances: &[GpuTileInstance],
        multisampled_view: &wgpu::TextureView,
        resolved_view: &wgpu::TextureView,
    ) -> Result<GpuPassTimings> {
        if instances.is_empty() {
            return Ok(GpuPassTimings::default());
        }
        let point_stride = scene.max_segments + 1;
        let point_count = u64::try_from(instances.len())
            .context("GPU tile instance count exceeds u64")?
            .checked_mul(u64::from(point_stride))
            .context("GPU sampled-point count overflows")?;
        let point_bytes = point_count
            .checked_mul(std::mem::size_of::<GpuControlPoint>() as u64)
            .context("GPU sampled-point buffer size overflows")?;
        let instance_buffer = self
            .device
            .create_buffer_init(&wgpu::util::BufferInitDescriptor {
                label: Some("code-atlas tile-local call/pass IDs"),
                contents: bytemuck::cast_slice(instances),
                usage: wgpu::BufferUsages::STORAGE,
            });
        let sampled_points = self.device.create_buffer(&wgpu::BufferDescriptor {
            label: Some("code-atlas GPU-evaluated spline points"),
            size: point_bytes,
            usage: wgpu::BufferUsages::STORAGE,
            mapped_at_creation: false,
        });
        let palette = Palette::for_theme(render.theme);
        let drawing_scale = (layout.width.min(layout.height) as f32 / 1080.0).max(0.5);
        let pass = |alpha_scale: f32, width_scale: f32, amplitude: f32| {
            let alpha = (f32::from(opacity) * alpha_scale).round().max(1.0) / 255.0;
            [
                render.call_width * drawing_scale * width_scale * 0.5,
                -(1.0 - alpha.min(0.999_999)).ln(),
                amplitude,
                0.0,
            ]
        };
        let params = GpuSplineParams {
            viewport: [
                viewport.x as f32,
                viewport.y as f32,
                viewport.width as f32,
                viewport.height as f32,
            ],
            source: linear_rgba(palette.direction_source),
            target: linear_rgba(palette.direction_target),
            pass0: pass(0.62, 0.82, 0.14),
            pass1: pass(0.38, 1.28, 0.34),
            settings: [layout.bundle_strength.clamp(0.0, 1.0) as f32, 0.0, 0.0, 0.0],
            counts: [
                point_stride,
                u32::try_from(instances.len()).context("GPU tile instance count exceeds u32")?,
                scene.max_segments,
                0,
            ],
        };
        let params_buffer = self
            .device
            .create_buffer_init(&wgpu::util::BufferInitDescriptor {
                label: Some("code-atlas GPU spline tile parameters"),
                contents: bytemuck::bytes_of(&params),
                usage: wgpu::BufferUsages::UNIFORM,
            });
        let compute_bind_group = self.create_spline_bind_group(
            &self.compute_bind_group_layout,
            scene,
            &instance_buffer,
            &sampled_points,
            &params_buffer,
        );
        let render_bind_group = self.create_spline_bind_group(
            &self.accumulate_bind_group_layout,
            scene,
            &instance_buffer,
            &sampled_points,
            &params_buffer,
        );
        let timestamp_resources = self.timestamp_queries.then(|| {
            let query_set = self.device.create_query_set(&wgpu::QuerySetDescriptor {
                label: Some("code-atlas GPU spline phase timestamps"),
                ty: wgpu::QueryType::Timestamp,
                count: 4,
            });
            let resolve = self.device.create_buffer(&wgpu::BufferDescriptor {
                label: Some("code-atlas GPU timestamp resolve"),
                size: 32,
