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
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
#[cfg(feature = "trace")]
use crate::device::trace::Action;
use crate::{
    api_log,
    command::{
        extract_texture_selector, validate_linear_texture_data, validate_texture_copy_range,
        ClearError, CommandAllocator, CommandBuffer, CopySide, ImageCopyTexture, TransferError,
    },
    conv,
    device::{life::ResourceMaps, DeviceError, WaitIdleError},
    get_lowest_common_denom,
    global::Global,
    hal_api::HalApi,
    hal_label,
    id::{self, DeviceId, QueueId},
    init_tracker::{has_copy_partial_init_tracker_coverage, TextureInitRange},
    lock::{rank, Mutex},
    resource::{
        Buffer, BufferAccessError, BufferMapState, DestroyedBuffer, DestroyedTexture, Resource,
        ResourceInfo, ResourceType, StagingBuffer, Texture, TextureInner,
    },
    resource_log, track, FastHashMap, SubmissionIndex,
};

use hal::{CommandEncoder as _, Device as _, Queue as _};
use smallvec::SmallVec;

use std::{
    iter, mem, ptr,
    sync::{atomic::Ordering, Arc},
};
use thiserror::Error;

use super::Device;

pub struct Queue<A: HalApi> {
    pub(crate) device: Option<Arc<Device<A>>>,
    pub(crate) raw: Option<A::Queue>,
    pub(crate) info: ResourceInfo<Queue<A>>,
}

impl<A: HalApi> Resource for Queue<A> {
    const TYPE: ResourceType = "Queue";

    type Marker = crate::id::markers::Queue;

    fn as_info(&self) -> &ResourceInfo<Self> {
        &self.info
    }

    fn as_info_mut(&mut self) -> &mut ResourceInfo<Self> {
        &mut self.info
    }
}

impl<A: HalApi> Drop for Queue<A> {
    fn drop(&mut self) {
        let queue = self.raw.take().unwrap();
        self.device.as_ref().unwrap().release_queue(queue);
    }
}

/// Number of command buffers that we generate from the same pool
/// for the write_xxx commands, before the pool is recycled.
///
/// If we don't stop at some point, the pool will grow forever,
/// without a concrete moment of when it can be cleared.
const WRITE_COMMAND_BUFFERS_PER_POOL: usize = 64;

#[repr(C)]
pub struct SubmittedWorkDoneClosureC {
    pub callback: unsafe extern "C" fn(user_data: *mut u8),
    pub user_data: *mut u8,
}

#[cfg(send_sync)]
unsafe impl Send for SubmittedWorkDoneClosureC {}

pub struct SubmittedWorkDoneClosure {
    // We wrap this so creating the enum in the C variant can be unsafe,
    // allowing our call function to be safe.
    inner: SubmittedWorkDoneClosureInner,
}

#[cfg(send_sync)]
type SubmittedWorkDoneCallback = Box<dyn FnOnce() + Send + 'static>;
#[cfg(not(send_sync))]
type SubmittedWorkDoneCallback = Box<dyn FnOnce() + 'static>;

enum SubmittedWorkDoneClosureInner {
    Rust { callback: SubmittedWorkDoneCallback },
    C { inner: SubmittedWorkDoneClosureC },
}

impl SubmittedWorkDoneClosure {
    pub fn from_rust(callback: SubmittedWorkDoneCallback) -> Self {
        Self {
            inner: SubmittedWorkDoneClosureInner::Rust { callback },
        }
    }

    /// # Safety
    ///
    /// - The callback pointer must be valid to call with the provided `user_data`
    ///   pointer.
    ///
    /// - Both pointers must point to `'static` data, as the callback may happen at
    ///   an unspecified time.
    pub unsafe fn from_c(inner: SubmittedWorkDoneClosureC) -> Self {
        Self {
            inner: SubmittedWorkDoneClosureInner::C { inner },
        }
    }

    pub(crate) fn call(self) {
        match self.inner {
            SubmittedWorkDoneClosureInner::Rust { callback } => callback(),
            // SAFETY: the contract of the call to from_c says that this unsafe is sound.
            SubmittedWorkDoneClosureInner::C { inner } => unsafe {
                (inner.callback)(inner.user_data)
            },
        }
    }
}

#[repr(C)]
#[derive(Debug, Copy, Clone)]
pub struct WrappedSubmissionIndex {
    pub queue_id: QueueId,
    pub index: SubmissionIndex,
}

/// A texture or buffer to be freed soon.
///
/// This is just a tagged raw texture or buffer, generally about to be added to
/// some other more specific container like:
///
/// - `PendingWrites::temp_resources`: resources used by queue writes and
///   unmaps, waiting to be folded in with the next queue submission
///
/// - `ActiveSubmission::last_resources`: temporary resources used by a queue
///   submission, to be freed when it completes
///
/// - `LifetimeTracker::free_resources`: resources to be freed in the next
///   `maintain` call, no longer used anywhere
#[derive(Debug)]
pub enum TempResource<A: HalApi> {
    Buffer(Arc<Buffer<A>>),
    StagingBuffer(Arc<StagingBuffer<A>>),
    DestroyedBuffer(Arc<DestroyedBuffer<A>>),
    DestroyedTexture(Arc<DestroyedTexture<A>>),
    Texture(Arc<Texture<A>>),
}

/// A series of raw [`CommandBuffer`]s that have been submitted to a
/// queue, and the [`wgpu_hal::CommandEncoder`] that built them.
///
/// [`CommandBuffer`]: hal::Api::CommandBuffer
/// [`wgpu_hal::CommandEncoder`]: hal::CommandEncoder
pub(crate) struct EncoderInFlight<A: HalApi> {
    raw: A::CommandEncoder,
    cmd_buffers: Vec<A::CommandBuffer>,
}

impl<A: HalApi> EncoderInFlight<A> {
    /// Free all of our command buffers.
    ///
    /// Return the command encoder, fully reset and ready to be
    /// reused.
    pub(crate) unsafe fn land(mut self) -> A::CommandEncoder {
        unsafe { self.raw.reset_all(self.cmd_buffers.into_iter()) };
        self.raw
    }
}

/// A private command encoder for writes made directly on the device
/// or queue.
///
/// Operations like `buffer_unmap`, `queue_write_buffer`, and
/// `queue_write_texture` need to copy data to the GPU. At the hal
/// level, this must be done by encoding and submitting commands, but
/// these operations are not associated with any specific wgpu command
/// buffer.
///
/// Instead, `Device::pending_writes` owns one of these values, which
/// has its own hal command encoder and resource lists. The commands
/// accumulated here are automatically submitted to the queue the next
/// time the user submits a wgpu command buffer, ahead of the user's
/// commands.
///
/// Important:
/// When locking pending_writes be sure that tracker is not locked
/// and try to lock trackers for the minimum timespan possible
///
/// All uses of [`StagingBuffer`]s end up here.
#[derive(Debug)]
pub(crate) struct PendingWrites<A: HalApi> {
    pub command_encoder: A::CommandEncoder,

