threers-codec 0.1.0

Pure-Rust media codecs: H.264/HEVC/VP9 encoders, H.264/HEVC/VP9/AV1 decoders, AAC/MP3/Opus/Vorbis/FLAC/WAV decoders, AAC-LC and MP3 encoders, MP4/WebM/Matroska/Ogg, GIF/APNG, captions, and VideoToolbox acceleration on macOS
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
//! HEVC parameter sets and the slice segment header (H.265 §7.3.2, §7.3.6).

use super::tables::{DEFAULT_SCALING, SCAN4, SCAN8X8};
use crate::decode::bits::BitReader;
use crate::decode::picture::check_picture_size;

pub type Result<T> = std::result::Result<T, String>;

fn ceil_log2(v: u32) -> u32 {
    if v <= 1 {
        0
    } else {
        32 - (v - 1).leading_zeros()
    }
}

/// Scaling factors per `sizeId` (4, 8, 16, 32) and `matrixId` (`cIdx` +
/// 3 for inter), expanded to raster order at full block size.
#[derive(Clone, Debug)]
pub struct ScalingFactors {
    /// `[sizeId][matrixId]`, `n * n` entries in raster order (`y * n + x`).
    pub m: [[Vec<u8>; 6]; 4],
}

/// Coded scaling lists, before expansion: per `sizeId`/`matrixId` the up to
/// 64 coefficients in diagonal order, and the DC for 16×16 and 32×32.
#[derive(Clone, Debug)]
struct ScalingLists {
    lists: [[[u8; 64]; 6]; 4],
    dc: [[u8; 6]; 4],
}

impl ScalingLists {
    fn default_lists() -> Self {
        let mut s = ScalingLists {
            lists: [[[16; 64]; 6]; 4],
            dc: [[16; 6]; 4],
        };
        for size in 1..4 {
            for m in 0..6 {
                s.lists[size][m] = DEFAULT_SCALING[(m >= 3) as usize];
            }
        }
        s
    }

    fn parse(r: &mut BitReader) -> Result<Self> {
        let mut s = Self::default_lists();
        for size in 0..4 {
            let step = if size == 3 { 3 } else { 1 };
            let mut m = 0;
            while m < 6 {
                if !r.flag() {
                    let delta = r.ue() as usize;
                    if delta != 0 {
                        let d = delta * step;
                        if d > m {
                            return Err("scaling_list_pred_matrix_id_delta out of range".into());
                        }
                        let rm = m - d;
                        s.lists[size][m] = s.lists[size][rm];
                        s.dc[size][m] = s.dc[size][rm];
                    } else {
                        // Default list.
                        s.lists[size][m] = if size == 0 {
                            [16; 64]
                        } else {
                            DEFAULT_SCALING[(m >= 3) as usize]
                        };
                        s.dc[size][m] = 16;
                    }
                } else {
                    let mut next = 8i32;
                    let n = (1usize << (4 + (size << 1))).min(64);
                    if size > 1 {
                        let dc = r.se().saturating_add(8);
                        if !(1..=255).contains(&dc) {
                            return Err("scaling_list_dc_coef out of range".into());
                        }
                        next = dc;
                        s.dc[size][m] = dc as u8;
                    }
                    for i in 0..n {
                        let delta = r.se();
                        // (next + delta + 256) % 256 for the legal −128..=127,
                        // without overflowing on garbage.
                        next = (next as i64 + delta as i64).rem_euclid(256) as i32;
                        s.lists[size][m][i] = next as u8;
                    }
                    if size <= 1 {
                        s.dc[size][m] = s.lists[size][m][0];
                    }
                }
                m += step;
            }
        }
        // 32×32 chroma lists (4:4:4 only) copy from 16×16 — keep them sane.
        for m in [1, 2, 4, 5] {
            s.lists[3][m] = s.lists[2][m];
            s.dc[3][m] = s.dc[2][m];
        }
        Ok(s)
    }

    fn expand(&self) -> ScalingFactors {
        let mut out: [[Vec<u8>; 6]; 4] = Default::default();
        for size in 0..4 {
            let n = 4usize << size;
            for m in 0..6 {
                let mut f = vec![0u8; n * n];
                if size == 0 {
                    for (i, &(x, y)) in SCAN4[0].iter().enumerate() {
                        f[y as usize * 4 + x as usize] = self.lists[0][m][i];
                    }
                } else {
                    let rep = n / 8;
                    for (i, &(x, y)) in SCAN8X8.iter().enumerate() {
                        for j in 0..rep {
                            for k in 0..rep {
                                let (xx, yy) = (x as usize * rep + k, y as usize * rep + j);
                                f[yy * n + xx] = self.lists[size][m][i];
                            }
                        }
                    }
                    if size >= 2 {
                        f[0] = self.dc[size][m];
                    }
                }
                out[size][m] = f;
            }
        }
        ScalingFactors { m: out }
    }
}

/// A short-term reference picture set (§7.4.8).
#[derive(Clone, Debug, Default)]
pub struct StRps {
    /// Negative deltas, closest first, with `used_by_curr_pic`.
    pub neg: Vec<(i32, bool)>,
    /// Positive deltas, closest first.
    pub pos: Vec<(i32, bool)>,
}

impl StRps {
    fn num_delta_pocs(&self) -> usize {
        self.neg.len() + self.pos.len()
    }

