threers-codec 0.0.10

Pure-Rust media codecs: H.264/HEVC/VP9 encoders, H.264/HEVC/VP9/AV1 decoders, AAC/MP3/Opus/Vorbis/FLAC/WAV decoders, 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
//! Frame set-up and the tile/superblock-row loop (dav1d `dav1d_submit_frame`,
//! `dav1d_decode_frame_init`, `setup_tile`, `dav1d_decode_tile_sbrow`,
//! `dav1d_decode_frame_main`).

use std::sync::Arc;

use super::cdf::CdfContext;
use super::decode::*;
use super::env::BlockContext;
use super::intra_edge;
use super::lf_mask::{self, Av1Filter, Av1FilterLut, Av1Restoration};
use super::msac::Msac;
use super::obu::*;
use super::picture::Picture;
use super::refmvs::{RefmvsFrame, RefmvsTile, TemporalBlock};
use super::tables::SGR_PARAMS;
use super::warpmv;

/// What a reference slot keeps (dav1d `c->refs[i]` + `c->cdf[i]`).
#[derive(Clone)]
pub struct RefSlot {
    pub pic: Arc<Picture>,
    pub hdr: Arc<FrameHeader>,
    pub seq: Arc<SeqHeader>,
    pub cdf: Arc<CdfContext>,
    pub segmap: Option<Arc<Vec<u8>>>,
    pub refmvs: Option<Arc<Vec<TemporalBlock>>>,
    pub refpoc: [u32; 7],
    pub showable: bool,
}

pub struct TileData {
    pub data: Vec<u8>,
    pub start: u32,
    pub end: u32,
}

fn scale_fac(r: i32, t: i32) -> i32 {
    ((r << 14) + (t >> 1)) / t
}

fn get_upscale_x0(in_w: i32, out_w: i32, step: i32) -> i32 {
    let err = out_w * step - (in_w << 14);
    let x0 = (-((out_w - in_w) << 13) + (out_w >> 1)) / out_w + 128 - (err / 2);
    x0 & 0x3fff
}


impl Fd {
    /// Allocate and initialise everything for one frame.
    pub fn new(seq: Arc<SeqHeader>, hdr: Arc<FrameHeader>, slots: &[Option<RefSlot>; 8]) -> Result<(Fd, Arc<CdfContext>), String> {
        let bpc = 8 + 2 * seq.hbd as u32;
        let layout = seq.layout;
        let ss_hor = (layout == Layout::I420 || layout == Layout::I422) as i32;
        let ss_ver = (layout == Layout::I420) as i32;
        let mut refp: [Option<RefFrame>; 7] = Default::default();
        let mut svc = [[Svc::default(); 2]; 7];
        let mut gmv_warp_allowed = [false; 7];
        let mut ref_coded_width = [0i32; 7];
        if hdr.is_inter_or_switch() {
            if hdr.primary_ref_frame != PRIMARY_REF_NONE && slots[hdr.refidx[hdr.primary_ref_frame as usize] as usize].is_none() {
                return Err("missing primary reference".into());
            }
            for i in 0..7 {
                let s = slots[hdr.refidx[i] as usize].as_ref().ok_or("missing reference")?;
                let (rw, rh) = (s.pic.w, s.pic.h);
                if hdr.width[0] * 2 < rw || hdr.height * 2 < rh || hdr.width[0] > rw * 16 || hdr.height > rh * 16 || s.pic.layout != layout || s.pic.bpc != bpc {
                    return Err("invalid reference".into());
                }
                ref_coded_width[i] = s.hdr.width[0];
                if hdr.width[0] != rw || hdr.height != rh {
                    svc[i][0].scale = scale_fac(rw, hdr.width[0]);
                    svc[i][1].scale = scale_fac(rh, hdr.height);
                    svc[i][0].step = (svc[i][0].scale + 8) >> 4;
                    svc[i][1].step = (svc[i][1].scale + 8) >> 4;
                }
                let mut g = hdr.gmv[i];
                gmv_warp_allowed[i] = hdr.gmv[i].typ > WM_TRANSLATION && !hdr.force_integer_mv && !warpmv::get_shear_params(&mut g) && svc[i][0].scale == 0;
                refp[i] = Some(RefFrame { pic: s.pic.clone(), hdr: s.hdr.clone() });
            }
        }
        // The shear parameters of the global motion models are used by MC.
        let mut hdr_m = (*hdr).clone();
        for g in hdr_m.gmv.iter_mut() {
            let _ = warpmv::get_shear_params(g);
        }
        let hdr = Arc::new(hdr_m);

