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
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
//! In-loop filters: deblocking (dav1d `lf_apply_tmpl.c`, `loopfilter_tmpl.c`),
//! CDEF (`cdef_apply_tmpl.c`, `cdef_tmpl.c`), superres (`mc.resize`) and loop
//! restoration (`lr_apply_tmpl.c`, `looprestoration_tmpl.c`).
//!
//! dav1d interleaves these per superblock row behind the decoder with line
//! buffers that keep the inputs each stage would see in whole-frame order
//! (decoding never reads filtered pixels of the current frame). Here they
//! run as whole-frame passes after the tiles: deblock all rows, then CDEF
//! from a copy of the deblocked frame, then superres, then restoration per
//! 64-row stripe with the deblocked rows at the stripe edges. The
//! per-pixel arithmetic is dav1d's.

use super::decode::Fd;
use super::lf_mask::{Av1FilterLut, RestorationUnit};
use super::mc::{self, Bd};
use super::obu::*;
use super::picture::Picture;
use super::tables::{CDEF_DIRECTIONS, SGR_PARAMS, SGR_X_BY_X};

// ------------------------------------------------------------------ deblock

#[allow(clippy::too_many_arguments)]
fn loop_filter(d: &mut [u16], off: usize, e: i32, i: i32, h: i32, stridea: isize, strideb: isize, wd: i32, bd: Bd) {
    let bd8 = bd.bits as i32 - 8;
    let f = 1 << bd8;
    let (e, i, h) = (e << bd8, i << bd8, h << bd8);
    let maxp = bd.max;
    let mut o = off as isize;
    for _ in 0..4 {
        let px = |k: isize| d[(o + strideb * k) as usize] as i32;
        let p1 = px(-2);
        let p0 = px(-1);
        let q0 = px(0);
        let q1 = px(1);
        let (mut p2, mut q2, mut p3, mut q3) = (0, 0, 0, 0);
        let mut fm = (p1 - p0).abs() <= i && (q1 - q0).abs() <= i && (p0 - q0).abs() * 2 + ((p1 - q1).abs() >> 1) <= e;
        if wd > 4 {
            p2 = px(-3);
            q2 = px(2);
            fm &= (p2 - p1).abs() <= i && (q2 - q1).abs() <= i;
            if wd > 6 {
                p3 = px(-4);
                q3 = px(3);
                fm &= (p3 - p2).abs() <= i && (q3 - q2).abs() <= i;
            }
        }
        if !fm {
            o += stridea;
            continue;
        }
        let (mut p6, mut p5, mut p4, mut q4, mut q5, mut q6) = (0, 0, 0, 0, 0, 0);
        let mut flat8out = false;
        let mut flat8in = false;
        if wd >= 16 {
            p6 = px(-7);
            p5 = px(-6);
            p4 = px(-5);
            q4 = px(4);
            q5 = px(5);
            q6 = px(6);
            flat8out = (p6 - p0).abs() <= f && (p5 - p0).abs() <= f && (p4 - p0).abs() <= f && (q4 - q0).abs() <= f && (q5 - q0).abs() <= f && (q6 - q0).abs() <= f;
        }
        if wd >= 6 {
            flat8in = (p2 - p0).abs() <= f && (p1 - p0).abs() <= f && (q1 - q0).abs() <= f && (q2 - q0).abs() <= f;
        }
        if wd >= 8 {
            flat8in &= (p3 - p0).abs() <= f && (q3 - q0).abs() <= f;
        }
        let mut set = |k: isize, v: i32| d[(o + strideb * k) as usize] = v as u16;
        if wd >= 16 && flat8out && flat8in {