                usage: wgpu::BufferUsages::QUERY_RESOLVE | wgpu::BufferUsages::COPY_SRC,
                mapped_at_creation: false,
            });
            let readback = self.device.create_buffer(&wgpu::BufferDescriptor {
                label: Some("code-atlas GPU timestamp readback"),
                size: 32,
                usage: wgpu::BufferUsages::COPY_DST | wgpu::BufferUsages::MAP_READ,
                mapped_at_creation: false,
            });
            (query_set, resolve, readback)
        });
        let mut encoder = self
            .device
            .create_command_encoder(&wgpu::CommandEncoderDescriptor {
                label: Some("code-atlas GPU spline evaluation and density encoder"),
            });
        {
            let mut pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
                label: Some("code-atlas GPU spline evaluation pass"),
                timestamp_writes: timestamp_resources.as_ref().map(|(query_set, _, _)| {
                    wgpu::ComputePassTimestampWrites {
                        query_set,
                        beginning_of_pass_write_index: Some(0),
                        end_of_pass_write_index: Some(1),
                    }
                }),
            });
            pass.set_pipeline(&self.spline_compute_pipeline);
            pass.set_bind_group(0, &compute_bind_group, &[]);
            pass.dispatch_workgroups(
                point_stride.div_ceil(8),
                u32::try_from(instances.len())
                    .context("GPU tile instance count exceeds u32")?
                    .div_ceil(8),
                1,
            );
        }
        {
            let mut pass = encoder.begin_render_pass(&wgpu::RenderPassDescriptor {
                label: Some("code-atlas GPU spline optical-density pass"),
                color_attachments: &[Some(wgpu::RenderPassColorAttachment {
                    view: multisampled_view,
                    depth_slice: None,
                    resolve_target: Some(resolved_view),
                    ops: wgpu::Operations {
                        load: wgpu::LoadOp::Load,
                        store: wgpu::StoreOp::Store,
                    },
                })],
                depth_stencil_attachment: None,
                timestamp_writes: timestamp_resources.as_ref().map(|(query_set, _, _)| {
                    wgpu::RenderPassTimestampWrites {
                        query_set,
                        beginning_of_pass_write_index: Some(2),
                        end_of_pass_write_index: Some(3),
                    }
                }),
                occlusion_query_set: None,
                multiview_mask: None,
            });
            pass.set_pipeline(&self.accumulate_pipeline);
            pass.set_bind_group(0, &render_bind_group, &[]);
            pass.draw(
                0..scene.max_segments.saturating_mul(6),
                0..u32::try_from(instances.len()).context("GPU instance count exceeds u32")?,
            );
        }
        if let Some((query_set, resolve, readback)) = &timestamp_resources {
            encoder.resolve_query_set(query_set, 0..4, resolve, 0);
            encoder.copy_buffer_to_buffer(resolve, 0, readback, 0, 32);
        }
        self.queue.submit(Some(encoder.finish()));
        if let Some((_, _, readback)) = timestamp_resources {
            let values = read_u64_buffer(&self.device, &readback)?;
            let period = f64::from(self.queue.get_timestamp_period()) / 1_000_000.0;
            return Ok(GpuPassTimings {
                spline_evaluation_ms: Some(values[1].saturating_sub(values[0]) as f64 * period),
                rasterization_ms: Some(values[3].saturating_sub(values[2]) as f64 * period),
            });
        }
        Ok(GpuPassTimings::default())
    }