        let in_cdf = if hdr.primary_ref_frame == PRIMARY_REF_NONE {
            Arc::new(*CdfContext::default_for(hdr.quant.yac))
        } else {
            slots[hdr.refidx[hdr.primary_ref_frame as usize] as usize].as_ref().unwrap().cdf.clone()
        };

        let cur = Picture::new(hdr.width[0], hdr.height, layout, bpc);
        let sr_cur = if hdr.width[0] != hdr.width[1] { Some(Picture::new(hdr.width[1], hdr.height, layout, bpc)) } else { None };
        let mut resize_step = [0; 2];
        let mut resize_start = [0; 2];
        if let Some(sr) = &sr_cur {
            resize_step[0] = scale_fac(cur.w, sr.w);
            let in_cw = (cur.w + ss_hor) >> ss_hor;
            let out_cw = (sr.w + ss_hor) >> ss_hor;
            resize_step[1] = scale_fac(in_cw, out_cw);
            resize_start[0] = get_upscale_x0(cur.w, sr.w, resize_step[0]);
            resize_start[1] = get_upscale_x0(in_cw, out_cw, resize_step[1]);
        }

        let w4 = (hdr.width[0] + 3) >> 2;
        let h4 = (hdr.height + 3) >> 2;
        let bw = ((hdr.width[0] + 7) >> 3) << 1;
        let bh = ((hdr.height + 7) >> 3) << 1;
        let sb128w = (bw + 31) >> 5;
        let sb128h = (bh + 31) >> 5;
        let sb_shift = 4 + seq.sb128 as i32;
        let sb_step = 16 << seq.sb128 as i32;
        let sbh = (bh + sb_step - 1) >> sb_shift;
        let b4_stride = (bw + 31) & !31;
        let sr_w = sr_cur.as_ref().map(|p| p.w).unwrap_or(cur.w);
        let sr_sb128w = (sr_w + 127) >> 7;

        // Reference MVs.
        let mut rf = None;
        let mut refpoc = [0u32; 7];
        if hdr.is_inter_or_switch() || hdr.allow_intrabc {
            if !hdr.allow_intrabc {
                for i in 0..7 {
                    refpoc[i] = refp[i].as_ref().unwrap().hdr.frame_offset as u32;
                }
            }
            let mut rp_ref: [Option<Arc<Vec<TemporalBlock>>>; 7] = Default::default();
            let mut refrefpoc = [[0u32; 7]; 7];
            if hdr.use_ref_frame_mvs {
                for i in 0..7 {
                    let s = slots[hdr.refidx[i] as usize].as_ref().unwrap();
                    let ref_w = ((ref_coded_width[i] + 7) >> 3) << 1;
                    let ref_h = ((s.pic.h + 7) >> 3) << 1;
                    if s.refmvs.is_some() && ref_w == bw && ref_h == bh {
                        rp_ref[i] = s.refmvs.clone();
                    }
                    refrefpoc[i] = s.refpoc;
                }
            }
            rf = Some(RefmvsFrame::new(&seq, &hdr, &refpoc, &refrefpoc, rp_ref));
        }

        // Segmentation maps.
        let mut prev_segmap = None;
        let mut cur_segmap = None;
        let segmap_size = b4_stride as usize * 32 * sb128h as usize;
        if hdr.segmentation.enabled {
            if hdr.segmentation.temporal || !hdr.segmentation.update_map {
                let pri = hdr.primary_ref_frame as usize;
                let ref_w = ((ref_coded_width[pri] + 7) >> 3) << 1;
                let ref_h = ((refp[pri].as_ref().unwrap().pic.h + 7) >> 3) << 1;
                if ref_w == bw && ref_h == bh {
                    prev_segmap = slots[hdr.refidx[pri] as usize].as_ref().unwrap().segmap.clone();
                }
            }
            cur_segmap = Some(if hdr.segmentation.update_map {
                vec![0u8; segmap_size]
            } else if let Some(p) = &prev_segmap {
                p.as_ref().clone()
            } else {
                vec![0u8; segmap_size]
            });
        }