    /// Parse `st_ref_pic_set(idx)`. `sets` are those already parsed from
    /// the SPS. `in_slice` marks the set in a slice header (whose `idx` is
    /// `num_short_term_ref_pic_sets`) — the only one that codes
    /// `delta_idx_minus1`. While the SPS itself is being parsed `idx` also
    /// equals the number parsed so far, so the index cannot tell them apart.
    fn parse(r: &mut BitReader, idx: usize, sets: &[StRps], in_slice: bool) -> Result<StRps> {
        let inter = idx != 0 && r.flag();
        if inter {
            let delta_idx = if in_slice { r.ue() as usize + 1 } else { 1 };
            if delta_idx > idx {
                return Err("delta_idx_minus1 out of range".into());
            }
            let rref = &sets[idx - delta_idx];
            let sign = r.flag();
            let abs = r.ue() as i32 + 1;
            let delta_rps = if sign { -abs } else { abs };
            let n = rref.num_delta_pocs();
            let mut used = vec![false; n + 1];
            let mut use_delta = vec![true; n + 1];
            for j in 0..=n {
                used[j] = r.flag();
                if !used[j] {
                    use_delta[j] = r.flag();
                }
            }
            // §7.4.8, equations 7-61 and 7-62. Entries 0..neg are the
            // reference set's negative pictures, then its positive ones,
            // then the reference picture itself.
            let nn = rref.neg.len();
            let mut out = StRps::default();
            for j in (0..rref.pos.len()).rev() {
                let d = rref.pos[j].0 + delta_rps;
                if d < 0 && use_delta[nn + j] {
                    out.neg.push((d, used[nn + j]));
                }
            }
            if delta_rps < 0 && use_delta[n] {
                out.neg.push((delta_rps, used[n]));
            }
            for j in 0..nn {
                let d = rref.neg[j].0 + delta_rps;
                if d < 0 && use_delta[j] {
                    out.neg.push((d, used[j]));
                }
            }
            for j in (0..nn).rev() {
                let d = rref.neg[j].0 + delta_rps;
                if d > 0 && use_delta[j] {
                    out.pos.push((d, used[j]));
                }
            }
            if delta_rps > 0 && use_delta[n] {
                out.pos.push((delta_rps, used[n]));
            }
            for j in 0..rref.pos.len() {
                let d = rref.pos[j].0 + delta_rps;
                if d > 0 && use_delta[nn + j] {
                    out.pos.push((d, used[nn + j]));
                }
            }
            Ok(out)
        } else {
            let nneg = r.ue() as usize;
            let npos = r.ue() as usize;
            if nneg > 16 || npos > 16 {
                return Err("too many pictures in a reference picture set".into());
            }
            let mut out = StRps::default();
            let mut poc = 0;
            for _ in 0..nneg {
                poc -= r.ue() as i32 + 1;
                out.neg.push((poc, r.flag()));
            }
            poc = 0;
            for _ in 0..npos {
                poc += r.ue() as i32 + 1;
                out.pos.push((poc, r.flag()));
            }
            Ok(out)
        }
    }
}

fn profile_tier_level(r: &mut BitReader, max_sub_layers_minus1: u32) -> u32 {
    r.skip(2 + 1);
    let profile_idc = r.u(5);
    r.skip(32 + 4 + 43 + 1 + 8);
    let mut prof = [false; 8];
    let mut lvl = [false; 8];
    for i in 0..max_sub_layers_minus1 as usize {
        prof[i] = r.flag();
        lvl[i] = r.flag();
    }
    if max_sub_layers_minus1 > 0 {
        for _ in max_sub_layers_minus1..8 {
            r.skip(2);
        }
    }
    for i in 0..max_sub_layers_minus1 as usize {
        if prof[i] {
            r.skip(88);
        }
        if lvl[i] {
            r.skip(8);
        }
    }
    profile_idc
}

fn sub_layer_hrd(r: &mut BitReader, cpb_cnt: u32, sub_pic: bool) {
    for _ in 0..=cpb_cnt {
        r.ue();
        r.ue();
        if sub_pic {
            r.ue();
            r.ue();
        }
        r.skip(1);
    }
}

fn hrd_parameters(r: &mut BitReader, common: bool, max_sub_layers_minus1: u32) -> Result<()> {
    let mut nal = false;
    let mut vcl = false;
    let mut sub_pic = false;
    if common {
        nal = r.flag();
        vcl = r.flag();
        if nal || vcl {
            sub_pic = r.flag();
            if sub_pic {
                r.skip(8 + 5 + 1 + 5);
            }
            r.skip(4 + 4);
            if sub_pic {
                r.skip(4);
            }
            r.skip(5 + 5 + 5);
        }
    }
    for _ in 0..=max_sub_layers_minus1 {
        let general = r.flag();
        let within_cvs = if general { true } else { r.flag() };
        let mut low_delay = false;
        if within_cvs {
            r.ue();
        } else {
            low_delay = r.flag();
        }
        let cpb_cnt = if !low_delay { r.ue() } else { 0 };
        // Bounded by the spec, and a loop count below: damage here must not
        // become four billion iterations.
        if cpb_cnt > 31 {
            return Err("cpb_cnt_minus1 out of range".into());
        }
        if nal {
            sub_layer_hrd(r, cpb_cnt, sub_pic);
        }
        if vcl {
            sub_layer_hrd(r, cpb_cnt, sub_pic);
        }
    }
    Ok(())
}

#[derive(Clone, Debug)]
pub struct Sps {
    pub id: u32,
    pub profile_idc: u32,
    pub chroma_format_idc: u32,
    pub width: u32,
    pub height: u32,
    /// Luma samples cropped: left, right, top, bottom.
    pub crop: [u32; 4],
    pub bit_depth_y: u32,
    pub bit_depth_c: u32,
    pub log2_max_poc_lsb: u32,
    pub max_dec_pic_buffering: u32,
    /// `sps_max_num_reorder_pics` of the highest sub-layer.
    pub max_num_reorder: u32,
    /// `sps_max_latency_increase_plus1` of the highest sub-layer.
    pub max_latency_increase_plus1: u32,
    pub log2_min_cb: u32,
    pub log2_ctb: u32,
    pub log2_min_tb: u32,
    pub log2_max_tb: u32,
    pub max_th_depth_inter: u32,
    pub max_th_depth_intra: u32,
    pub scaling: Option<ScalingFactors>,
    pub amp: bool,
    pub sao: bool,
    pub pcm: bool,
    pub pcm_bit_depth_y: u32,
    pub pcm_bit_depth_c: u32,
    pub log2_min_pcm: u32,
    pub log2_max_pcm: u32,
    pub pcm_loop_filter_disabled: bool,
    pub st_rps: Vec<StRps>,
    pub long_term_present: bool,
    pub lt_ref_pics: Vec<(u32, bool)>,
    pub temporal_mvp: bool,
    pub strong_intra_smoothing: bool,
    /// Range extension: no reference smoothing in intra prediction.
    pub intra_smoothing_disabled: bool,
    /// Range extension: 4×4 intra residuals without a transform are
    /// rotated by 180°.
    pub transform_skip_rotation: bool,
    /// Range extension: one significance context for untransformed blocks.
    pub transform_skip_context: bool,
    /// Range extension: residual DPCM, implied for intra blocks predicted
    /// straight across and signalled for inter ones.
    pub implicit_rdpcm: bool,
    pub explicit_rdpcm: bool,
    /// Range extension: weighted-prediction offsets at full bit depth.
    pub high_precision_offsets: bool,
    /// Range extension: Rice parameters carried between sub-blocks.
    pub persistent_rice_adaptation: bool,
    /// The VUI's colour description, when it has one.
    pub colour_primaries: Option<u32>,
    pub transfer_characteristics: Option<u32>,
    pub matrix_coefficients: Option<u32>,
    pub full_range: bool,
}