            set(-6, (p6 + p6 + p6 + p6 + p6 + p6 * 2 + p5 * 2 + p4 * 2 + p3 + p2 + p1 + p0 + q0 + 8) >> 4);
            set(-5, (p6 + p6 + p6 + p6 + p6 + p5 * 2 + p4 * 2 + p3 * 2 + p2 + p1 + p0 + q0 + q1 + 8) >> 4);
            set(-4, (p6 + p6 + p6 + p6 + p5 + p4 * 2 + p3 * 2 + p2 * 2 + p1 + p0 + q0 + q1 + q2 + 8) >> 4);
            set(-3, (p6 + p6 + p6 + p5 + p4 + p3 * 2 + p2 * 2 + p1 * 2 + p0 + q0 + q1 + q2 + q3 + 8) >> 4);
            set(-2, (p6 + p6 + p5 + p4 + p3 + p2 * 2 + p1 * 2 + p0 * 2 + q0 + q1 + q2 + q3 + q4 + 8) >> 4);
            set(-1, (p6 + p5 + p4 + p3 + p2 + p1 * 2 + p0 * 2 + q0 * 2 + q1 + q2 + q3 + q4 + q5 + 8) >> 4);
            set(0, (p5 + p4 + p3 + p2 + p1 + p0 * 2 + q0 * 2 + q1 * 2 + q2 + q3 + q4 + q5 + q6 + 8) >> 4);
            set(1, (p4 + p3 + p2 + p1 + p0 + q0 * 2 + q1 * 2 + q2 * 2 + q3 + q4 + q5 + q6 + q6 + 8) >> 4);
            set(2, (p3 + p2 + p1 + p0 + q0 + q1 * 2 + q2 * 2 + q3 * 2 + q4 + q5 + q6 + q6 + q6 + 8) >> 4);
            set(3, (p2 + p1 + p0 + q0 + q1 + q2 * 2 + q3 * 2 + q4 * 2 + q5 + q6 + q6 + q6 + q6 + 8) >> 4);
            set(4, (p1 + p0 + q0 + q1 + q2 + q3 * 2 + q4 * 2 + q5 * 2 + q6 + q6 + q6 + q6 + q6 + 8) >> 4);
            set(5, (p0 + q0 + q1 + q2 + q3 + q4 * 2 + q5 * 2 + q6 * 2 + q6 + q6 + q6 + q6 + q6 + 8) >> 4);
        } else if wd >= 8 && flat8in {
            set(-3, (p3 + p3 + p3 + 2 * p2 + p1 + p0 + q0 + 4) >> 3);
            set(-2, (p3 + p3 + p2 + 2 * p1 + p0 + q0 + q1 + 4) >> 3);
            set(-1, (p3 + p2 + p1 + 2 * p0 + q0 + q1 + q2 + 4) >> 3);
            set(0, (p2 + p1 + p0 + 2 * q0 + q1 + q2 + q3 + 4) >> 3);
            set(1, (p1 + p0 + q0 + 2 * q1 + q2 + q3 + q3 + 4) >> 3);
            set(2, (p0 + q0 + q1 + 2 * q2 + q3 + q3 + q3 + 4) >> 3);
        } else if wd == 6 && flat8in {
            set(-2, (p2 + 2 * p2 + 2 * p1 + 2 * p0 + q0 + 4) >> 3);
            set(-1, (p2 + 2 * p1 + 2 * p0 + 2 * q0 + q1 + 4) >> 3);
            set(0, (p1 + 2 * p0 + 2 * q0 + 2 * q1 + q2 + 4) >> 3);
            set(1, (p0 + 2 * q0 + 2 * q1 + 2 * q2 + q2 + 4) >> 3);
        } else {
            let hev = (p1 - p0).abs() > h || (q1 - q0).abs() > h;
            let cd = |v: i32| v.clamp(-128 * (1 << bd8), 128 * (1 << bd8) - 1);
            let cp = |v: i32| v.clamp(0, maxp);
            let lim = (128 << bd8) - 1;
            if hev {
                let mut ff = cd(p1 - q1);
                ff = cd(3 * (q0 - p0) + ff);
                let f1 = (ff + 4).min(lim) >> 3;
                let f2 = (ff + 3).min(lim) >> 3;
                set(-1, cp(p0 + f2));
                set(0, cp(q0 - f1));
            } else {
                let ff = cd(3 * (q0 - p0));
                let f1 = (ff + 4).min(lim) >> 3;
                let f2 = (ff + 3).min(lim) >> 3;
                set(-1, cp(p0 + f2));
                set(0, cp(q0 - f1));
                let ff = (f1 + 1) >> 1;
                set(-2, cp(p1 + ff));
                set(1, cp(q1 - ff));
            }
        }
        o += stridea;
    }
}