        // Loop filter state.
        let num_sb128 = (sb128w * sb128h) as usize;
        let mut lim_lut = Av1FilterLut::default();
        lf_mask::calc_eih(&mut lim_lut, hdr.loopfilter.sharpness as i32);
        let mut lvl = [[[[0u8; 2]; 8]; 4]; 8];
        lf_mask::calc_lf_values(&mut lvl, &hdr, [0; 4]);
        let restore_planes = (hdr.restoration.typ[0] != RESTORATION_NONE) as u32
            | (((hdr.restoration.typ[1] != RESTORATION_NONE) as u32) << 1)
            | (((hdr.restoration.typ[2] != RESTORATION_NONE) as u32) << 2);
        let mut start_of_tile_row = vec![0u8; sbh as usize];
        let mut sby = 0usize;
        for tile_row in 0..hdr.tiling.rows as usize {
            start_of_tile_row[sby] = tile_row as u8;
            sby += 1;
            while sby < hdr.tiling.row_start_sb[tile_row + 1] as usize {
                start_of_tile_row[sby] = 0;
                sby += 1;
            }
        }
        let re_sz = (sb128h * hdr.tiling.cols as i32) as usize;
        let lf = LfState {
            level: vec![[0u8; 4]; num_sb128 * 32 * 32 + 3],
            mask: vec![Av1Filter::default(); num_sb128],
            lr_mask: vec![Av1Restoration::default(); (sr_sb128w * sb128h) as usize],
            lim_lut,
            lvl,
            tx_lpf_right_edge: [vec![0u8; re_sz * 32], vec![0u8; re_sz * 32]],
            start_of_tile_row,
            restore_planes,
        };

        let mut dq = [[[0u16; 2]; 3]; 8];
        init_quant_tables(&seq, &hdr, hdr.quant.yac as i32, &mut dq);
        let mut qm = [[None; 3]; 19];
        if hdr.quant.qm {
            for row in qm.iter_mut() {
                *row = [
                    if hdr.quant.qm_y < 15 { Some((hdr.quant.qm_y as usize, 0)) } else { None },
                    if hdr.quant.qm_u < 15 { Some((hdr.quant.qm_u as usize, 1)) } else { None },
                    if hdr.quant.qm_v < 15 { Some((hdr.quant.qm_v as usize, 1)) } else { None },
                ];
            }
        }

        let mut jnt_weights = [[0u8; 7]; 7];
        if hdr.switchable_comp_refs {
            const QDW: [[i32; 2]; 3] = [[2, 3], [2, 5], [2, 7]];
            const QDL: [[u8; 2]; 4] = [[9, 7], [11, 5], [12, 4], [13, 3]];
            let nb = seq.order_hint_n_bits;
            for i in 0..7 {
                let r0 = refp[i].as_ref().unwrap().hdr.frame_offset as i32;
                for j in i + 1..7 {
                    let r1 = refp[j].as_ref().unwrap().hdr.frame_offset as i32;
                    let d1 = get_poc_diff(nb, r0, hdr.frame_offset as i32).abs().min(31);
                    let d0 = get_poc_diff(nb, r1, hdr.frame_offset as i32).abs().min(31);
                    let order = (d0 <= d1) as usize;
                    let mut k = 0;
                    while k < 3 {
                        let c0 = QDW[k][order];
                        let c1 = QDW[k][1 - order];
                        let d0c0 = d0 * c0;
                        let d1c1 = d1 * c1;
                        if (d0 > d1 && d0c0 < d1c1) || (d0 <= d1 && d0c0 > d1c1) {
                            break;
                        }
                        k += 1;
                    }
                    jnt_weights[i][j] = QDL[k][order];
                }
            }
        }