impl Sps {
    /// Chroma subsampling as shifts, `(horizontal, vertical)`: (1, 1) for
    /// 4:2:0, (1, 0) for 4:2:2, (0, 0) for 4:4:4.
    pub fn chroma_shift(&self) -> (usize, usize) {
        (
            (self.chroma_format_idc != 3) as usize,
            (self.chroma_format_idc == 1) as usize,
        )
    }

    pub fn parse(rbsp: &[u8]) -> Result<Sps> {
        let mut r = BitReader::new(rbsp);
        r.skip(4);
        let max_sub_layers_minus1 = r.u(3);
        r.skip(1);
        let profile_idc = profile_tier_level(&mut r, max_sub_layers_minus1);
        let id = r.ue();
        if id > 15 {
            return Err("SPS id out of range".into());
        }
        let chroma_format_idc = r.ue();
        if chroma_format_idc == 3 && r.flag() {
            return Err("separate colour planes are not supported".into());
        }
        if !(1..=3).contains(&chroma_format_idc) {
            return Err(format!(
                "HEVC chroma_format_idc {chroma_format_idc} is not supported (4:2:0, 4:2:2 and 4:4:4 only)"
            ));
        }
        let width = r.ue();
        let height = r.ue();
        let mut crop = [0u32; 4];
        if r.flag() {
            // In chroma units: SubWidthC and SubHeightC.
            let sw = if chroma_format_idc == 3 { 1 } else { 2 };
            let sh = if chroma_format_idc == 1 { 2 } else { 1 };
            let c = [r.ue(), r.ue(), r.ue(), r.ue()];
            crop = [
                c[0].saturating_mul(sw),
                c[1].saturating_mul(sw),
                c[2].saturating_mul(sh),
                c[3].saturating_mul(sh),
            ];
        }
        // Bounds before anything is sized from these (§7.4.3.2.1): garbage
        // reads as all-ones Exp-Golomb codes.
        if width == 0 || height == 0 || width > 1 << 14 || height > 1 << 14 {
            return Err("SPS picture size out of range".into());
        }
        check_picture_size(width, height)?;
        if crop[0].saturating_add(crop[1]) >= width || crop[2].saturating_add(crop[3]) >= height {
            return Err("SPS conformance window exceeds the picture".into());
        }
        let bit_depth_y = r.ue().saturating_add(8);
        let bit_depth_c = r.ue().saturating_add(8);
        if bit_depth_y > 12 || bit_depth_c > 12 {
            return Err("bit depths above 12 are not supported".into());
        }
        let log2_max_poc_lsb = r.ue().saturating_add(4);
        if log2_max_poc_lsb > 16 {
            return Err("log2_max_pic_order_cnt_lsb out of range".into());
        }
        let ordering_all = r.flag();
        let mut max_dec_pic_buffering = 1;
        let mut max_num_reorder = 0;
        let mut max_latency_increase_plus1 = 0;
        let start = if ordering_all {
            0
        } else {
            max_sub_layers_minus1
        };
        for _ in start..=max_sub_layers_minus1 {
            max_dec_pic_buffering = r.ue().saturating_add(1);
            if max_dec_pic_buffering > 16 {
                return Err("sps_max_dec_pic_buffering out of range".into());
            }
            max_num_reorder = r.ue();
            max_latency_increase_plus1 = r.ue();
        }
        let log2_min_cb = r.ue().saturating_add(3);
        let log2_ctb = log2_min_cb.saturating_add(r.ue());
        let log2_min_tb = r.ue().saturating_add(2);
        let log2_max_tb = log2_min_tb.saturating_add(r.ue());
        // Whole minimum coding blocks: a block past this edge could not be
        // split back inside the picture, and every per-block array is sized
        // to the picture.
        let min_cb = 1u64.checked_shl(log2_min_cb).unwrap_or(u64::MAX);
        if (width as u64 | height as u64) & (min_cb - 1) != 0 {
            return Err("picture size is not a multiple of the minimum coding block".into());
        }
        if !(4..=6).contains(&log2_ctb)
            || log2_max_tb > 5
            || log2_min_tb >= log2_min_cb
            || log2_max_tb > log2_ctb
        {
            return Err("coding/transform block sizes out of range".into());
        }
        let max_th_depth_inter = r.ue();
        let max_th_depth_intra = r.ue();
        let mut scaling = None;
        if r.flag() {
            let lists = if r.flag() {
                ScalingLists::parse(&mut r)?
            } else {
                ScalingLists::default_lists()
            };
            scaling = Some(lists.expand());
        }
        let amp = r.flag();
        let sao = r.flag();
        let pcm = r.flag();
        let (mut pcm_bit_depth_y, mut pcm_bit_depth_c, mut log2_min_pcm, mut log2_max_pcm) =
            (8, 8, 0, 0);
        let mut pcm_loop_filter_disabled = false;
        if pcm {
            pcm_bit_depth_y = r.u(4) + 1;
            pcm_bit_depth_c = r.u(4) + 1;
            log2_min_pcm = r.ue().saturating_add(3);
            log2_max_pcm = log2_min_pcm.saturating_add(r.ue());
            pcm_loop_filter_disabled = r.flag();
        }
        let n = r.ue() as usize;
        if n > 64 {
            return Err("num_short_term_ref_pic_sets out of range".into());
        }
        let mut st_rps = Vec::with_capacity(n);
        for i in 0..n {
            let s = StRps::parse(&mut r, i, &st_rps, false)?;
            st_rps.push(s);
        }
        let long_term_present = r.flag();
        let mut lt_ref_pics = Vec::new();
        if long_term_present {
            let n = r.ue();
            if n > 32 {
                return Err("num_long_term_ref_pics_sps out of range".into());
            }
            for _ in 0..n {
                lt_ref_pics.push((r.u(log2_max_poc_lsb), r.flag()));
            }
        }
        let temporal_mvp = r.flag();
        let strong_intra_smoothing = r.flag();
        let (mut colour_primaries, mut transfer_characteristics) = (None, None);
        let mut matrix_coefficients = None;
        let mut full_range = false;
        if r.flag() {
            // vui_parameters()
            if r.flag() && r.u(8) == 255 {
                r.skip(32);
            }
            if r.flag() {
                r.skip(1);
            }
            if r.flag() {
                r.skip(3);
                full_range = r.flag();
                if r.flag() {
                    colour_primaries = Some(r.u(8));
                    transfer_characteristics = Some(r.u(8));
                    matrix_coefficients = Some(r.u(8));
                }
            }
            if r.flag() {
                r.ue();
                r.ue();
            }
            r.skip(3);
            if r.flag() {
                r.ue();
                r.ue();
                r.ue();
                r.ue();
            }
            if r.flag() {
                r.skip(64);
                if r.flag() {
                    r.ue();
                }
                if r.flag() {
                    hrd_parameters(&mut r, true, max_sub_layers_minus1)?;
                }
            }
            if r.flag() {
                r.skip(3);
                r.ue();
                r.ue();
                r.ue();
                r.ue();
                r.ue();
            }
        }
        let mut tools = 0;
        if r.flag() {
            // As for the PPS: before version 2 these were arbitrary extension
            // data bits, so only RExt and later profiles have a range
            // extension. JCT-VC EXT_A_ericsson_3 sets them on a Main stream;
            // read this way it decodes to exactly what HM makes of its
            // reissue, EXT_A_ericsson_4. (ffmpeg gates only the PPS one, and
            // refuses _3.)
            let range = r.flag() && profile_idc >= 4;
            let _others = r.u(7);
            if range {
                // Nine flags, first to last: transform skip rotation and
                // context, implicit and explicit RDPCM, extended precision,
                // intra smoothing disabled, high precision offsets,
                // persistent Rice adaptation, CABAC bypass alignment.
                tools = r.u(9);
                if tools & 0x10 != 0 {
                    return Err("extended precision processing is not supported".into());
                }
                if tools & 0x01 != 0 {
                    return Err("CABAC bypass alignment is not supported".into());
                }
            }
        }
        if r.overrun() {
            return Err("SPS runs past the NAL".into());
        }
        if width == 0 || height == 0 || width > 16384 || height > 16384 {
            return Err("picture size out of range".into());
        }
        let _ = profile_idc;
        Ok(Sps {
            id,
            profile_idc,
            chroma_format_idc,
            width,
            height,
            crop,
            bit_depth_y,
            bit_depth_c,
            log2_max_poc_lsb,
            max_dec_pic_buffering,
            max_num_reorder,
            max_latency_increase_plus1,
            log2_min_cb,
            log2_ctb,
            log2_min_tb,
            log2_max_tb,
            max_th_depth_inter,
            max_th_depth_intra,
            scaling,
            amp,
            sao,
            pcm,
            pcm_bit_depth_y,
            pcm_bit_depth_c,
            log2_min_pcm,
            log2_max_pcm,
            pcm_loop_filter_disabled,
            st_rps,
            long_term_present,
            lt_ref_pics,
            temporal_mvp,
            strong_intra_smoothing,
            intra_smoothing_disabled: tools & 0x08 != 0,
            transform_skip_rotation: tools & 0x100 != 0,
            transform_skip_context: tools & 0x80 != 0,
            implicit_rdpcm: tools & 0x40 != 0,
            explicit_rdpcm: tools & 0x20 != 0,
            high_precision_offsets: tools & 0x04 != 0,
            persistent_rice_adaptation: tools & 0x02 != 0,
            colour_primaries,
            transfer_characteristics,
            matrix_coefficients,
            full_range,
        })
    }