    /// True if `command_encoder` is in the "recording" state, as
    /// described in the docs for the [`wgpu_hal::CommandEncoder`]
    /// trait.
    ///
    /// [`wgpu_hal::CommandEncoder`]: hal::CommandEncoder
    pub is_recording: bool,

    pub temp_resources: Vec<TempResource<A>>,
    pub dst_buffers: FastHashMap<id::BufferId, Arc<Buffer<A>>>,
    pub dst_textures: FastHashMap<id::TextureId, Arc<Texture<A>>>,

    /// All command buffers allocated from `command_encoder`.
    pub executing_command_buffers: Vec<A::CommandBuffer>,
}

impl<A: HalApi> PendingWrites<A> {
    pub fn new(command_encoder: A::CommandEncoder) -> Self {
        Self {
            command_encoder,
            is_recording: false,
            temp_resources: Vec::new(),
            dst_buffers: FastHashMap::default(),
            dst_textures: FastHashMap::default(),
            executing_command_buffers: Vec::new(),
        }
    }

    pub fn dispose(mut self, device: &A::Device) {
        unsafe {
            if self.is_recording {
                self.command_encoder.discard_encoding();
            }
            self.command_encoder
                .reset_all(self.executing_command_buffers.into_iter());
            device.destroy_command_encoder(self.command_encoder);
        }

        self.temp_resources.clear();
    }

    pub fn consume_temp(&mut self, resource: TempResource<A>) {
        self.temp_resources.push(resource);
    }

    fn consume(&mut self, buffer: Arc<StagingBuffer<A>>) {
        self.temp_resources
            .push(TempResource::StagingBuffer(buffer));
    }

    fn pre_submit(&mut self) -> Result<Option<&A::CommandBuffer>, DeviceError> {
        self.dst_buffers.clear();
        self.dst_textures.clear();
        if self.is_recording {
            let cmd_buf = unsafe { self.command_encoder.end_encoding()? };
            self.is_recording = false;
            self.executing_command_buffers.push(cmd_buf);

            return Ok(self.executing_command_buffers.last());
        }

        Ok(None)
    }

    #[must_use]
    fn post_submit(
        &mut self,
        command_allocator: &CommandAllocator<A>,
        device: &A::Device,
        queue: &A::Queue,
    ) -> Option<EncoderInFlight<A>> {
        if self.executing_command_buffers.len() >= WRITE_COMMAND_BUFFERS_PER_POOL {
            let new_encoder = command_allocator.acquire_encoder(device, queue).unwrap();
            Some(EncoderInFlight {
                raw: mem::replace(&mut self.command_encoder, new_encoder),
                cmd_buffers: mem::take(&mut self.executing_command_buffers),
            })
        } else {
            None
        }
    }

    pub fn activate(&mut self) -> &mut A::CommandEncoder {
        if !self.is_recording {
            unsafe {
                self.command_encoder
                    .begin_encoding(Some("(wgpu internal) PendingWrites"))
                    .unwrap();
            }
            self.is_recording = true;
        }
        &mut self.command_encoder
    }

    pub fn deactivate(&mut self) {
        if self.is_recording {
            unsafe {
                self.command_encoder.discard_encoding();
            }
            self.is_recording = false;
        }
    }
}

fn prepare_staging_buffer<A: HalApi>(
    device: &Arc<Device<A>>,
    size: wgt::BufferAddress,
    instance_flags: wgt::InstanceFlags,
) -> Result<(StagingBuffer<A>, *mut u8), DeviceError> {
    profiling::scope!("prepare_staging_buffer");
    let stage_desc = hal::BufferDescriptor {
        label: hal_label(Some("(wgpu internal) Staging"), instance_flags),
        size,
        usage: hal::BufferUses::MAP_WRITE | hal::BufferUses::COPY_SRC,
        memory_flags: hal::MemoryFlags::TRANSIENT,
    };

    let buffer = unsafe { device.raw().create_buffer(&stage_desc)? };
    let mapping = unsafe { device.raw().map_buffer(&buffer, 0..size) }?;

    let staging_buffer = StagingBuffer {
        raw: Mutex::new(rank::STAGING_BUFFER_RAW, Some(buffer)),
        device: device.clone(),
        size,
        info: ResourceInfo::new(
            "<StagingBuffer>",
            Some(device.tracker_indices.staging_buffers.clone()),
        ),
        is_coherent: mapping.is_coherent,
    };

    Ok((staging_buffer, mapping.ptr.as_ptr()))
}

impl<A: HalApi> StagingBuffer<A> {
    unsafe fn flush(&self, device: &A::Device) -> Result<(), DeviceError> {
        if !self.is_coherent {
            unsafe {
                device.flush_mapped_ranges(
                    self.raw.lock().as_ref().unwrap(),
                    iter::once(0..self.size),
                )
            };
        }
        unsafe { device.unmap_buffer(self.raw.lock().as_ref().unwrap())? };
        Ok(())
    }
}

#[derive(Clone, Debug, Error)]
#[error("Queue is invalid")]
pub struct InvalidQueue;

#[derive(Clone, Debug, Error)]
#[non_exhaustive]
pub enum QueueWriteError {
    #[error(
        "Device of queue ({:?}) does not match device of write recipient ({:?})",
        queue_device_id,
        target_device_id
    )]
    DeviceMismatch {
        queue_device_id: DeviceId,
        target_device_id: DeviceId,
    },
    #[error(transparent)]
    Queue(#[from] DeviceError),
    #[error(transparent)]
    Transfer(#[from] TransferError),
    #[error(transparent)]
    MemoryInitFailure(#[from] ClearError),
}

#[derive(Clone, Debug, Error)]
#[non_exhaustive]
pub enum QueueSubmitError {
    #[error(transparent)]
    Queue(#[from] DeviceError),
    #[error("Buffer {0:?} is destroyed")]
    DestroyedBuffer(id::BufferId),
    #[error("Texture {0:?} is destroyed")]
    DestroyedTexture(id::TextureId),
    #[error(transparent)]
    Unmap(#[from] BufferAccessError),
    #[error("Buffer {0:?} is still mapped")]
    BufferStillMapped(id::BufferId),
    #[error("Surface output was dropped before the command buffer got submitted")]
    SurfaceOutputDropped,
    #[error("Surface was unconfigured before the command buffer got submitted")]
    SurfaceUnconfigured,
    #[error("GPU got stuck :(")]
    StuckGpu,
}

//TODO: move out common parts of write_xxx.