        let n_ts = hdr.tiling.cols as usize * hdr.tiling.rows as usize;
        let ipred_sz = (sbh * sb128w) as usize * 128;
        let fd = Fd {
            seq: seq.clone(),
            hdr: hdr.clone(),
            layout,
            ss_hor,
            ss_ver,
            bpc,
            bitdepth_max: (1 << bpc) - 1,
            cur,
            sr_cur,
            refp,
            svc,
            gmv_warp_allowed,
            w4,
            h4,
            bw,
            bh,
            sb128w,
            sb_shift,
            sb_step,
            sbh,
            sr_sb128w,
            b4_stride,
            resize_step,
            resize_start,
            dq,
            qm,
            a: vec![BlockContext::default(); (sb128w * hdr.tiling.rows as i32) as usize],
            ai: 0,
            rf,
            jnt_weights,
            lf,
            ipred_edge: [vec![0; ipred_sz], vec![0; ipred_sz], vec![0; ipred_sz]],
            cur_segmap,
            prev_segmap,
            ts: Vec::with_capacity(n_ts),
            tsi: 0,
            bx: 0,
            by: 0,
            l: BlockContext::default(),
            rt: RefmvsTile { col_start: 0, col_end: 0, row_start: 0, row_end: 0 },
            lf_mask: 0,
            cdef_idx: 0,
            warpmv: DEFAULT_WM_PARAMS,
            tl_4x4_filter: 0,
            pal: [[0; 8]; 3],
            al_pal: [[[[0; 8]; 3]; 32]; 2],
            pal_sz_uv: [[0; 32]; 2],
            pal_idx_y: vec![0; 64 * 64],
            pal_idx_uv: vec![0; 64 * 64],
            txtp_map: [0; 32 * 32],
            cf: vec![0; 32 * 32],
            levels: vec![0; 32 * 34],
            refpoc,
            mc_buf: Vec::new(),
        };
        Ok((fd, in_cdf))
    }

    fn setup_tile(&mut self, in_cdf: &CdfContext, data: &[u8], tile_row: usize, tile_col: usize) {
        let hdr = self.hdr.clone();
        let col_sb_start = hdr.tiling.col_start_sb[tile_col] as i32;
        let col_sb128_start = col_sb_start >> (!self.seq.sb128) as i32;
        let col_sb_end = hdr.tiling.col_start_sb[tile_col + 1] as i32;
        let row_sb_start = hdr.tiling.row_start_sb[tile_row] as i32;
        let row_sb_end = hdr.tiling.row_start_sb[tile_row + 1] as i32;
        let sb_shift = self.sb_shift;
        let tiling = TilePos {
            row: tile_row,
            col: tile_col,
            col_start: col_sb_start << sb_shift,
            col_end: (col_sb_end << sb_shift).min(self.bw),
            row_start: row_sb_start << sb_shift,
            row_end: (row_sb_end << sb_shift).min(self.bh),
        };
        let mut lr_ref = [None; 3];
        let (sb_idx, unit_idx);
        if hdr.width[0] != hdr.width[1] {
            sb_idx = (tiling.row_start >> 5) * self.sr_sb128w;
            unit_idx = (tiling.row_start & 16) >> 3;
        } else {
            sb_idx = (tiling.row_start >> 5) * self.sb128w + col_sb128_start;
            unit_idx = ((tiling.row_start & 16) >> 3) + ((tiling.col_start & 16) >> 4);
        }
        for (p, lr) in lr_ref.iter_mut().enumerate() {
            if (self.lf.restore_planes >> p) & 1 == 0 {
                continue;
            }
            let (si, ui);
            if hdr.width[0] != hdr.width[1] {
                let ss_hor = (p != 0 && self.layout != Layout::I444) as i32;
                let d = hdr.super_res_width_scale_denominator as i32;
                let unit_size_log2 = hdr.restoration.unit_size[(p != 0) as usize] as i32;
                let rnd = (8 << unit_size_log2) - 1;
                let shift = unit_size_log2 + 3;
                let x = ((4 * tiling.col_start * d >> ss_hor) + rnd) >> shift;
                let px_x = x << (unit_size_log2 + ss_hor);
                let u_idx = unit_idx + ((px_x & 64) >> 6);
                let sb128x = px_x >> 7;
                if sb128x >= self.sr_sb128w {
                    continue;
                }
                si = (sb_idx + sb128x) as usize;
                ui = u_idx as usize;
            } else {
                si = sb_idx as usize;
                ui = unit_idx as usize;
            }
            let u = &mut self.lf.lr_mask[si].lr[p][ui];
            u.filter_v = [3, -7, 15];
            u.filter_h = [3, -7, 15];
            u.sgr_weights = [-32, 31];
            *lr = Some((si, ui));
        }
        self.ts.push(TileState {
            cdf: Box::new(*in_cdf),
            msac: Msac::new(data, hdr.disable_cdf_update),
            tiling,
            dqmem: [[[0; 2]; 3]; 8],
            dq_frame: true,
            last_qidx: hdr.quant.yac as i32,
            last_delta_lf: [0; 4],
            lflvlmem: [[[[0; 2]; 8]; 4]; 8],
            lflvl_frame: true,
            lr_ref,
        });
    }