/// `loop_filter_{h,v}_sb128{y,uv}_c`. `dir` 0 = vertical edges (filter
/// along rows), 1 = horizontal edges. Levels are read at `lvl[li][comp]`
/// stepping `lstep`, the neighbour across the edge at `li + lprev`.
#[allow(clippy::too_many_arguments)]
fn lf_sb(d: &mut [u16], off: usize, stride: usize, vmask: [u32; 3], lvl: &[[u8; 4]], l0: usize, lstep: usize, lprev: isize, comp: usize, lut: &Av1FilterLut, uv: bool, dir: usize, bd: Bd) {
    let vm = vmask[0] | vmask[1] | vmask[2];
    let mut y = 1u32;
    let mut o = off;
    let mut li = l0;
    while vm & !(y.wrapping_sub(1)) != 0 {
        if vm & y != 0 {
            let cur = lvl[li][comp];
            let l = if cur != 0 { cur } else { lvl[(li as isize + lprev) as usize][comp] } as usize;
            if l != 0 {
                let h = (l >> 4) as i32;
                let e = lut.e[l] as i32;
                let i = lut.i[l] as i32;
                let wd = if uv {
                    4 + 2 * ((vmask[1] & y) != 0) as i32
                } else {
                    let idx = if vmask[2] & y != 0 { 2 } else { (vmask[1] & y != 0) as i32 };
                    4 << idx
                };
                if dir == 0 {
                    loop_filter(d, o, e, i, h, stride as isize, 1, wd, bd);
                } else {
                    loop_filter(d, o, e, i, h, 1, stride as isize, wd, bd);
                }
            }
        }
        if y == 1 << 31 {
            break;
        }
        y <<= 1;
        o += if dir == 0 { 4 * stride } else { 4 };
        li += lstep;
    }
}

// ------------------------------------------------------------------ cdef

fn constrain(diff: i32, threshold: i32, shift: i32) -> i32 {
    let adiff = diff.abs();
    let v = adiff.min(0.max(threshold - (adiff >> shift)));
    if diff < 0 {
        -v
    } else {
        v
    }
}

fn cdef_find_dir(img: &[u16], off: usize, stride: usize, bd8: i32) -> (i32, u32) {
    let mut hv = [[0i32; 8]; 2];
    let mut diag = [[0i32; 15]; 2];
    let mut alt = [[0i32; 11]; 4];
    for y in 0..8 {
        for x in 0..8 {
            let px = (img[off + y * stride + x] as i32 >> bd8) - 128;
            diag[0][y + x] += px;
            alt[0][y + (x >> 1)] += px;
            hv[0][y] += px;
            alt[1][3 + y - (x >> 1)] += px;
            diag[1][7 + y - x] += px;
            alt[2][3 - (y >> 1) + x] += px;
            hv[1][x] += px;
            alt[3][(y >> 1) + x] += px;
        }
    }
    let mut cost = [0u32; 8];
    for n in 0..8 {
        cost[2] = cost[2].wrapping_add((hv[0][n] * hv[0][n]) as u32);
        cost[6] = cost[6].wrapping_add((hv[1][n] * hv[1][n]) as u32);
    }
    cost[2] = cost[2].wrapping_mul(105);
    cost[6] = cost[6].wrapping_mul(105);
    const DIV: [u32; 7] = [840, 420, 280, 210, 168, 140, 120];
    for n in 0..7 {
        let d = DIV[n];
        cost[0] = cost[0].wrapping_add(((diag[0][n] * diag[0][n] + diag[0][14 - n] * diag[0][14 - n]) as u32).wrapping_mul(d));
        cost[4] = cost[4].wrapping_add(((diag[1][n] * diag[1][n] + diag[1][14 - n] * diag[1][14 - n]) as u32).wrapping_mul(d));
    }
    cost[0] = cost[0].wrapping_add((diag[0][7] * diag[0][7] * 105) as u32);
    cost[4] = cost[4].wrapping_add((diag[1][7] * diag[1][7] * 105) as u32);
    for n in 0..4 {
        let c = &mut cost[n * 2 + 1];
        for m in 0..5 {
            *c = c.wrapping_add((alt[n][3 + m] * alt[n][3 + m]) as u32);
        }
        *c = c.wrapping_mul(105);
        for m in 0..3 {
            let d = DIV[2 * m + 1];
            *c = c.wrapping_add(((alt[n][m] * alt[n][m] + alt[n][10 - m] * alt[n][10 - m]) as u32).wrapping_mul(d));
        }
    }
    let mut best_dir = 0;
    let mut best = cost[0];
    for n in 1..8 {
        if cost[n] > best {
            best = cost[n];
            best_dir = n;
        }
    }
    (best_dir as i32, best.wrapping_sub(cost[best_dir ^ 4]) >> 10)
}