impl Global {
    pub fn queue_write_buffer<A: HalApi>(
        &self,
        queue_id: QueueId,
        buffer_id: id::BufferId,
        buffer_offset: wgt::BufferAddress,
        data: &[u8],
    ) -> Result<(), QueueWriteError> {
        profiling::scope!("Queue::write_buffer");
        api_log!("Queue::write_buffer {buffer_id:?} {}bytes", data.len());

        let hub = A::hub(self);

        let buffer_device_id = hub
            .buffers
            .get(buffer_id)
            .map_err(|_| TransferError::InvalidBuffer(buffer_id))?
            .device
            .as_info()
            .id();

        let queue = hub
            .queues
            .get(queue_id)
            .map_err(|_| DeviceError::InvalidQueueId)?;

        let device = queue.device.as_ref().unwrap();

        {
            let queue_device_id = device.as_info().id();
            if buffer_device_id != queue_device_id {
                return Err(QueueWriteError::DeviceMismatch {
                    queue_device_id,
                    target_device_id: buffer_device_id,
                });
            }
        }

        let data_size = data.len() as wgt::BufferAddress;

        #[cfg(feature = "trace")]
        if let Some(ref mut trace) = *device.trace.lock() {
            let data_path = trace.make_binary("bin", data);
            trace.add(Action::WriteBuffer {
                id: buffer_id,
                data: data_path,
                range: buffer_offset..buffer_offset + data_size,
                queued: true,
            });
        }

        if data_size == 0 {
            log::trace!("Ignoring write_buffer of size 0");
            return Ok(());
        }

        // Platform validation requires that the staging buffer always be
        // freed, even if an error occurs. All paths from here must call
        // `device.pending_writes.consume`.
        let (staging_buffer, staging_buffer_ptr) =
            prepare_staging_buffer(device, data_size, device.instance_flags)?;
        let mut pending_writes = device.pending_writes.lock();
        let pending_writes = pending_writes.as_mut().unwrap();

        let stage_fid = hub.staging_buffers.request();
        let staging_buffer = stage_fid.init(staging_buffer);

        if let Err(flush_error) = unsafe {
            profiling::scope!("copy");
            ptr::copy_nonoverlapping(data.as_ptr(), staging_buffer_ptr, data.len());
            staging_buffer.flush(device.raw())
        } {
            pending_writes.consume(staging_buffer);
            return Err(flush_error.into());
        }

        let result = self.queue_write_staging_buffer_impl(
            device,
            pending_writes,
            &staging_buffer,
            buffer_id,
            buffer_offset,
        );

        pending_writes.consume(staging_buffer);
        result
    }

    pub fn queue_create_staging_buffer<A: HalApi>(
        &self,
        queue_id: QueueId,
        buffer_size: wgt::BufferSize,
        id_in: Option<id::StagingBufferId>,
    ) -> Result<(id::StagingBufferId, *mut u8), QueueWriteError> {
        profiling::scope!("Queue::create_staging_buffer");
        let hub = A::hub(self);

        let queue = hub
            .queues
            .get(queue_id)
            .map_err(|_| DeviceError::InvalidQueueId)?;

        let device = queue.device.as_ref().unwrap();

        let (staging_buffer, staging_buffer_ptr) =
            prepare_staging_buffer(device, buffer_size.get(), device.instance_flags)?;

        let fid = hub.staging_buffers.prepare(id_in);
        let (id, _) = fid.assign(Arc::new(staging_buffer));
        resource_log!("Queue::create_staging_buffer {id:?}");

        Ok((id, staging_buffer_ptr))
    }

    pub fn queue_write_staging_buffer<A: HalApi>(
        &self,
        queue_id: QueueId,
        buffer_id: id::BufferId,
        buffer_offset: wgt::BufferAddress,
        staging_buffer_id: id::StagingBufferId,
    ) -> Result<(), QueueWriteError> {
        profiling::scope!("Queue::write_staging_buffer");
        let hub = A::hub(self);

        let queue = hub
            .queues
            .get(queue_id)
            .map_err(|_| DeviceError::InvalidQueueId)?;

        let device = queue.device.as_ref().unwrap();

        let staging_buffer = hub.staging_buffers.unregister(staging_buffer_id);
        if staging_buffer.is_none() {
            return Err(QueueWriteError::Transfer(TransferError::InvalidBuffer(
                buffer_id,
            )));
        }
        let staging_buffer = staging_buffer.unwrap();
        let mut pending_writes = device.pending_writes.lock();
        let pending_writes = pending_writes.as_mut().unwrap();

        // At this point, we have taken ownership of the staging_buffer from the
        // user. Platform validation requires that the staging buffer always
        // be freed, even if an error occurs. All paths from here must call
        // `device.pending_writes.consume`.
        if let Err(flush_error) = unsafe { staging_buffer.flush(device.raw()) } {
            pending_writes.consume(staging_buffer);
            return Err(flush_error.into());
        }

        let result = self.queue_write_staging_buffer_impl(
            device,
            pending_writes,
            &staging_buffer,
            buffer_id,
            buffer_offset,
        );

        pending_writes.consume(staging_buffer);
        result
    }

    pub fn queue_validate_write_buffer<A: HalApi>(
        &self,
        _queue_id: QueueId,
        buffer_id: id::BufferId,
        buffer_offset: u64,
        buffer_size: u64,
    ) -> Result<(), QueueWriteError> {
        profiling::scope!("Queue::validate_write_buffer");
        let hub = A::hub(self);

        let buffer = hub
            .buffers
            .get(buffer_id)
            .map_err(|_| TransferError::InvalidBuffer(buffer_id))?;

        self.queue_validate_write_buffer_impl(&buffer, buffer_id, buffer_offset, buffer_size)?;

        Ok(())
    }

    fn queue_validate_write_buffer_impl<A: HalApi>(
        &self,
        buffer: &Buffer<A>,
        buffer_id: id::BufferId,
        buffer_offset: u64,
        buffer_size: u64,
    ) -> Result<(), TransferError> {
        if !buffer.usage.contains(wgt::BufferUsages::COPY_DST) {
            return Err(TransferError::MissingCopyDstUsageFlag(
                Some(buffer_id),
                None,
            ));
        }
        if buffer_size % wgt::COPY_BUFFER_ALIGNMENT != 0 {
            return Err(TransferError::UnalignedCopySize(buffer_size));
        }
        if buffer_offset % wgt::COPY_BUFFER_ALIGNMENT != 0 {
            return Err(TransferError::UnalignedBufferOffset(buffer_offset));
        }
        if buffer_offset + buffer_size > buffer.size {
            return Err(TransferError::BufferOverrun {
                start_offset: buffer_offset,
                end_offset: buffer_offset + buffer_size,
                buffer_size: buffer.size,
                side: CopySide::Destination,
            });
        }