    pub fn ctb_size(&self) -> u32 {
        1 << self.log2_ctb
    }
    pub fn width_ctbs(&self) -> u32 {
        self.width.div_ceil(self.ctb_size())
    }
    pub fn height_ctbs(&self) -> u32 {
        self.height.div_ceil(self.ctb_size())
    }
}

#[derive(Clone, Debug)]
pub struct Pps {
    pub id: u32,
    pub sps_id: u32,
    pub dependent_slices: bool,
    pub output_flag_present: bool,
    pub extra_slice_header_bits: u32,
    pub sign_data_hiding: bool,
    pub cabac_init_present: bool,
    pub num_ref_idx_default: [u32; 2],
    pub init_qp: i32,
    pub constrained_intra_pred: bool,
    pub transform_skip: bool,
    pub cu_qp_delta: bool,
    pub diff_cu_qp_delta_depth: u32,
    pub cb_qp_offset: i32,
    pub cr_qp_offset: i32,
    pub slice_chroma_qp_offsets_present: bool,
    /// Range extension: the (Cb, Cr) QP offsets a coding unit can pick from,
    /// empty when `chroma_qp_offset_list_enabled_flag` is off.
    pub chroma_qp_offset_list: Vec<[i32; 2]>,
    /// Depth below the CTB of the groups that each code one choice.
    pub diff_cu_chroma_qp_offset_depth: u32,
    /// Range extension: largest transform-skip block, log2 (2 without it).
    pub log2_max_transform_skip: u32,
    /// Range extension: chroma residuals may add a scaled luma residual.
    pub cross_component_prediction: bool,
    /// Range extension: SAO offsets are scaled up by these, luma and chroma.
    pub log2_sao_offset_scale: [u32; 2],
    pub weighted_pred: bool,
    pub weighted_bipred: bool,
    pub transquant_bypass: bool,
    pub tiles: bool,
    pub entropy_sync: bool,
    /// Explicit tile column widths / row heights in CTBs, or `None` for
    /// uniform spacing with the given count.
    pub tile_cols: TileSpacing,
    pub tile_rows: TileSpacing,
    pub loop_filter_across_tiles: bool,
    pub loop_filter_across_slices: bool,
    pub deblocking_override_enabled: bool,
    pub deblocking_disabled: bool,
    pub beta_offset: i32,
    pub tc_offset: i32,
    pub scaling: Option<ScalingFactors>,
    pub lists_modification_present: bool,
    pub log2_parallel_merge_level: u32,
    pub slice_header_extension_present: bool,
}