    fn create_spline_bind_group(
        &self,
        layout: &wgpu::BindGroupLayout,
        scene: &GpuSplineScene,
        instances: &wgpu::Buffer,
        sampled_points: &wgpu::Buffer,
        params: &wgpu::Buffer,
    ) -> wgpu::BindGroup {
        self.device.create_bind_group(&wgpu::BindGroupDescriptor {
            label: Some("code-atlas GPU spline bind group"),
            layout,
            entries: &[
                wgpu::BindGroupEntry {
                    binding: 0,
                    resource: scene.controls.as_entire_binding(),
                },
                wgpu::BindGroupEntry {
                    binding: 1,
                    resource: scene.calls.as_entire_binding(),
                },
                wgpu::BindGroupEntry {
                    binding: 2,
                    resource: instances.as_entire_binding(),
                },
                wgpu::BindGroupEntry {
                    binding: 3,
                    resource: sampled_points.as_entire_binding(),
                },
                wgpu::BindGroupEntry {
                    binding: 4,
                    resource: params.as_entire_binding(),
                },
            ],
        })
    }
}

pub(crate) fn render_wgpu_png(
    atlas: &Atlas,
    output: &Path,
    layout: &LayoutOptions,
    render: &RenderOptions,
) -> Result<RenderStats> {
    let started = Instant::now();
    let mut stats = render_software_base_png(atlas, output, layout, render)?;
    let mut pixmap = Pixmap::load_png(output)
        .with_context(|| format!("cannot reload software base {}", output.display()))?;
    let renderer = WgpuDensityRenderer::new()?;
    let scene = renderer.prepare_scene(atlas)?;
    let (tiles, calls_drawn, overlap) = plan_call_tiles(atlas, layout, render)?;
    for tile in &tiles {
        let (call_tile, _) = render_tile(&renderer, &scene, layout, render, tile)?;
        composite_tile_core(&mut pixmap, &call_tile, tile);
    }
    let labels = draw_labels(&mut pixmap, atlas, layout, Palette::for_theme(render.theme))?;
    pixmap
        .save_png(output)
        .with_context(|| format!("cannot save {}", output.display()))?;
    stats.backend = "wgpu spline-evaluated optical-density tiled".to_owned();
    stats.gpu_adapter = Some(renderer.adapter_name().to_owned());
    stats.call_layer_tiles = tiles.len();
    stats.call_layer_tile_size = Some(render.pdf_call_tile_size);
    stats.call_layer_overlap = Some(overlap);
    stats.calls_drawn = calls_drawn;
    stats.file_labels_drawn = labels.files;
    stats.directory_labels_drawn = labels.directories;
    stats.output_bytes = fs::metadata(output)?.len();
    stats.effective_call_opacity = effective_call_opacity(atlas.calls.len(), render);
    stats.call_compositing = optical_density_description();
    stats.render_ms = started.elapsed().as_millis().min(u128::from(u64::MAX)) as u64;
    Ok(stats)
}

pub(crate) fn render_wgpu_call_layer_png(
    atlas: &Atlas,
    output: &Path,
    layout: &LayoutOptions,
    render: &RenderOptions,
) -> Result<RenderStats> {
    let started = Instant::now();
    let renderer = WgpuDensityRenderer::new()?;
    let scene = renderer.prepare_scene(atlas)?;
    let (tiles, calls_drawn, overlap) = plan_call_tiles(atlas, layout, render)?;
    let mut pixmap = Pixmap::new(layout.width, layout.height)
        .context("output dimensions are too large for the GPU call renderer")?;
    for tile in &tiles {
        let (call_tile, _) = render_tile(&renderer, &scene, layout, render, tile)?;
        composite_tile_core(&mut pixmap, &call_tile, tile);
    }
    if let Some(parent) = output.parent() {
        fs::create_dir_all(parent)
            .with_context(|| format!("cannot create output directory {}", parent.display()))?;
    }
    pixmap
        .save_png(output)
        .with_context(|| format!("cannot save {}", output.display()))?;
    let palette = Palette::for_theme(render.theme);
    Ok(RenderStats {
        backend: "wgpu spline-evaluated optical-density tiled".to_owned(),
        gpu_adapter: Some(renderer.adapter_name().to_owned()),
        width: layout.width,
        height: layout.height,
        call_layer_width: layout.width,
        call_layer_height: layout.height,
        call_layer_dpi: None,
        call_layer_tiles: tiles.len(),
        call_layer_tile_size: Some(render.pdf_call_tile_size),
        call_layer_overlap: Some(overlap),
        files_drawn: atlas.files().count(),
        file_labels_drawn: 0,
        directory_labels_drawn: 0,
        calls_drawn,
        directory_boundaries_drawn: atlas
            .nodes
            .iter()
            .filter(|node| node.kind == crate::model::NodeKind::Directory && node.id != 0)
            .count(),
        output_bytes: fs::metadata(output)?.len(),
        effective_call_opacity: effective_call_opacity(atlas.calls.len(), render),
        call_compositing: optical_density_description(),
        direction_source_color: palette.direction_source.hex(),
        direction_target_color: palette.direction_target.hex(),
        render_ms: started.elapsed().as_millis().min(u128::from(u64::MAX)) as u64,
        pdf_tile_cache: None,
    })
}

fn render_tile(
    renderer: &WgpuDensityRenderer,
    scene: &GpuSplineScene,
    layout: &LayoutOptions,
    render: &RenderOptions,
    tile: &CallTile,
) -> Result<(Pixmap, usize)> {
    renderer.render_call_tile(
        scene,
        layout,
        render,
        &tile.call_indices,
        CallTileViewport {
            x: tile.render_x,
            y: tile.render_y,
            width: tile.render_width,
            height: tile.render_height,
        },
    )
}