        Ok(())
    }

    fn queue_write_staging_buffer_impl<A: HalApi>(
        &self,
        device: &Device<A>,
        pending_writes: &mut PendingWrites<A>,
        staging_buffer: &StagingBuffer<A>,
        buffer_id: id::BufferId,
        buffer_offset: u64,
    ) -> Result<(), QueueWriteError> {
        let hub = A::hub(self);

        let (dst, transition) = {
            let buffer_guard = hub.buffers.read();
            let dst = buffer_guard
                .get(buffer_id)
                .map_err(|_| TransferError::InvalidBuffer(buffer_id))?;
            let mut trackers = device.trackers.lock();
            trackers
                .buffers
                .set_single(dst, hal::BufferUses::COPY_DST)
                .ok_or(TransferError::InvalidBuffer(buffer_id))?
        };
        let snatch_guard = device.snatchable_lock.read();
        let dst_raw = dst
            .raw
            .get(&snatch_guard)
            .ok_or(TransferError::InvalidBuffer(buffer_id))?;

        if dst.device.as_info().id() != device.as_info().id() {
            return Err(DeviceError::WrongDevice.into());
        }

        let src_buffer_size = staging_buffer.size;
        self.queue_validate_write_buffer_impl(&dst, buffer_id, buffer_offset, src_buffer_size)?;

        dst.info
            .use_at(device.active_submission_index.load(Ordering::Relaxed) + 1);

        let region = wgt::BufferSize::new(src_buffer_size).map(|size| hal::BufferCopy {
            src_offset: 0,
            dst_offset: buffer_offset,
            size,
        });
        let inner_buffer = staging_buffer.raw.lock();
        let barriers = iter::once(hal::BufferBarrier {
            buffer: inner_buffer.as_ref().unwrap(),
            usage: hal::BufferUses::MAP_WRITE..hal::BufferUses::COPY_SRC,
        })
        .chain(transition.map(|pending| pending.into_hal(&dst, &snatch_guard)));
        let encoder = pending_writes.activate();
        unsafe {
            encoder.transition_buffers(barriers);
            encoder.copy_buffer_to_buffer(
                inner_buffer.as_ref().unwrap(),
                dst_raw,
                region.into_iter(),
            );
        }
        let dst = hub.buffers.get(buffer_id).unwrap();
        pending_writes.dst_buffers.insert(buffer_id, dst.clone());

        // Ensure the overwritten bytes are marked as initialized so
        // they don't need to be nulled prior to mapping or binding.
        {
            dst.initialization_status
                .write()
                .drain(buffer_offset..(buffer_offset + src_buffer_size));
        }

        Ok(())
    }

    pub fn queue_write_texture<A: HalApi>(
        &self,
        queue_id: QueueId,
        destination: &ImageCopyTexture,
        data: &[u8],
        data_layout: &wgt::ImageDataLayout,
        size: &wgt::Extent3d,
    ) -> Result<(), QueueWriteError> {
        profiling::scope!("Queue::write_texture");
        api_log!("Queue::write_texture {:?} {size:?}", destination.texture);

        let hub = A::hub(self);

        let queue = hub
            .queues
            .get(queue_id)
            .map_err(|_| DeviceError::InvalidQueueId)?;

        let device = queue.device.as_ref().unwrap();

        #[cfg(feature = "trace")]
        if let Some(ref mut trace) = *device.trace.lock() {
            let data_path = trace.make_binary("bin", data);
            trace.add(Action::WriteTexture {
                to: *destination,
                data: data_path,
                layout: *data_layout,
                size: *size,
            });
        }

        if size.width == 0 || size.height == 0 || size.depth_or_array_layers == 0 {
            log::trace!("Ignoring write_texture of size 0");
            return Ok(());
        }

        let dst = hub
            .textures
            .get(destination.texture)
            .map_err(|_| TransferError::InvalidTexture(destination.texture))?;

        if dst.device.as_info().id().into_queue_id() != queue_id {
            return Err(DeviceError::WrongDevice.into());
        }

        if !dst.desc.usage.contains(wgt::TextureUsages::COPY_DST) {
            return Err(
                TransferError::MissingCopyDstUsageFlag(None, Some(destination.texture)).into(),
            );
        }

        // Note: Doing the copy range validation early is important because ensures that the
        // dimensions are not going to cause overflow in other parts of the validation.
        let (hal_copy_size, array_layer_count) =
            validate_texture_copy_range(destination, &dst.desc, CopySide::Destination, size)?;

        let (selector, dst_base) = extract_texture_selector(destination, size, &dst)?;

        if !dst_base.aspect.is_one() {
            return Err(TransferError::CopyAspectNotOne.into());
        }

        if !conv::is_valid_copy_dst_texture_format(dst.desc.format, destination.aspect) {
            return Err(TransferError::CopyToForbiddenTextureFormat {
                format: dst.desc.format,
                aspect: destination.aspect,
            }
            .into());
        }

        // Note: `_source_bytes_per_array_layer` is ignored since we
        // have a staging copy, and it can have a different value.
        let (_, _source_bytes_per_array_layer) = validate_linear_texture_data(
            data_layout,
            dst.desc.format,
            destination.aspect,
            data.len() as wgt::BufferAddress,
            CopySide::Source,
            size,
            false,
        )?;

        if dst.desc.format.is_depth_stencil_format() {
            device
                .require_downlevel_flags(wgt::DownlevelFlags::DEPTH_TEXTURE_AND_BUFFER_COPIES)
                .map_err(TransferError::from)?;
        }

        let (block_width, block_height) = dst.desc.format.block_dimensions();
        let width_blocks = size.width / block_width;
        let height_blocks = size.height / block_height;

        let block_rows_per_image = data_layout.rows_per_image.unwrap_or(
            // doesn't really matter because we need this only if we copy
            // more than one layer, and then we validate for this being not
            // None
            height_blocks,
        );

        let block_size = dst
            .desc
            .format
            .block_copy_size(Some(destination.aspect))
            .unwrap();
        let bytes_per_row_alignment =
            get_lowest_common_denom(device.alignments.buffer_copy_pitch.get() as u32, block_size);
        let stage_bytes_per_row =
            wgt::math::align_to(block_size * width_blocks, bytes_per_row_alignment);

        let block_rows_in_copy =
            (size.depth_or_array_layers - 1) * block_rows_per_image + height_blocks;
        let stage_size = stage_bytes_per_row as u64 * block_rows_in_copy as u64;

        let mut pending_writes = device.pending_writes.lock();
        let pending_writes = pending_writes.as_mut().unwrap();
        let encoder = pending_writes.activate();