#[derive(Clone, Debug)]
pub enum TileSpacing {
    Uniform(u32),
    Explicit(Vec<u32>),
}

impl TileSpacing {
    /// Boundaries in CTBs: `n + 1` entries from 0 to `total`.
    pub fn boundaries(&self, total: u32) -> Vec<u32> {
        let mut b = vec![0];
        match self {
            TileSpacing::Uniform(n) => {
                for i in 0..*n {
                    b.push(((i as u64 + 1) * total as u64 / *n as u64) as u32);
                }
            }
            TileSpacing::Explicit(sizes) => {
                let mut acc = 0;
                // Saturating: a damaged parameter set can state sizes that
                // add up past `u32`.
                for &s in sizes {
                    acc = u32::saturating_add(acc, s);
                    b.push(acc.min(total));
                }
                b.push(total);
            }
        }
        b
    }
}

impl Pps {
    /// `spss` are the sequence parameter sets received so far, indexed by id:
    /// the referenced one's profile decides whether a range extension is read.
    pub fn parse(rbsp: &[u8], spss: &[Option<Sps>]) -> Result<Pps> {
        let mut r = BitReader::new(rbsp);
        let id = r.ue();
        let sps_id = r.ue();
        if id > 63 || sps_id > 15 {
            return Err("PPS id out of range".into());
        }
        let dependent_slices = r.flag();
        let output_flag_present = r.flag();
        let extra_slice_header_bits = r.u(3);
        let sign_data_hiding = r.flag();
        let cabac_init_present = r.flag();
        let num_ref_idx_default = [r.ue().saturating_add(1), r.ue().saturating_add(1)];
        if num_ref_idx_default[0] > 15 || num_ref_idx_default[1] > 15 {
            return Err("num_ref_idx_default out of range".into());
        }
        let init_qp = r.se().saturating_add(26);
        let constrained_intra_pred = r.flag();
        let transform_skip = r.flag();
        let cu_qp_delta = r.flag();
        let diff_cu_qp_delta_depth = if cu_qp_delta { r.ue() } else { 0 };
        let cb_qp_offset = r.se();
        let cr_qp_offset = r.se();
        if !(-12..=12).contains(&cb_qp_offset) || !(-12..=12).contains(&cr_qp_offset) {
            return Err("pps_cb_qp_offset or pps_cr_qp_offset out of range".into());
        }
        let slice_chroma_qp_offsets_present = r.flag();
        let weighted_pred = r.flag();
        let weighted_bipred = r.flag();
        let transquant_bypass = r.flag();
        let tiles = r.flag();
        let entropy_sync = r.flag();
        let mut tile_cols = TileSpacing::Uniform(1);
        let mut tile_rows = TileSpacing::Uniform(1);
        let mut loop_filter_across_tiles = true;
        if tiles {
            let ncols = r.ue().saturating_add(1);
            let nrows = r.ue().saturating_add(1);
            // Level 6.2's limits are 20 × 22; past those it is not a stream.
            if ncols > 20 || nrows > 22 {
                return Err("tile counts out of range".into());
            }
            if ncols > 64 || nrows > 64 {
                return Err("too many tiles".into());
            }
            if r.flag() {
                tile_cols = TileSpacing::Uniform(ncols);
                tile_rows = TileSpacing::Uniform(nrows);
            } else {
                tile_cols =
                    TileSpacing::Explicit((1..ncols).map(|_| r.ue().saturating_add(1)).collect());
                tile_rows =
                    TileSpacing::Explicit((1..nrows).map(|_| r.ue().saturating_add(1)).collect());
            }
            loop_filter_across_tiles = r.flag();
        }
        let loop_filter_across_slices = r.flag();
        let mut deblocking_override_enabled = false;
        let mut deblocking_disabled = false;
        let mut beta_offset = 0;
        let mut tc_offset = 0;
        if r.flag() {
            deblocking_override_enabled = r.flag();
            deblocking_disabled = r.flag();
            if !deblocking_disabled {
                beta_offset = r.se().saturating_mul(2);
                tc_offset = r.se().saturating_mul(2);
            }
        }
        let mut scaling = None;
        if r.flag() {
            scaling = Some(ScalingLists::parse(&mut r)?.expand());
        }
        let lists_modification_present = r.flag();
        let log2_parallel_merge_level = r.ue().saturating_add(2);
        // At most CtbLog2SizeY, which is at most 6; merge candidates shift by
        // it, so a damaged value has to stop here.
        if log2_parallel_merge_level > 6 {
            return Err("log2_parallel_merge_level_minus2 out of range".into());
        }
        let slice_header_extension_present = r.flag();
        let mut chroma_qp_offset_list = Vec::new();
        let mut diff_cu_chroma_qp_offset_depth = 0;
        let mut log2_max_transform_skip = 2;
        let mut cross_component_prediction = false;
        let mut log2_sao_offset_scale = [0, 0];
        if r.flag() {
            // Streams older than the range extensions (version 2) set
            // pps_extension_flag and follow it with arbitrary extension data,
            // which a Main or Main 10 decoder must ignore (JCT-VC PS_A_VIDYO_3
            // tests exactly that). So, as ffmpeg does, read the range extension
            // only for RExt and later profiles; without the SPS, assume one.
            let rext = spss
                .get(sps_id as usize)
                .and_then(Option::as_ref)
                .is_none_or(|s| s.profile_idc >= 4);
            let range = r.flag() && rext;
            let _others = r.u(7);
            if range {
                if transform_skip {
                    // Only ever compared with block sizes; HM accepts the
                    // out-of-range 8 in JCT-VC PERSIST_RPARAM_A_RExt_Sony_1.
                    log2_max_transform_skip = r.ue().saturating_add(2);
                }
                cross_component_prediction = r.flag();
                if r.flag() {
                    diff_cu_chroma_qp_offset_depth = r.ue();
                    let len = r.ue().saturating_add(1);
                    if len > 6 {
                        return Err("chroma_qp_offset_list_len_minus1 out of range".into());
                    }
                    for _ in 0..len {
                        let pair = [r.se(), r.se()];
                        if pair.iter().any(|o| !(-12..=12).contains(o)) {
                            return Err("chroma QP offset list entry out of range".into());
                        }
                        chroma_qp_offset_list.push(pair);
                    }
                }
                log2_sao_offset_scale = [r.ue(), r.ue()];
                // At most bitDepth − 10, so never above 6 (16-bit).
                if log2_sao_offset_scale.iter().any(|&v| v > 6) {
                    return Err("log2_sao_offset_scale out of range".into());
                }
            }
        }
        Ok(Pps {
            id,
            sps_id,
            dependent_slices,
            output_flag_present,
            extra_slice_header_bits,
            sign_data_hiding,
            cabac_init_present,
            num_ref_idx_default,
            init_qp,
            constrained_intra_pred,
            transform_skip,
            cu_qp_delta,
            diff_cu_qp_delta_depth,
            cb_qp_offset,
            cr_qp_offset,
            slice_chroma_qp_offsets_present,
            chroma_qp_offset_list,
            diff_cu_chroma_qp_offset_depth,
            log2_max_transform_skip,
            cross_component_prediction,
            log2_sao_offset_scale,
            weighted_pred,
            weighted_bipred,
            transquant_bypass,
            tiles,
            entropy_sync,
            tile_cols,
            tile_rows,
            loop_filter_across_tiles,
            loop_filter_across_slices,
            deblocking_override_enabled,
            deblocking_disabled,
            beta_offset,
            tc_offset,
            scaling,
            lists_modification_present,
            log2_parallel_merge_level,
            slice_header_extension_present,
        })
    }
}