fn linear_rgba(color: crate::render::Rgba) -> [f32; 4] {
    let convert = |value: u8| {
        let value = f32::from(value) / 255.0;
        if value <= 0.04045 {
            value / 12.92
        } else {
            ((value + 0.055) / 1.055).powf(2.4)
        }
    };
    [
        convert(color.r),
        convert(color.g),
        convert(color.b),
        f32::from(color.a) / 255.0,
    ]
}

fn sum_optional(left: Option<f64>, right: Option<f64>) -> Option<f64> {
    match (left, right) {
        (Some(left), Some(right)) => Some(left + right),
        (None, Some(right)) => Some(right),
        (Some(left), None) => Some(left),
        (None, None) => None,
    }
}

fn composite_tile_core(destination: &mut Pixmap, source: &Pixmap, tile: &CallTile) {
    for y in tile.core_y..tile.core_y + tile.core_height {
        for x in tile.core_x..tile.core_x + tile.core_width {
            let source_x = x - tile.render_x;
            let source_y = y - tile.render_y;
            let source_offset =
                (source_y as usize * source.width() as usize + source_x as usize) * 4;
            let destination_offset = (y as usize * destination.width() as usize + x as usize) * 4;
            let source_pixel = &source.data()[source_offset..source_offset + 4];
            let destination_pixel =
                &mut destination.data_mut()[destination_offset..destination_offset + 4];
            let inverse_alpha = 255_u32 - u32::from(source_pixel[3]);
            for channel in 0..3 {
                destination_pixel[channel] = (u32::from(source_pixel[channel])
                    + (u32::from(destination_pixel[channel]) * inverse_alpha + 127) / 255)
                    .min(255) as u8;
            }
            destination_pixel[3] = (u32::from(source_pixel[3])
                + (u32::from(destination_pixel[3]) * inverse_alpha + 127) / 255)
                .min(255) as u8;
        }
    }
}

fn read_texture(
    device: &wgpu::Device,
    queue: &wgpu::Queue,
    texture: &wgpu::Texture,
    extent: wgpu::Extent3d,
) -> Result<Vec<u8>> {
    let unpadded_bytes_per_row = extent.width * 4;
    let alignment = wgpu::COPY_BYTES_PER_ROW_ALIGNMENT;
    let padded_bytes_per_row = unpadded_bytes_per_row.div_ceil(alignment) * alignment;
    let output_buffer = device.create_buffer(&wgpu::BufferDescriptor {
        label: Some("code-atlas optical-density readback buffer"),
        size: u64::from(padded_bytes_per_row) * u64::from(extent.height),
        usage: wgpu::BufferUsages::COPY_DST | wgpu::BufferUsages::MAP_READ,
        mapped_at_creation: false,
    });
    let mut encoder = device.create_command_encoder(&wgpu::CommandEncoderDescriptor {
        label: Some("code-atlas optical-density readback encoder"),
    });
    encoder.copy_texture_to_buffer(
        wgpu::TexelCopyTextureInfo {
            texture,
            mip_level: 0,
            origin: wgpu::Origin3d::ZERO,
            aspect: wgpu::TextureAspect::All,
        },
        wgpu::TexelCopyBufferInfo {
            buffer: &output_buffer,
            layout: wgpu::TexelCopyBufferLayout {
                offset: 0,
                bytes_per_row: Some(padded_bytes_per_row),
                rows_per_image: Some(extent.height),
            },
        },
        extent,
    );
    queue.submit(Some(encoder.finish()));

    let slice = output_buffer.slice(..);
    let (sender, receiver) = mpsc::channel();
    slice.map_async(wgpu::MapMode::Read, move |result| {
        let _ = sender.send(result);
    });
    device
        .poll(wgpu::PollType::wait_indefinitely())
        .map_err(|error| anyhow!("GPU optical-density readback poll failed: {error}"))?;
    receiver
        .recv()
        .context("GPU optical-density readback callback was dropped")?
        .context("cannot map the GPU optical-density readback buffer")?;