        // If the copy does not fully cover the layers, we need to initialize to
        // zero *first* as we don't keep track of partial texture layer inits.
        //
        // Strictly speaking we only need to clear the areas of a layer
        // untouched, but this would get increasingly messy.
        let init_layer_range = if dst.desc.dimension == wgt::TextureDimension::D3 {
            // volume textures don't have a layer range as array volumes aren't supported
            0..1
        } else {
            destination.origin.z..destination.origin.z + size.depth_or_array_layers
        };
        let mut dst_initialization_status = dst.initialization_status.write();
        if dst_initialization_status.mips[destination.mip_level as usize]
            .check(init_layer_range.clone())
            .is_some()
        {
            if has_copy_partial_init_tracker_coverage(size, destination.mip_level, &dst.desc) {
                for layer_range in dst_initialization_status.mips[destination.mip_level as usize]
                    .drain(init_layer_range)
                    .collect::<Vec<std::ops::Range<u32>>>()
                {
                    let mut trackers = device.trackers.lock();
                    crate::command::clear_texture(
                        &dst,
                        TextureInitRange {
                            mip_range: destination.mip_level..(destination.mip_level + 1),
                            layer_range,
                        },
                        encoder,
                        &mut trackers.textures,
                        &device.alignments,
                        device.zero_buffer.as_ref().unwrap(),
                        &device.snatchable_lock.read(),
                    )
                    .map_err(QueueWriteError::from)?;
                }
            } else {
                dst_initialization_status.mips[destination.mip_level as usize]
                    .drain(init_layer_range);
            }
        }

        let snatch_guard = device.snatchable_lock.read();

        // Re-get `dst` immutably here, so that the mutable borrow of the
        // `texture_guard.get` above ends in time for the `clear_texture`
        // call above. Since we've held `texture_guard` the whole time, we know
        // the texture hasn't gone away in the mean time, so we can unwrap.
        let dst = hub.textures.get(destination.texture).unwrap();
        dst.info
            .use_at(device.active_submission_index.load(Ordering::Relaxed) + 1);

        let dst_raw = dst
            .raw(&snatch_guard)
            .ok_or(TransferError::InvalidTexture(destination.texture))?;

        let bytes_per_row = data_layout
            .bytes_per_row
            .unwrap_or(width_blocks * block_size);

        // Platform validation requires that the staging buffer always be
        // freed, even if an error occurs. All paths from here must call
        // `device.pending_writes.consume`.
        let (staging_buffer, staging_buffer_ptr) =
            prepare_staging_buffer(device, stage_size, device.instance_flags)?;

        let stage_fid = hub.staging_buffers.request();
        let staging_buffer = stage_fid.init(staging_buffer);

        if stage_bytes_per_row == bytes_per_row {
            profiling::scope!("copy aligned");
            // Fast path if the data is already being aligned optimally.
            unsafe {
                ptr::copy_nonoverlapping(
                    data.as_ptr().offset(data_layout.offset as isize),
                    staging_buffer_ptr,
                    stage_size as usize,
                );
            }
        } else {
            profiling::scope!("copy chunked");
            // Copy row by row into the optimal alignment.
            let copy_bytes_per_row = stage_bytes_per_row.min(bytes_per_row) as usize;
            for layer in 0..size.depth_or_array_layers {
                let rows_offset = layer * block_rows_per_image;
                for row in 0..height_blocks {
                    unsafe {
                        ptr::copy_nonoverlapping(
                            data.as_ptr().offset(
                                data_layout.offset as isize
                                    + (rows_offset + row) as isize * bytes_per_row as isize,
                            ),
                            staging_buffer_ptr.offset(
                                (rows_offset + row) as isize * stage_bytes_per_row as isize,
                            ),
                            copy_bytes_per_row,
                        );
                    }
                }
            }
        }

        if let Err(e) = unsafe { staging_buffer.flush(device.raw()) } {
            pending_writes.consume(staging_buffer);
            return Err(e.into());
        }

        let regions = (0..array_layer_count).map(|rel_array_layer| {
            let mut texture_base = dst_base.clone();
            texture_base.array_layer += rel_array_layer;
            hal::BufferTextureCopy {
                buffer_layout: wgt::ImageDataLayout {
                    offset: rel_array_layer as u64
                        * block_rows_per_image as u64
                        * stage_bytes_per_row as u64,
                    bytes_per_row: Some(stage_bytes_per_row),
                    rows_per_image: Some(block_rows_per_image),
                },
                texture_base,
                size: hal_copy_size,
            }
        });

        {
            let inner_buffer = staging_buffer.raw.lock();
            let barrier = hal::BufferBarrier {
                buffer: inner_buffer.as_ref().unwrap(),
                usage: hal::BufferUses::MAP_WRITE..hal::BufferUses::COPY_SRC,
            };

            let mut trackers = device.trackers.lock();
            let transition = trackers
                .textures
                .set_single(&dst, selector, hal::TextureUses::COPY_DST)
                .ok_or(TransferError::InvalidTexture(destination.texture))?;
            unsafe {
                encoder.transition_textures(transition.map(|pending| pending.into_hal(dst_raw)));
                encoder.transition_buffers(iter::once(barrier));
                encoder.copy_buffer_to_texture(inner_buffer.as_ref().unwrap(), dst_raw, regions);
            }
        }

        pending_writes.consume(staging_buffer);
        pending_writes
            .dst_textures
            .insert(destination.texture, dst.clone());

        Ok(())
    }

    #[cfg(webgl)]
    pub fn queue_copy_external_image_to_texture<A: HalApi>(
        &self,
        queue_id: QueueId,
        source: &wgt::ImageCopyExternalImage,
        destination: crate::command::ImageCopyTextureTagged,
        size: wgt::Extent3d,
    ) -> Result<(), QueueWriteError> {
        profiling::scope!("Queue::copy_external_image_to_texture");

        let hub = A::hub(self);

        let queue = hub
            .queues
            .get(queue_id)
            .map_err(|_| DeviceError::InvalidQueueId)?;

        let device = queue.device.as_ref().unwrap();

        if size.width == 0 || size.height == 0 || size.depth_or_array_layers == 0 {
            log::trace!("Ignoring write_texture of size 0");
            return Ok(());
        }

        let mut needs_flag = false;
        needs_flag |= matches!(source.source, wgt::ExternalImageSource::OffscreenCanvas(_));
        needs_flag |= source.origin != wgt::Origin2d::ZERO;
        needs_flag |= destination.color_space != wgt::PredefinedColorSpace::Srgb;
        #[allow(clippy::bool_comparison)]
        if matches!(source.source, wgt::ExternalImageSource::ImageBitmap(_)) {
            needs_flag |= source.flip_y != false;
            needs_flag |= destination.premultiplied_alpha != false;
        }

        if needs_flag {
            device
                .require_downlevel_flags(wgt::DownlevelFlags::UNRESTRICTED_EXTERNAL_TEXTURE_COPIES)
                .map_err(TransferError::from)?;
        }

        let src_width = source.source.width();
        let src_height = source.source.height();

        let dst = hub.textures.get(destination.texture).unwrap();