    fn read_restoration_info(&mut self, si: usize, ui: usize, p: usize, frame_type: u8) {
        let mut lr = self.lf.lr_mask[si].lr[p][ui];
        let (rsi, rui) = self.ts[self.tsi].lr_ref[p].unwrap();
        let rr = self.lf.lr_mask[rsi].lr[p][rui];
        if frame_type == RESTORATION_SWITCHABLE {
            let filter = sa!(self, 2, m.restore_switchable) as u8;
            lr.typ = filter + (filter != 0) as u8;
        } else {
            let t = if frame_type == RESTORATION_WIENER { ba!(self, m.restore_wiener) } else { ba!(self, m.restore_sgrproj) };
            lr.typ = if t { frame_type } else { RESTORATION_NONE };
        }
        if lr.typ == RESTORATION_WIENER {
            let ms = self.msac();
            lr.filter_v[0] = if p != 0 { 0 } else { (ms.subexp(rr.filter_v[0] as i32 + 5, 16, 1) - 5) as i8 };
            lr.filter_v[1] = (ms.subexp(rr.filter_v[1] as i32 + 23, 32, 2) - 23) as i8;
            lr.filter_v[2] = (ms.subexp(rr.filter_v[2] as i32 + 17, 64, 3) - 17) as i8;
            lr.filter_h[0] = if p != 0 { 0 } else { (ms.subexp(rr.filter_h[0] as i32 + 5, 16, 1) - 5) as i8 };
            lr.filter_h[1] = (ms.subexp(rr.filter_h[1] as i32 + 23, 32, 2) - 23) as i8;
            lr.filter_h[2] = (ms.subexp(rr.filter_h[2] as i32 + 17, 64, 3) - 17) as i8;
            lr.sgr_weights = rr.sgr_weights;
            self.ts[self.tsi].lr_ref[p] = Some((si, ui));
        } else if lr.typ == RESTORATION_SGRPROJ {
            let ms = self.msac();
            let idx = ms.bools(4) as usize;
            let sgr = SGR_PARAMS[idx];
            lr.typ += idx as u8;
            lr.sgr_weights[0] = if sgr[0] != 0 { (ms.subexp(rr.sgr_weights[0] as i32 + 96, 128, 4) - 96) as i8 } else { 0 };
            lr.sgr_weights[1] = if sgr[1] != 0 { (ms.subexp(rr.sgr_weights[1] as i32 + 32, 128, 4) - 32) as i8 } else { 95 };
            lr.filter_v = rr.filter_v;
            lr.filter_h = rr.filter_h;
            self.ts[self.tsi].lr_ref[p] = Some((si, ui));
        }
        self.lf.lr_mask[si].lr[p][ui] = lr;
    }