    let mapped = slice
        .get_mapped_range()
        .context("cannot access the GPU optical-density readback buffer")?;
    let mut pixels = Vec::with_capacity((unpadded_bytes_per_row * extent.height) as usize);
    for row in mapped.chunks_exact(padded_bytes_per_row as usize) {
        pixels.extend_from_slice(&row[..unpadded_bytes_per_row as usize]);
    }
    drop(mapped);
    output_buffer.unmap();
    Ok(pixels)
}

fn read_u64_buffer(device: &wgpu::Device, buffer: &wgpu::Buffer) -> Result<[u64; 4]> {
    let slice = buffer.slice(..);
    let (sender, receiver) = mpsc::channel();
    slice.map_async(wgpu::MapMode::Read, move |result| {
        let _ = sender.send(result);
    });
    device
        .poll(wgpu::PollType::wait_indefinitely())
        .map_err(|error| anyhow!("GPU timestamp readback poll failed: {error}"))?;
    receiver
        .recv()
        .context("GPU timestamp readback callback was dropped")?
        .context("cannot map GPU timestamps")?;
    let mapped = slice
        .get_mapped_range()
        .context("cannot access GPU timestamps")?;
    let values = bytemuck::try_from_bytes::<[u64; 4]>(&mapped)
        .map_err(|error| anyhow!("GPU timestamp buffer has an invalid size: {error:?}"))?
        .to_owned();
    drop(mapped);
    buffer.unmap();
    Ok(values)
}

#[cfg(test)]
mod tests {
    use std::{collections::HashMap, path::PathBuf};

    use crate::{
        model::{AnalyzerReport, Atlas, BuildTimings, Callsite, Node, NodeKind, Rect},
        render::{Theme, plan_call_tiles, render_software_call_tile},
    };

    use super::*;

    #[test]
    fn gpu_density_preserves_order_overlap_reference_and_tile_invariants() {
        let Ok(renderer) = WgpuDensityRenderer::new() else {
            eprintln!("skipping GPU density invariants because no compatible adapter is available");
            return;
        };
        let (atlas, layout) = fixture();
        let scene = renderer.prepare_scene(&atlas).unwrap();
        let indices = [0, 1, 2];
        let viewport = CallTileViewport {
            x: 0,
            y: 0,
            width: layout.width,
            height: layout.height,
        };
        for theme in [Theme::Architect, Theme::Night] {
            let render = RenderOptions {
                theme,
                pdf_call_tile_size: 64,
                ..RenderOptions::default()
            };
            let (forward, _) = renderer
                .render_call_tile(&scene, &layout, &render, &indices, viewport)
                .unwrap();
            let (reverse, _) = renderer
                .render_call_tile(&scene, &layout, &render, &[2, 1, 0], viewport)
                .unwrap();
            assert_eq!(
                forward.data(),
                reverse.data(),
                "GPU density resolve depends on input order for {theme}"
            );

            let (single, _) = renderer
                .render_call_tile(&scene, &layout, &render, &[0], viewport)
                .unwrap();
            let (double, _) = renderer
                .render_call_tile(&scene, &layout, &render, &[0, 1], viewport)
                .unwrap();
            assert!(
                alpha_sum(&double) > alpha_sum(&single),
                "repeated GPU routes do not increase optical density for {theme}"
            );

            let (cpu, _) =
                render_software_call_tile(&atlas, &layout, &render, &indices, viewport).unwrap();
            assert_reference_similarity(&cpu, &forward, theme);
            assert_gpu_tiles_match(&renderer, &scene, &atlas, &layout, &render, &forward);
        }
    }

    fn assert_reference_similarity(cpu: &Pixmap, gpu: &Pixmap, theme: Theme) {
        let mut alpha_difference = 0_u64;
        let mut cpu_alpha = 0_u64;
        let mut intersection = 0_u64;
        let mut union = 0_u64;
        for (cpu, gpu) in cpu.data().chunks_exact(4).zip(gpu.data().chunks_exact(4)) {
            alpha_difference += u64::from(cpu[3].abs_diff(gpu[3]));
            cpu_alpha += u64::from(cpu[3]);
            let cpu_covered = cpu[3] >= 4;
            let gpu_covered = gpu[3] >= 4;
            intersection += u64::from(cpu_covered && gpu_covered);
            union += u64::from(cpu_covered || gpu_covered);
        }
        let relative_alpha_error = alpha_difference as f64 / cpu_alpha.max(1) as f64;
        let coverage_iou = intersection as f64 / union.max(1) as f64;
        assert!(
            relative_alpha_error < 0.45,
            "GPU/CPU relative alpha error {relative_alpha_error:.3} is too high for {theme}"
        );
        assert!(
            coverage_iou > 0.72,
            "GPU/CPU coverage IoU {coverage_iou:.3} is too low for {theme}"
        );
    }