fn adjust_strength(strength: i32, var: u32) -> i32 {
    if var == 0 {
        return 0;
    }
    let i = if var >> 6 != 0 { (31 - (var >> 6).leading_zeros() as i32).min(12) } else { 0 };
    (strength * (4 + i) + 8) >> 4
}

#[inline]
fn ulog2(v: i32) -> i32 {
    31 - (v as u32).leading_zeros() as i32
}

/// `cdef_filter_block_c` with the padded input `tmp` (12 wide, origin at
/// (2, 2), `i16::MIN` outside the frame).
#[allow(clippy::too_many_arguments)]
fn cdef_filter_block(dst: &mut [u16], off: usize, stride: usize, tmp: &[i32], pri: i32, sec: i32, dir: i32, damping: i32, w: usize, h: usize, bd8: i32) {
    const TS: isize = 12;
    let t = |y: usize, x: usize, o: isize| tmp[((y as isize + 2) * TS + x as isize + 2 + o) as usize];
    let dirs = |k: i32, j: usize| {
        
        CDEF_DIRECTIONS[k as usize][j] as isize
    };
    let umin = |a: i32, b: i32| if (a as u32) < (b as u32) { a } else { b };
    if pri != 0 {
        let pri_tap = 4 - ((pri >> bd8) & 1);
        let pri_shift = 0.max(damping - ulog2(pri));
        if sec != 0 {
            let sec_shift = damping - ulog2(sec);
            for y in 0..h {
                for x in 0..w {
                    let px = dst[off + y * stride + x] as i32;
                    let mut sum = 0;
                    let (mut max, mut min) = (px, px);
                    let mut ptk = pri_tap;
                    for k in 0..2 {
                        let off1 = dirs(dir + 2, k);
                        let p0 = t(y, x, off1);
                        let p1 = t(y, x, -off1);
                        sum += ptk * constrain(p0 - px, pri, pri_shift);
                        sum += ptk * constrain(p1 - px, pri, pri_shift);
                        ptk = (ptk & 3) | 2;
                        min = umin(p0, min);
                        max = max.max(p0);
                        min = umin(p1, min);
                        max = max.max(p1);
                        let off2 = dirs(dir + 4, k);
                        let off3 = dirs(dir, k);
                        let s = [t(y, x, off2), t(y, x, -off2), t(y, x, off3), t(y, x, -off3)];
                        let st = 2 - k as i32;
                        for &sv in &s {
                            sum += st * constrain(sv - px, sec, sec_shift);
                            min = umin(sv, min);
                            max = max.max(sv);
                        }
                    }
                    dst[off + y * stride + x] = (px + ((sum - (sum < 0) as i32 + 8) >> 4)).clamp(min, max) as u16;
                }
            }
        } else {
            for y in 0..h {
                for x in 0..w {
                    let px = dst[off + y * stride + x] as i32;
                    let mut sum = 0;
                    let mut ptk = pri_tap;
                    for k in 0..2 {
                        let o = dirs(dir + 2, k);
                        sum += ptk * constrain(t(y, x, o) - px, pri, pri_shift);
                        sum += ptk * constrain(t(y, x, -o) - px, pri, pri_shift);
                        ptk = (ptk & 3) | 2;
                    }
                    dst[off + y * stride + x] = (px + ((sum - (sum < 0) as i32 + 8) >> 4)) as u16;
                }
            }
        }
    } else {
        let sec_shift = damping - ulog2(sec);
        for y in 0..h {
            for x in 0..w {
                let px = dst[off + y * stride + x] as i32;
                let mut sum = 0;
                for k in 0..2 {
                    let o1 = dirs(dir + 4, k);
                    let o2 = dirs(dir, k);
                    let st = 2 - k as i32;
                    for &sv in &[t(y, x, o1), t(y, x, -o1), t(y, x, o2), t(y, x, -o2)] {
                        sum += st * constrain(sv - px, sec, sec_shift);
                    }
                }
                dst[off + y * stride + x] = (px + ((sum - (sum < 0) as i32 + 8) >> 4)) as u16;
            }
        }
    }
}