    fn decode_tile_sbrow(&mut self) -> DResult {
        let hdr = self.hdr.clone();
        let seq = self.seq.clone();
        let root_bl = if seq.sb128 { 0 } else { 1 };
        let sb_step = self.sb_step;
        let (tile_row, tile_col) = (self.ts[self.tsi].tiling.row, self.ts[self.tsi].tiling.col);
        let col_sb_start = hdr.tiling.col_start_sb[tile_col] as i32;
        let col_sb128_start = col_sb_start >> (!seq.sb128) as i32;
        let t = &self.ts[self.tsi].tiling;
        if hdr.is_inter_or_switch() || hdr.allow_intrabc {
            let rf = self.rf.as_ref().unwrap();
            self.rt = RefmvsTile { col_start: t.col_start, col_end: t.col_end.min(rf.iw4), row_start: t.row_start, row_end: t.row_end.min(rf.ih4) };
        }
        reset_context(&mut self.l, hdr.is_key_or_intra());
        self.pal_sz_uv[1] = [0; 32];
        let sb128y = self.by >> 5;
        let (cs, ce) = (t.col_start, t.col_end);
        self.ai = (col_sb128_start + tile_row as i32 * self.sb128w) as usize;
        self.lf_mask = (sb128y * self.sb128w + col_sb128_start) as usize;
        let tree = intra_edge::tree();
        self.bx = cs;
        while self.bx < ce {
            if root_bl == 0 {
                self.cdef_idx = 0;
                self.lf.mask[self.lf_mask].cdef_idx = [-1; 4];
            } else {
                self.cdef_idx = (((self.bx & 16) >> 4) + ((self.by & 16) >> 3)) as usize;
                self.lf.mask[self.lf_mask].cdef_idx[self.cdef_idx] = -1;
            }
            // Restoration units.
            for p in 0..3 {
                if (self.lf.restore_planes >> p) & 1 == 0 {
                    continue;
                }
                let ss_ver = (p != 0 && self.layout == Layout::I420) as i32;
                let ss_hor = (p != 0 && self.layout != Layout::I444) as i32;
                let unit_size_log2 = hdr.restoration.unit_size[(p != 0) as usize] as i32;
                let y = self.by * 4 >> ss_ver;
                let h = (self.cur.h + ss_ver) >> ss_ver;
                let unit_size = 1 << unit_size_log2;
                let mask = unit_size - 1;
                if y & mask != 0 {
                    continue;
                }
                let half_unit = unit_size >> 1;
                if y != 0 && y + half_unit > h {
                    continue;
                }
                let frame_type = hdr.restoration.typ[p];
                if hdr.width[0] != hdr.width[1] {
                    let w = (self.sr_cur.as_ref().unwrap().w + ss_hor) >> ss_hor;
                    let n_units = 1.max((w + half_unit) >> unit_size_log2);
                    let d = hdr.super_res_width_scale_denominator as i32;
                    let rnd = unit_size * 8 - 1;
                    let shift = unit_size_log2 + 3;
                    let x0 = ((4 * self.bx * d >> ss_hor) + rnd) >> shift;
                    let x1 = ((4 * (self.bx + sb_step) * d >> ss_hor) + rnd) >> shift;
                    for x in x0..x1.min(n_units) {
                        let px_x = x << (unit_size_log2 + ss_hor);
                        let sb_idx = (self.by >> 5) * self.sr_sb128w + (px_x >> 7);
                        let unit_idx = ((self.by & 16) >> 3) + ((px_x & 64) >> 6);
                        self.read_restoration_info(sb_idx as usize, unit_idx as usize, p, frame_type);
                    }
                } else {
                    let x = 4 * self.bx >> ss_hor;
                    if x & mask != 0 {
                        continue;
                    }
                    let w = (self.cur.w + ss_hor) >> ss_hor;
                    if x != 0 && x + half_unit > w {
                        continue;
                    }
                    let sb_idx = (self.by >> 5) * self.sr_sb128w + (self.bx >> 5);
                    let unit_idx = ((self.by & 16) >> 3) + ((self.bx & 16) >> 4);
                    self.read_restoration_info(sb_idx as usize, unit_idx as usize, p, frame_type);
                }
            }
            self.decode_sb(root_bl, tree.root[root_bl])?;
            if self.bx & 16 != 0 || seq.sb128 {
                self.ai += 1;
                self.lf_mask += 1;
            }
            self.bx += sb_step;
        }

        self.backup_ipred_edge();

        // Back up the right-edge tx_lpf contexts for the loop filter.
        let align_h = ((self.bh + 31) & !31) as usize;
        let by = self.by as usize;
        let o0 = align_h * tile_col + by;
        let src0 = (self.by & 16) as usize;
        let n = sb_step as usize;
        self.lf.tx_lpf_right_edge[0][o0..o0 + n].copy_from_slice(&self.l.tx_lpf_y[src0..src0 + n]);
        let ss_ver = self.ss_ver as usize;
        let align_h = align_h >> ss_ver;
        let o1 = align_h * tile_col + (by >> ss_ver);
        let src1 = ((self.by & 16) >> ss_ver) as usize;
        let n1 = n >> ss_ver;
        self.lf.tx_lpf_right_edge[1][o1..o1 + n1].copy_from_slice(&self.l.tx_lpf_uv[src1..src1 + n1]);