#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub enum SliceType {
    B,
    P,
    I,
}

/// Explicit weighted prediction: `[list][ref] = [(weight, offset); Y, Cb, Cr]`
/// with offsets already scaled to the bit depth.
#[derive(Clone, Debug, Default)]
pub struct PredWeights {
    pub luma_log2_denom: u32,
    pub chroma_log2_denom: u32,
    pub w: [Vec<[(i32, i32); 3]>; 2],
}

#[derive(Clone, Debug)]
pub struct SliceHeader {
    pub first_slice_in_pic: bool,
    pub no_output_of_prior_pics: bool,
    pub pps_id: u32,
    pub dependent: bool,
    pub segment_address: u32,
    pub slice_type: SliceType,
    pub pic_output: bool,
    pub poc_lsb: u32,
    pub st_rps: StRps,
    /// Long-term entries: `(poc value, msb present, used by current)`.
    pub lt: Vec<(i32, bool, bool)>,
    pub temporal_mvp: bool,
    pub sao_luma: bool,
    pub sao_chroma: bool,
    pub num_ref_idx: [u32; 2],
    pub list_entry: [Option<Vec<u32>>; 2],
    pub mvd_l1_zero: bool,
    pub cabac_init_flag: bool,
    pub collocated_from_l0: bool,
    pub collocated_ref_idx: u32,
    pub weights: Option<PredWeights>,
    pub max_merge_cand: u32,
    pub qp_delta: i32,
    pub cb_qp_offset: i32,
    pub cr_qp_offset: i32,
    /// `cu_chroma_qp_offset_enabled_flag`: coding units may pick from the
    /// PPS chroma QP offset list.
    pub cu_chroma_qp_offset: bool,
    pub deblocking_disabled: bool,
    pub beta_offset: i32,
    pub tc_offset: i32,
    pub loop_filter_across_slices: bool,
    /// Byte offset of `slice_segment_data()` in the RBSP.
    pub data_offset: usize,
    /// Bits the slice header spent on its own `st_ref_pic_set()`; 0 when it
    /// named one of the SPS's.
    pub st_rps_bits: u32,
    /// `num_entry_point_offsets`.
    pub num_entry_points: u32,
}