        if !conv::is_valid_external_image_copy_dst_texture_format(dst.desc.format) {
            return Err(
                TransferError::ExternalCopyToForbiddenTextureFormat(dst.desc.format).into(),
            );
        }
        if dst.desc.dimension != wgt::TextureDimension::D2 {
            return Err(TransferError::InvalidDimensionExternal(destination.texture).into());
        }
        if !dst.desc.usage.contains(wgt::TextureUsages::COPY_DST) {
            return Err(
                TransferError::MissingCopyDstUsageFlag(None, Some(destination.texture)).into(),
            );
        }
        if !dst
            .desc
            .usage
            .contains(wgt::TextureUsages::RENDER_ATTACHMENT)
        {
            return Err(
                TransferError::MissingRenderAttachmentUsageFlag(destination.texture).into(),
            );
        }
        if dst.desc.sample_count != 1 {
            return Err(TransferError::InvalidSampleCount {
                sample_count: dst.desc.sample_count,
            }
            .into());
        }

        if source.origin.x + size.width > src_width {
            return Err(TransferError::TextureOverrun {
                start_offset: source.origin.x,
                end_offset: source.origin.x + size.width,
                texture_size: src_width,
                dimension: crate::resource::TextureErrorDimension::X,
                side: CopySide::Source,
            }
            .into());
        }
        if source.origin.y + size.height > src_height {
            return Err(TransferError::TextureOverrun {
                start_offset: source.origin.y,
                end_offset: source.origin.y + size.height,
                texture_size: src_height,
                dimension: crate::resource::TextureErrorDimension::Y,
                side: CopySide::Source,
            }
            .into());
        }
        if size.depth_or_array_layers != 1 {
            return Err(TransferError::TextureOverrun {
                start_offset: 0,
                end_offset: size.depth_or_array_layers,
                texture_size: 1,
                dimension: crate::resource::TextureErrorDimension::Z,
                side: CopySide::Source,
            }
            .into());
        }

        // Note: Doing the copy range validation early is important because ensures that the
        // dimensions are not going to cause overflow in other parts of the validation.
        let (hal_copy_size, _) = validate_texture_copy_range(
            &destination.to_untagged(),
            &dst.desc,
            CopySide::Destination,
            &size,
        )?;

        let (selector, dst_base) =
            extract_texture_selector(&destination.to_untagged(), &size, &dst)?;

        let mut pending_writes = device.pending_writes.lock();
        let encoder = pending_writes.as_mut().unwrap().activate();

        // If the copy does not fully cover the layers, we need to initialize to
        // zero *first* as we don't keep track of partial texture layer inits.
        //
        // Strictly speaking we only need to clear the areas of a layer
        // untouched, but this would get increasingly messy.
        let init_layer_range = if dst.desc.dimension == wgt::TextureDimension::D3 {
            // volume textures don't have a layer range as array volumes aren't supported
            0..1
        } else {
            destination.origin.z..destination.origin.z + size.depth_or_array_layers
        };
        let mut dst_initialization_status = dst.initialization_status.write();
        if dst_initialization_status.mips[destination.mip_level as usize]
            .check(init_layer_range.clone())
            .is_some()
        {
            if has_copy_partial_init_tracker_coverage(&size, destination.mip_level, &dst.desc) {
                for layer_range in dst_initialization_status.mips[destination.mip_level as usize]
                    .drain(init_layer_range)
                    .collect::<Vec<std::ops::Range<u32>>>()
                {
                    let mut trackers = device.trackers.lock();
                    crate::command::clear_texture(
                        &dst,
                        TextureInitRange {
                            mip_range: destination.mip_level..(destination.mip_level + 1),
                            layer_range,
                        },
                        encoder,
                        &mut trackers.textures,
                        &device.alignments,
                        device.zero_buffer.as_ref().unwrap(),
                        &device.snatchable_lock.read(),
                    )
                    .map_err(QueueWriteError::from)?;
                }
            } else {
                dst_initialization_status.mips[destination.mip_level as usize]
                    .drain(init_layer_range);
            }
        }
        dst.info
            .use_at(device.active_submission_index.load(Ordering::Relaxed) + 1);

        let snatch_guard = device.snatchable_lock.read();
        let dst_raw = dst
            .raw(&snatch_guard)
            .ok_or(TransferError::InvalidTexture(destination.texture))?;

        let regions = hal::TextureCopy {
            src_base: hal::TextureCopyBase {
                mip_level: 0,
                array_layer: 0,
                origin: source.origin.to_3d(0),
                aspect: hal::FormatAspects::COLOR,
            },
            dst_base,
            size: hal_copy_size,
        };

        unsafe {
            let mut trackers = device.trackers.lock();
            let transitions = trackers
                .textures
                .set_single(&dst, selector, hal::TextureUses::COPY_DST)
                .ok_or(TransferError::InvalidTexture(destination.texture))?;
            encoder.transition_textures(transitions.map(|pending| pending.into_hal(dst_raw)));
            encoder.copy_external_image_to_texture(
                source,
                dst_raw,
                destination.premultiplied_alpha,
                iter::once(regions),
            );
        }

        Ok(())
    }

    pub fn queue_submit<A: HalApi>(
        &self,
        queue_id: QueueId,
        command_buffer_ids: &[id::CommandBufferId],
    ) -> Result<WrappedSubmissionIndex, QueueSubmitError> {
        profiling::scope!("Queue::submit");
        api_log!("Queue::submit {queue_id:?}");

        let (submit_index, callbacks) = {
            let hub = A::hub(self);

            let queue = hub
                .queues
                .get(queue_id)
                .map_err(|_| DeviceError::InvalidQueueId)?;

            let device = queue.device.as_ref().unwrap();

            let snatch_guard = device.snatchable_lock.read();

            // Fence lock must be acquired after the snatch lock everywhere to avoid deadlocks.
            let mut fence = device.fence.write();
            let fence = fence.as_mut().unwrap();
            let submit_index = device
                .active_submission_index
                .fetch_add(1, Ordering::Relaxed)
                + 1;
            let mut active_executions = Vec::new();

            let mut used_surface_textures = track::TextureUsageScope::default();

            // Use a hashmap here to deduplicate the surface textures that are used in the command buffers.
            // This avoids vulkan deadlocking from the same surface texture being submitted multiple times.
            let mut submit_surface_textures_owned = FastHashMap::default();

            {
                let mut command_buffer_guard = hub.command_buffers.write();

                if !command_buffer_ids.is_empty() {
                    profiling::scope!("prepare");

                    //TODO: if multiple command buffers are submitted, we can re-use the last
                    // native command buffer of the previous chain instead of always creating
                    // a temporary one, since the chains are not finished.
                    let mut temp_suspected = device.temp_suspected.lock();
                    {
                        let mut suspected = temp_suspected.replace(ResourceMaps::new()).unwrap();
                        suspected.clear();
                    }

                    // finish all the command buffers first
                    for &cmb_id in command_buffer_ids {
                        // we reset the used surface textures every time we use
                        // it, so make sure to set_size on it.
                        used_surface_textures.set_size(device.tracker_indices.textures.size());

                        #[allow(unused_mut)]
                        let mut cmdbuf = match command_buffer_guard.replace_with_error(cmb_id) {
                            Ok(cmdbuf) => cmdbuf,
                            Err(_) => continue,
                        };