        if self.ts[self.tsi].msac.overread() {
            return Err("symbol decoder overread".into());
        }
        Ok(())
    }

    fn backup_ipred_edge(&mut self) {
        let ts = &self.ts[self.tsi];
        let sby = self.by >> self.sb_shift;
        let sby_off = (self.sb128w * 128 * sby) as usize;
        let x_off = ts.tiling.col_start as usize;
        let n = 4 * (ts.tiling.col_end as usize - x_off);
        let ys = self.cur.stride[0];
        let row = ((self.by + self.sb_step) * 4 - 1) as usize;
        let src = row * ys + x_off * 4;
        self.ipred_edge[0][sby_off + x_off * 4..sby_off + x_off * 4 + n].copy_from_slice(&self.cur.planes[0][src..src + n]);
        if self.layout != Layout::I400 {
            let (ss_hor, ss_ver) = (self.ss_hor as usize, self.ss_ver as usize);
            let uvs = self.cur.stride[1];
            let crow = (((self.by + self.sb_step) * 4) as usize >> ss_ver) - 1;
            let uv_off = (x_off * 4 >> ss_hor) + crow * uvs;
            let nc = n >> ss_hor;
            for pl in 1..=2 {
                let d = sby_off + (x_off * 4 >> ss_hor);
                self.ipred_edge[pl][d..d + nc].copy_from_slice(&self.cur.planes[pl][uv_off..uv_off + nc]);
            }
        }
    }

    /// Decode the frame's tiles with in-loop filtering (`dav1d_decode_frame`).
    pub fn decode_tiles(&mut self, in_cdf: &CdfContext, tiles: &[TileData]) -> Result<Option<CdfContext>, String> {
        let hdr = self.hdr.clone();
        let (mut tile_row, mut tile_col) = (0usize, 0usize);
        let mut update_set = false;
        for tg in tiles {
            let mut data: &[u8] = &tg.data;
            for j in tg.start..=tg.end {
                let tile_sz;
                if j == tg.end {
                    tile_sz = data.len();
                } else {
                    let nb = hdr.tiling.n_bytes as usize;
                    if nb > data.len() {
                        return Err("tile size overrun".into());
                    }
                    let mut s = 0usize;
                    for k in 0..nb {
                        s |= (data[k] as usize) << (k * 8);
                    }
                    data = &data[nb..];
                    tile_sz = s + 1;
                    if tile_sz > data.len() {
                        return Err("tile size overrun".into());
                    }
                }
                self.setup_tile(in_cdf, &data[..tile_sz], tile_row, tile_col);
                tile_col += 1;
                if tile_col == hdr.tiling.cols as usize {
                    tile_col = 0;
                    tile_row += 1;
                }
                if j == hdr.tiling.update as u32 && hdr.refresh_context {
                    update_set = true;
                }
                data = &data[tile_sz..];
            }
        }

        for a in self.a.iter_mut() {
            reset_context(a, hdr.is_key_or_intra());
        }
        for tile_row in 0..hdr.tiling.rows as usize {
            let sbh_end = (hdr.tiling.row_start_sb[tile_row + 1] as i32).min(self.sbh);
            for sby in hdr.tiling.row_start_sb[tile_row] as i32..sbh_end {
                self.by = sby << (4 + self.seq.sb128 as i32);
                let by_end = (self.by + self.sb_step) >> 1;
                if hdr.use_ref_frame_mvs {
                    let (bw, by) = (self.bw, self.by);
                    self.rf.as_mut().unwrap().load_tmvs(0, bw >> 1, by >> 1, by_end);
                }
                for tile_col in 0..hdr.tiling.cols as usize {
                    self.tsi = tile_row * hdr.tiling.cols as usize + tile_col;
                    self.decode_tile_sbrow()?;
                }
                if hdr.is_inter_or_switch() {
                    let (bw, by) = (self.bw, self.by);
                    self.rf.as_mut().unwrap().save_tmvs(0, bw >> 1, by >> 1, by_end);
                }
            }
        }
        self.apply_filters();
        if hdr.refresh_context && update_set {
            let mut out = *in_cdf;
            out.update_from(&self.ts[hdr.tiling.update as usize].cdf, hdr.is_key_or_intra());
            return Ok(Some(out));
        }
        Ok(None)
    }
}