impl SliceHeader {
    /// Parse a slice segment header. A dependent segment copies everything
    /// but its address from `prev`, the last independent one.
    pub fn parse(
        rbsp: &[u8],
        nal_type: u32,
        spss: &[Option<Sps>],
        ppss: &[Option<Pps>],
        prev: Option<&SliceHeader>,
    ) -> Result<SliceHeader> {
        let mut r = BitReader::new(rbsp);
        let first = r.flag();
        let mut no_output_of_prior_pics = false;
        if (16..=23).contains(&nal_type) {
            no_output_of_prior_pics = r.flag();
        }
        let pps_id = r.ue();
        let pps = ppss
            .get(pps_id as usize)
            .and_then(|p| p.as_ref())
            .ok_or_else(|| format!("slice names missing PPS {pps_id}"))?;
        let sps = spss[pps.sps_id as usize]
            .as_ref()
            .ok_or("slice's PPS names a missing SPS")?;
        let mut dependent = false;
        let mut segment_address = 0;
        if !first {
            if pps.dependent_slices {
                dependent = r.flag();
            }
            let ctbs = sps.width_ctbs() * sps.height_ctbs();
            segment_address = r.u(ceil_log2(ctbs));
            if segment_address >= ctbs {
                return Err("slice_segment_address out of range".into());
            }
        }
        let mut h = if dependent {
            let mut h = prev
                .ok_or("dependent slice segment without a preceding slice")?
                .clone();
            h.first_slice_in_pic = first;
            h.dependent = true;
            h.segment_address = segment_address;
            h
        } else {
            let mut h = SliceHeader {
                first_slice_in_pic: first,
                no_output_of_prior_pics,
                pps_id,
                dependent: false,
                segment_address,
                slice_type: SliceType::I,
                pic_output: true,
                poc_lsb: 0,
                st_rps: StRps::default(),
                lt: Vec::new(),
                temporal_mvp: false,
                sao_luma: false,
                sao_chroma: false,
                num_ref_idx: [0, 0],
                list_entry: [None, None],
                mvd_l1_zero: false,
                cabac_init_flag: false,
                collocated_from_l0: true,
                collocated_ref_idx: 0,
                weights: None,
                max_merge_cand: 5,
                qp_delta: 0,
                cb_qp_offset: 0,
                cr_qp_offset: 0,
                cu_chroma_qp_offset: false,
                deblocking_disabled: pps.deblocking_disabled,
                beta_offset: pps.beta_offset,
                tc_offset: pps.tc_offset,
                loop_filter_across_slices: pps.loop_filter_across_slices,
                data_offset: 0,
                st_rps_bits: 0,
                num_entry_points: 0,
            };
            r.skip(pps.extra_slice_header_bits);
            h.slice_type = match r.ue() {
                0 => SliceType::B,
                1 => SliceType::P,
                2 => SliceType::I,
                t => return Err(format!("slice_type {t} out of range")),
            };
            if pps.output_flag_present {
                h.pic_output = r.flag();
            }
            let idr = nal_type == 19 || nal_type == 20;
            let mut num_pic_total_curr = 0u32;
            if !idr {
                h.poc_lsb = r.u(sps.log2_max_poc_lsb);
                if !r.flag() {
                    let at = r.pos();
                    h.st_rps = StRps::parse(&mut r, sps.st_rps.len(), &sps.st_rps, true)?;
                    h.st_rps_bits = (r.pos() - at) as u32;
                } else {
                    if sps.st_rps.is_empty() {
                        return Err("slice refers to an SPS RPS but there are none".into());
                    }
                    let idx = if sps.st_rps.len() > 1 {
                        r.u(ceil_log2(sps.st_rps.len() as u32)) as usize
                    } else {
                        0
                    };
                    h.st_rps = sps
                        .st_rps
                        .get(idx)
                        .ok_or("short_term_ref_pic_set_idx out of range")?
                        .clone();
                }
                if sps.long_term_present {
                    let n_sps = if !sps.lt_ref_pics.is_empty() {
                        r.ue()
                    } else {
                        0
                    };
                    let n_pics = r.ue();
                    if n_sps as usize > sps.lt_ref_pics.len() || n_sps + n_pics > 32 {
                        return Err("too many long-term pictures".into());
                    }
                    let max_lsb = 1i32 << sps.log2_max_poc_lsb;
                    let mut msb_cycle = 0i32;
                    for i in 0..(n_sps + n_pics) {
                        let (lsb, used) = if i < n_sps {
                            let idx = if sps.lt_ref_pics.len() > 1 {
                                r.u(ceil_log2(sps.lt_ref_pics.len() as u32)) as usize
                            } else {
                                0
                            };
                            let e = sps.lt_ref_pics.get(idx).ok_or("lt_idx_sps out of range")?;
                            (e.0 as i32, e.1)
                        } else {
                            (r.u(sps.log2_max_poc_lsb) as i32, r.flag())
                        };
                        let msb_present = r.flag();
                        let mut poc = lsb;
                        if msb_present {
                            let delta = r.ue() as i32;
                            msb_cycle = if i == 0 || i == n_sps {
                                delta
                            } else {
                                delta + msb_cycle
                            };
                            // The rest of the MSB comes from the current
                            // POC, which the caller adds: §7.4.7.1 (7-52).
                            poc = lsb - msb_cycle * max_lsb;
                        }
                        h.lt.push((poc, msb_present, used));
                    }
                }
                if sps.temporal_mvp {
                    h.temporal_mvp = r.flag();
                }
                num_pic_total_curr = h.st_rps.neg.iter().filter(|e| e.1).count() as u32
                    + h.st_rps.pos.iter().filter(|e| e.1).count() as u32
                    + h.lt.iter().filter(|e| e.2).count() as u32;
            }
            if sps.sao {
                h.sao_luma = r.flag();
                h.sao_chroma = r.flag();
            }
            if h.slice_type != SliceType::I {
                h.num_ref_idx = pps.num_ref_idx_default;
                if r.flag() {
                    h.num_ref_idx[0] = r.ue().saturating_add(1);
                    if h.slice_type == SliceType::B {
                        h.num_ref_idx[1] = r.ue().saturating_add(1);
                    }
                }
                if h.slice_type != SliceType::B {
                    h.num_ref_idx[1] = 0;
                }
                if h.num_ref_idx[0] > 15 || h.num_ref_idx[1] > 15 {
                    return Err("num_ref_idx_active out of range".into());
                }
                if pps.lists_modification_present && num_pic_total_curr > 1 {
                    let bits = ceil_log2(num_pic_total_curr);
                    for l in 0..(1 + (h.slice_type == SliceType::B) as usize) {
                        if r.flag() {
                            h.list_entry[l] =
                                Some((0..h.num_ref_idx[l]).map(|_| r.u(bits)).collect());
                        }
                    }
                }
                if h.slice_type == SliceType::B {
                    h.mvd_l1_zero = r.flag();
                }
                if pps.cabac_init_present {
                    h.cabac_init_flag = r.flag();
                }
                if h.temporal_mvp {
                    if h.slice_type == SliceType::B {
                        h.collocated_from_l0 = r.flag();
                    }
                    let l = if h.collocated_from_l0 { 0 } else { 1 };
                    if h.num_ref_idx[l] > 1 {
                        h.collocated_ref_idx = r.ue();
                        if h.collocated_ref_idx >= h.num_ref_idx[l] {
                            return Err("collocated_ref_idx out of range".into());
                        }
                    }
                }
                if (pps.weighted_pred && h.slice_type == SliceType::P)
                    || (pps.weighted_bipred && h.slice_type == SliceType::B)
                {
                    h.weights = Some(pred_weight_table(&mut r, sps, &h)?);
                }
                // Checked before subtracting: a damaged value above 4 would
                // underflow.
                let five_minus = r.ue();
                if five_minus > 4 {
                    return Err("five_minus_max_num_merge_cand out of range".into());
                }
                h.max_merge_cand = 5 - five_minus;
            }
            h.qp_delta = r.se();
            // QP arithmetic below indexes tables by these, and only stays in
            // range if the slice starts in range.
            let qp_bd_offset = 6 * (sps.bit_depth_y as i32 - 8);
            if !(-qp_bd_offset..=51).contains(&pps.init_qp.saturating_add(h.qp_delta)) {
                return Err("SliceQpY out of range".into());
            }
            if pps.slice_chroma_qp_offsets_present {
                h.cb_qp_offset = r.se();
                h.cr_qp_offset = r.se();
                let sum = |a: i32, b: i32| a.saturating_add(b);
                if !(-12..=12).contains(&sum(pps.cb_qp_offset, h.cb_qp_offset))
                    || !(-12..=12).contains(&sum(pps.cr_qp_offset, h.cr_qp_offset))
                {
                    return Err("slice chroma QP offset out of range".into());
                }
            }
            if !pps.chroma_qp_offset_list.is_empty() {
                h.cu_chroma_qp_offset = r.flag();
            }
            let mut override_flag = false;
            if pps.deblocking_override_enabled {
                override_flag = r.flag();
            }
            if override_flag {
                h.deblocking_disabled = r.flag();
                if !h.deblocking_disabled {
                    h.beta_offset = r.se().saturating_mul(2);
                    h.tc_offset = r.se().saturating_mul(2);
                }
            }
            if pps.loop_filter_across_slices
                && (h.sao_luma || h.sao_chroma || !h.deblocking_disabled)
            {
                h.loop_filter_across_slices = r.flag();
            }
            h
        };
        h.num_entry_points = 0;
        if pps.tiles || pps.entropy_sync {
            let n = r.ue();
            h.num_entry_points = n;
            if n > 0 {
                let len = r.ue().saturating_add(1);
                if len > 32 {
                    return Err("offset_len_minus1 out of range".into());
                }
                for _ in 0..n {
                    r.skip(len);
                }
            }
        }
        if pps.slice_header_extension_present {
            let len = r.ue();
            r.skip(8 * len);
        }
        // byte_alignment(): a one bit, then zeros.
        r.skip(1);
        r.byte_align();
        if r.overrun() {
            return Err("slice header runs past the NAL".into());
        }
        h.data_offset = r.pos() / 8;
        Ok(h)
    }
}