                        if cmdbuf.device.as_info().id().into_queue_id() != queue_id {
                            return Err(DeviceError::WrongDevice.into());
                        }

                        #[cfg(feature = "trace")]
                        if let Some(ref mut trace) = *device.trace.lock() {
                            trace.add(Action::Submit(
                                submit_index,
                                cmdbuf
                                    .data
                                    .lock()
                                    .as_mut()
                                    .unwrap()
                                    .commands
                                    .take()
                                    .unwrap(),
                            ));
                        }
                        if !cmdbuf.is_finished() {
                            let cmdbuf = Arc::into_inner(cmdbuf).expect(
                                "Command buffer cannot be destroyed because is still in use",
                            );
                            device.destroy_command_buffer(cmdbuf);
                            continue;
                        }

                        // optimize the tracked states
                        // cmdbuf.trackers.optimize();
                        {
                            let cmd_buf_data = cmdbuf.data.lock();
                            let cmd_buf_trackers = &cmd_buf_data.as_ref().unwrap().trackers;

                            // update submission IDs
                            for buffer in cmd_buf_trackers.buffers.used_resources() {
                                let tracker_index = buffer.info.tracker_index();
                                let raw_buf = match buffer.raw.get(&snatch_guard) {
                                    Some(raw) => raw,
                                    None => {
                                        return Err(QueueSubmitError::DestroyedBuffer(
                                            buffer.info.id(),
                                        ));
                                    }
                                };
                                buffer.info.use_at(submit_index);
                                if buffer.is_unique() {
                                    if let BufferMapState::Active { .. } = *buffer.map_state.lock()
                                    {
                                        log::warn!("Dropped buffer has a pending mapping.");
                                        unsafe { device.raw().unmap_buffer(raw_buf) }
                                            .map_err(DeviceError::from)?;
                                    }
                                    temp_suspected
                                        .as_mut()
                                        .unwrap()
                                        .buffers
                                        .insert(tracker_index, buffer.clone());
                                } else {
                                    match *buffer.map_state.lock() {
                                        BufferMapState::Idle => (),
                                        _ => {
                                            return Err(QueueSubmitError::BufferStillMapped(
                                                buffer.info.id(),
                                            ))
                                        }
                                    }
                                }
                            }
                            for texture in cmd_buf_trackers.textures.used_resources() {
                                let tracker_index = texture.info.tracker_index();
                                let should_extend = match texture.inner.get(&snatch_guard) {
                                    None => {
                                        return Err(QueueSubmitError::DestroyedTexture(
                                            texture.info.id(),
                                        ));
                                    }
                                    Some(TextureInner::Native { .. }) => false,
                                    Some(TextureInner::Surface { ref raw, .. }) => {
                                        if raw.is_some() {
                                            // Compare the Arcs by pointer as Textures don't implement Eq.
                                            submit_surface_textures_owned
                                                .insert(Arc::as_ptr(&texture), texture.clone());
                                        }

                                        true
                                    }
                                };
                                texture.info.use_at(submit_index);
                                if texture.is_unique() {
                                    temp_suspected
                                        .as_mut()
                                        .unwrap()
                                        .textures
                                        .insert(tracker_index, texture.clone());
                                }
                                if should_extend {
                                    unsafe {
                                        used_surface_textures
                                            .merge_single(&texture, None, hal::TextureUses::PRESENT)
                                            .unwrap();
                                    };
                                }
                            }
                            for texture_view in cmd_buf_trackers.views.used_resources() {
                                texture_view.info.use_at(submit_index);
                                if texture_view.is_unique() {
                                    temp_suspected.as_mut().unwrap().texture_views.insert(
                                        texture_view.as_info().tracker_index(),
                                        texture_view.clone(),
                                    );
                                }
                            }
                            {
                                for bg in cmd_buf_trackers.bind_groups.used_resources() {
                                    bg.info.use_at(submit_index);
                                    // We need to update the submission indices for the contained
                                    // state-less (!) resources as well, so that they don't get
                                    // deleted too early if the parent bind group goes out of scope.
                                    for view in bg.used.views.used_resources() {
                                        view.info.use_at(submit_index);
                                    }
                                    for sampler in bg.used.samplers.used_resources() {
                                        sampler.info.use_at(submit_index);
                                    }
                                    if bg.is_unique() {
                                        temp_suspected
                                            .as_mut()
                                            .unwrap()
                                            .bind_groups
                                            .insert(bg.as_info().tracker_index(), bg.clone());
                                    }
                                }
                            }
                            // assert!(cmd_buf_trackers.samplers.is_empty());
                            for compute_pipeline in
                                cmd_buf_trackers.compute_pipelines.used_resources()
                            {
                                compute_pipeline.info.use_at(submit_index);
                                if compute_pipeline.is_unique() {
                                    temp_suspected.as_mut().unwrap().compute_pipelines.insert(
                                        compute_pipeline.as_info().tracker_index(),
                                        compute_pipeline.clone(),
                                    );
                                }
                            }
                            for render_pipeline in
                                cmd_buf_trackers.render_pipelines.used_resources()
                            {
                                render_pipeline.info.use_at(submit_index);
                                if render_pipeline.is_unique() {
                                    temp_suspected.as_mut().unwrap().render_pipelines.insert(
                                        render_pipeline.as_info().tracker_index(),
                                        render_pipeline.clone(),
                                    );
                                }
                            }
                            for query_set in cmd_buf_trackers.query_sets.used_resources() {
                                query_set.info.use_at(submit_index);
                                if query_set.is_unique() {
                                    temp_suspected.as_mut().unwrap().query_sets.insert(
                                        query_set.as_info().tracker_index(),
                                        query_set.clone(),
                                    );
                                }
                            }
                            for bundle in cmd_buf_trackers.bundles.used_resources() {
                                bundle.info.use_at(submit_index);
                                // We need to update the submission indices for the contained
                                // state-less (!) resources as well, excluding the bind groups.
                                // They don't get deleted too early if the bundle goes out of scope.
                                for render_pipeline in
                                    bundle.used.render_pipelines.read().used_resources()
                                {
                                    render_pipeline.info.use_at(submit_index);
                                }
                                for query_set in bundle.used.query_sets.read().used_resources() {
                                    query_set.info.use_at(submit_index);
                                }
                                if bundle.is_unique() {
                                    temp_suspected
                                        .as_mut()
                                        .unwrap()
                                        .render_bundles
                                        .insert(bundle.as_info().tracker_index(), bundle.clone());
                                }
                            }
                        }
                        let mut baked = cmdbuf.from_arc_into_baked();
                        // execute resource transitions
                        unsafe {
                            baked
                                .encoder
                                .begin_encoding(hal_label(
                                    Some("(wgpu internal) Transit"),
                                    device.instance_flags,
                                ))
                                .map_err(DeviceError::from)?
                        };
                        log::trace!("Stitching command buffer {:?} before submission", cmb_id);