/// `padding()` for CDEF: a 12x12 window (origin (2,2)) of the deblocked
/// frame, `i16::MIN` where an edge is unavailable.
fn fill_cdef_tmp(tmp: &mut [i32], src: &[u16], stride: usize, off: usize, w: usize, h: usize, (have_left, have_right, have_top, have_bottom): (bool, bool, bool, bool)) {
    for v in tmp.iter_mut() {
        *v = i16::MIN as i32;
    }
    let y0 = if have_top { -2 } else { 0 };
    let y1 = h as isize + if have_bottom { 2 } else { 0 };
    let x0 = if have_left { -2 } else { 0 };
    let x1 = w as isize + if have_right { 2 } else { 0 };
    for y in y0..y1 {
        for x in x0..x1 {
            tmp[((y + 2) * 12 + x + 2) as usize] = src[(off as isize + y * stride as isize + x) as usize] as i32;
        }
    }
}

// ------------------------------------------------------------------ restoration

const LR_HAVE_LEFT: u32 = 1;
const LR_HAVE_RIGHT: u32 = 2;
const LR_HAVE_TOP: u32 = 4;
const LR_HAVE_BOTTOM: u32 = 8;

/// A stripe of one restoration unit with its surroundings resolved per the
/// edge flags: `at(x, y)` for x in [-4, w+4), y in [-3, h+3).
struct StripeSrc {
    buf: Vec<i32>,
    w: usize,
}

impl StripeSrc {
    #[inline]
    fn at(&self, x: isize, y: isize) -> i32 {
        self.buf[((y + 3) * (self.w as isize + 8) + x + 4) as usize]
    }
}

/// Resolve the samples a restoration unit's stripe reads: the stripe
/// itself (CDEF output), the two deblocked rows above and below it
/// (dav1d's `lr_lpf_line`), replicating where an edge is missing.
#[allow(clippy::too_many_arguments)]
fn build_stripe(cdef: &[u16], lpf: &[u16], stride: usize, lpf_stride: usize, x0: i32, y0: i32, w: i32, h: i32, plane_w: i32, plane_h: i32, edges: u32) -> StripeSrc {
    let bw = w as usize + 8;
    let mut buf = vec![0i32; bw * (h as usize + 6)];
    for yy in -3..h + 3 {
        let (rows, rstride, ry): (&[u16], usize, i32) = if yy < 0 {
            if edges & LR_HAVE_TOP != 0 {
                (lpf, lpf_stride, y0 + yy.max(-2))
            } else {
                (cdef, stride, y0)
            }
        } else if yy >= h {
            if edges & LR_HAVE_BOTTOM != 0 {
                (lpf, lpf_stride, (y0 + h + (yy - h).min(1)).min(plane_h - 1))
            } else {
                (cdef, stride, y0 + h - 1)
            }
        } else {
            (cdef, stride, y0 + yy)
        };
        for xx in -4..w + 4 {
            let mut ax = x0 + xx;
            if xx < 0 && edges & LR_HAVE_LEFT == 0 {
                ax = x0;
            }
            if xx >= w && edges & LR_HAVE_RIGHT == 0 {
                ax = x0 + w - 1;
            }
            let ax = ax.clamp(0, plane_w + 3);
            buf[(yy + 3) as usize * bw + (xx + 4) as usize] = rows[ry as usize * rstride + ax as usize] as i32;
        }
    }
    StripeSrc { buf, w: w as usize }
}