    fn assert_gpu_tiles_match(
        renderer: &WgpuDensityRenderer,
        scene: &GpuSplineScene,
        atlas: &Atlas,
        layout: &LayoutOptions,
        render: &RenderOptions,
        full: &Pixmap,
    ) {
        let (tiles, _, _) = plan_call_tiles(atlas, layout, render).unwrap();
        assert!(tiles.len() > 1);
        for tile in &tiles {
            let (pixmap, _) = render_tile(renderer, scene, layout, render, tile).unwrap();
            for y in tile.core_y..tile.core_y + tile.core_height {
                for x in tile.core_x..tile.core_x + tile.core_width {
                    let full_offset = (y as usize * layout.width as usize + x as usize) * 4;
                    let local_x = x - tile.render_x;
                    let local_y = y - tile.render_y;
                    let tile_offset =
                        (local_y as usize * tile.render_width as usize + local_x as usize) * 4;
                    for channel in 0..4 {
                        assert!(
                            full.data()[full_offset + channel]
                                .abs_diff(pixmap.data()[tile_offset + channel])
                                <= 2,
                            "GPU tile differs at ({x}, {y}) channel {channel}"
                        );
                    }
                }
            }
        }
    }

    fn alpha_sum(pixmap: &Pixmap) -> u64 {
        pixmap
            .data()
            .chunks_exact(4)
            .map(|pixel| u64::from(pixel[3]))
            .sum()
    }

    fn fixture() -> (Atlas, LayoutOptions) {
        let layout = LayoutOptions {
            width: 160,
            height: 120,
            ..LayoutOptions::default()
        };
        let file = |id, path: &str, rect| Node {
            id,
            parent: Some(0),
            children: Vec::new(),
            kind: NodeKind::File,
            name: path.to_owned(),
            path: path.to_owned(),
            depth: 1,
            bytes: 20,
            loc: 20,
            commits: 1,
            weight: 20.0,
            language: "rust".to_owned(),
            rect,
        };
        let call = |id, source, source_line, target, target_line, callee: &str| Callsite {
            id,
            source,
            source_line,
            target,
            target_line: Some(target_line),
            callee: callee.to_owned(),
            kind: "runtime".to_owned(),
            analyzer: "fixture".to_owned(),
            confidence: 1.0,
        };
        let atlas = Atlas {
            root_path: PathBuf::from("fixture"),
            revision: "test".to_owned(),
            dirty: false,
            excluded_test_files: 0,
            excluded_hidden_files: 0,
            excluded_custom_files: 0,
            excluded_paths: Vec::new(),
            nodes: vec![
                Node {
                    id: 0,
                    parent: None,
                    children: vec![1, 2],
                    kind: NodeKind::Directory,
                    name: "fixture".to_owned(),
                    path: String::new(),
                    depth: 0,
                    bytes: 40,
                    loc: 40,
                    commits: 2,
                    weight: 40.0,
                    language: "directory".to_owned(),
                    rect: Rect {
                        x0: 0.0,
                        y0: 0.0,
                        x1: 160.0,
                        y1: 120.0,
                    },
                },
                file(
                    1,
                    "a.rs",
                    Rect {
                        x0: 4.0,
                        y0: 4.0,
                        x1: 70.0,
                        y1: 116.0,
                    },
                ),
                file(
                    2,
                    "b.rs",
                    Rect {
                        x0: 90.0,
                        y0: 4.0,
                        x1: 156.0,
                        y1: 116.0,
                    },
                ),
            ],
            calls: vec![
                call(11, 1, 2, 2, 19, "b::low"),
                call(11, 1, 2, 2, 19, "b::low"),
                call(7, 2, 3, 1, 18, "a::high"),
            ],
            path_to_id: HashMap::new(),
            report: AnalyzerReport::default(),
            timings: BuildTimings::default(),
        };
        (atlas, layout)
    }
}