                        //Note: locking the trackers has to be done after the storages
                        let mut trackers = device.trackers.lock();
                        baked
                            .initialize_buffer_memory(&mut *trackers, &snatch_guard)
                            .map_err(|err| QueueSubmitError::DestroyedBuffer(err.0))?;
                        baked
                            .initialize_texture_memory(&mut *trackers, device, &snatch_guard)
                            .map_err(|err| QueueSubmitError::DestroyedTexture(err.0))?;
                        //Note: stateless trackers are not merged:
                        // device already knows these resources exist.
                        CommandBuffer::insert_barriers_from_tracker(
                            &mut baked.encoder,
                            &mut *trackers,
                            &baked.trackers,
                            &snatch_guard,
                        );

                        let transit = unsafe { baked.encoder.end_encoding().unwrap() };
                        baked.list.insert(0, transit);

                        // Transition surface textures into `Present` state.
                        // Note: we could technically do it after all of the command buffers,
                        // but here we have a command encoder by hand, so it's easier to use it.
                        if !used_surface_textures.is_empty() {
                            unsafe {
                                baked
                                    .encoder
                                    .begin_encoding(hal_label(
                                        Some("(wgpu internal) Present"),
                                        device.instance_flags,
                                    ))
                                    .map_err(DeviceError::from)?
                            };
                            trackers
                                .textures
                                .set_from_usage_scope(&used_surface_textures);
                            let (transitions, textures) =
                                trackers.textures.drain_transitions(&snatch_guard);
                            let texture_barriers = transitions
                                .into_iter()
                                .enumerate()
                                .map(|(i, p)| p.into_hal(textures[i].unwrap().raw().unwrap()));
                            let present = unsafe {
                                baked.encoder.transition_textures(texture_barriers);
                                baked.encoder.end_encoding().unwrap()
                            };
                            baked.list.push(present);
                            used_surface_textures = track::TextureUsageScope::default();
                        }

                        // done
                        active_executions.push(EncoderInFlight {
                            raw: baked.encoder,
                            cmd_buffers: baked.list,
                        });
                    }

                    log::trace!("Device after submission {}", submit_index);
                }
            }

            let mut pending_writes = device.pending_writes.lock();
            let pending_writes = pending_writes.as_mut().unwrap();

            {
                used_surface_textures.set_size(hub.textures.read().len());
                for (&id, texture) in pending_writes.dst_textures.iter() {
                    match texture.inner.get(&snatch_guard) {
                        None => {
                            return Err(QueueSubmitError::DestroyedTexture(id));
                        }
                        Some(TextureInner::Native { .. }) => {}
                        Some(TextureInner::Surface { ref raw, .. }) => {
                            if raw.is_some() {
                                // Compare the Arcs by pointer as Textures don't implement Eq
                                submit_surface_textures_owned
                                    .insert(Arc::as_ptr(texture), texture.clone());
                            }

                            unsafe {
                                used_surface_textures
                                    .merge_single(texture, None, hal::TextureUses::PRESENT)
                                    .unwrap()
                            };
                        }
                    }
                }

                if !used_surface_textures.is_empty() {
                    let mut trackers = device.trackers.lock();

                    trackers
                        .textures
                        .set_from_usage_scope(&used_surface_textures);
                    let (transitions, textures) =
                        trackers.textures.drain_transitions(&snatch_guard);
                    let texture_barriers = transitions
                        .into_iter()
                        .enumerate()
                        .map(|(i, p)| p.into_hal(textures[i].unwrap().raw().unwrap()));
                    unsafe {
                        pending_writes
                            .command_encoder
                            .transition_textures(texture_barriers);
                    };
                }
            }

            let refs = pending_writes
                .pre_submit()?
                .into_iter()
                .chain(
                    active_executions
                        .iter()
                        .flat_map(|pool_execution| pool_execution.cmd_buffers.iter()),
                )
                .collect::<Vec<_>>();

            let mut submit_surface_textures =
                SmallVec::<[_; 2]>::with_capacity(submit_surface_textures_owned.len());

            for texture in submit_surface_textures_owned.values() {
                submit_surface_textures.extend(match texture.inner.get(&snatch_guard) {
                    Some(TextureInner::Surface { raw, .. }) => raw.as_ref(),
                    _ => None,
                });
            }

            unsafe {
                queue
                    .raw
                    .as_ref()
                    .unwrap()
                    .submit(&refs, &submit_surface_textures, (fence, submit_index))
                    .map_err(DeviceError::from)?;
            }

            profiling::scope!("cleanup");
            if let Some(pending_execution) = pending_writes.post_submit(
                &device.command_allocator,
                device.raw(),
                queue.raw.as_ref().unwrap(),
            ) {
                active_executions.push(pending_execution);
            }

            // this will register the new submission to the life time tracker
            let mut pending_write_resources = mem::take(&mut pending_writes.temp_resources);
            device.lock_life().track_submission(
                submit_index,
                pending_write_resources.drain(..),
                active_executions,
            );

            // This will schedule destruction of all resources that are no longer needed
            // by the user but used in the command stream, among other things.
            let (closures, _) = match device.maintain(fence, wgt::Maintain::Poll, snatch_guard) {
                Ok(closures) => closures,
                Err(WaitIdleError::Device(err)) => return Err(QueueSubmitError::Queue(err)),
                Err(WaitIdleError::StuckGpu) => return Err(QueueSubmitError::StuckGpu),
                Err(WaitIdleError::WrongSubmissionIndex(..)) => unreachable!(),
            };

            // pending_write_resources has been drained, so it's empty, but we
            // want to retain its heap allocation.
            pending_writes.temp_resources = pending_write_resources;
            device.lock_life().post_submit();

            (submit_index, closures)
        };

        // the closures should execute with nothing locked!
        callbacks.fire();

        Ok(WrappedSubmissionIndex {
            queue_id,
            index: submit_index,
        })
    }

    pub fn queue_get_timestamp_period<A: HalApi>(
        &self,
        queue_id: QueueId,
    ) -> Result<f32, InvalidQueue> {
        let hub = A::hub(self);
        match hub.queues.get(queue_id) {
            Ok(queue) => Ok(unsafe { queue.raw.as_ref().unwrap().get_timestamp_period() }),
            Err(_) => Err(InvalidQueue),
        }
    }

    pub fn queue_on_submitted_work_done<A: HalApi>(
        &self,
        queue_id: QueueId,
        closure: SubmittedWorkDoneClosure,
    ) -> Result<(), InvalidQueue> {
        api_log!("Queue::on_submitted_work_done {queue_id:?}");

        //TODO: flush pending writes
        let hub = A::hub(self);
        match hub.queues.get(queue_id) {
            Ok(queue) => queue
                .device
                .as_ref()
                .unwrap()
                .lock_life()
                .add_work_done_closure(closure),
            Err(_) => return Err(InvalidQueue),
        }
        Ok(())
    }
}