fn pred_weight_table(r: &mut BitReader, sps: &Sps, h: &SliceHeader) -> Result<PredWeights> {
    let mut w = PredWeights {
        luma_log2_denom: r.ue(),
        ..Default::default()
    };
    if w.luma_log2_denom > 7 {
        return Err("luma_log2_weight_denom out of range".into());
    }
    let cd = (w.luma_log2_denom as i32).saturating_add(r.se());
    if !(0..=7).contains(&cd) {
        return Err("chroma weight denominator out of range".into());
    }
    w.chroma_log2_denom = cd as u32;
    // Offsets are coded at 8 bits and scaled up, or with
    // high_precision_offsets_enabled_flag at the full bit depth.
    let (sy, sc, half) = if sps.high_precision_offsets {
        (0, 0, 1 << (sps.bit_depth_c - 1))
    } else {
        (sps.bit_depth_y as i32 - 8, sps.bit_depth_c as i32 - 8, 128)
    };
    for l in 0..(1 + (h.slice_type == SliceType::B) as usize) {
        let n = h.num_ref_idx[l] as usize;
        let luma: Vec<bool> = (0..n).map(|_| r.flag()).collect();
        let chroma: Vec<bool> = (0..n).map(|_| r.flag()).collect();
        for i in 0..n {
            let mut e = [
                (1 << w.luma_log2_denom, 0),
                (1 << w.chroma_log2_denom, 0),
                (1 << w.chroma_log2_denom, 0),
            ];
            if luma[i] {
                e[0].0 += r.se();
                e[0].1 = r.se() << sy;
            }
            if chroma[i] {
                for ec in e.iter_mut().skip(1).take(2) {
                    let dw = r.se();
                    let doff = r.se();
                    let wc = (1 << w.chroma_log2_denom) + dw;
                    let off =
                        (half - ((half * wc) >> w.chroma_log2_denom) + doff).clamp(-half, half - 1);
                    *ec = (wc, off << sc);
                }
            }
            w.w[l].push(e);
        }
    }
    Ok(w)
}

#[cfg(test)]
mod tests {
    use super::*;

    /// `cpb_cnt_minus1` is a loop count, so a value past the spec's 31 has
    /// to stop parsing: damage can make it four billion.
    #[test]
    fn hrd_refuses_a_cpb_count_past_the_limit() {
        // nal_hrd = 1, vcl_hrd = 0, sub_pic = 0, 23 skipped bits, then
        // fixed_pic_rate_general = 1, elemental_duration ue(0) = 1, and
        // cpb_cnt_minus1 as ue(v).
        fn hrd(cpb_code: &str) -> Result<()> {
            let bits = format!("100{}11{cpb_code}{}", "0".repeat(23), "1".repeat(64));
            let bytes: Vec<u8> = bits
                .as_bytes()
                .chunks(8)
                .map(|c| c.iter().fold(0, |b, &d| b << 1 | (d - b'0')) << (8 - c.len()))
                .collect();
            hrd_parameters(&mut BitReader::new(&bytes), true, 0)
        }
        assert!(hrd("00000100000").is_ok(), "31 is allowed");
        assert!(hrd("00000100001").is_err(), "32 is not");
    }
}