#[allow(clippy::too_many_arguments)]
fn filter_unit(dst: &mut [u16], stride: usize, x0: usize, y0: usize, w: usize, h: usize, src: &StripeSrc, lr: &RestorationUnit, bitdepth_max: i32) {
    let bitdepth = 32 - (bitdepth_max as u32).leading_zeros() as i32;
    if lr.typ == RESTORATION_WIENER {
        let mut fh = [0i32; 7];
        let mut fv = [0i32; 7];
        for k in 0..3 {
            fh[k] = lr.filter_h[k] as i32;
            fh[6 - k] = fh[k];
            fv[k] = lr.filter_v[k] as i32;
            fv[6 - k] = fv[k];
        }
        // dav1d adds the centre +128 into the tap (16 bpc) or as
        // `src * 128` (8 bpc): the same sum.
        fh[3] = -(fh[0] + fh[1] + fh[2]) * 2 + 128;
        fv[3] = 128 - (fv[0] + fv[1] + fv[2]) * 2;
        let round_bits_h = 3 + (bitdepth == 12) as i32 * 2;
        let rounding_off_h = 1 << (round_bits_h - 1);
        let clip_limit = 1 << (bitdepth + 1 + 7 - round_bits_h);
        let round_bits_v = 11 - (bitdepth == 12) as i32 * 2;
        let rounding_off_v = 1 << (round_bits_v - 1);
        let round_offset = 1 << (bitdepth + (round_bits_v - 1));
        let rows = h + 6;
        let mut hor = vec![0i32; rows * w];
        for r in 0..rows {
            let yy = r as isize - 3;
            for x in 0..w {
                let mut sum = 1 << (bitdepth + 6);
                for i in 0..7 {
                    sum += src.at(x as isize + i as isize - 3, yy) * fh[i];
                }
                hor[r * w + x] = ((sum + rounding_off_h) >> round_bits_h).clamp(0, clip_limit - 1);
            }
        }
        for y in 0..h {
            for x in 0..w {
                let mut sum = -round_offset;
                for k in 0..7 {
                    sum += hor[(y + k) * w + x] * fv[k];
                }
                dst[(y0 + y) * stride + x0 + x] = ((sum + rounding_off_v) >> round_bits_v).clamp(0, bitdepth_max) as u16;
            }
        }
        return;
    }
    let idx = (lr.typ - RESTORATION_SGRPROJ) as usize;
    let s0 = SGR_PARAMS[idx][0] as u32;
    let s1 = SGR_PARAMS[idx][1] as u32;
    let w0 = lr.sgr_weights[0] as i32;
    let w1 = 128 - (lr.sgr_weights[0] as i32 + lr.sgr_weights[1] as i32);
    let bd8 = bitdepth - 8;
    let cols = w + 2;
    // (A, B) of `sgr_calc_row_ab` at rows -1..=h, columns -1..=w.
    let ab = |r: isize, n: i32, s: u32, one_by_x: u32, odd_only: bool| -> (Vec<i32>, Vec<i32>) {
        let mut a = vec![0i32; (h + 2) * cols];
        let mut b = vec![0i32; (h + 2) * cols];
        for i in -1..=h as isize {
            if odd_only && i & 1 == 0 {
                continue;
            }
            for j in -1..=w as isize {
                let (mut ss, mut sq) = (0i32, 0i32);
                for dy in -r..=r {
                    for dx in -r..=r {
                        let c = src.at(j + dx, i + dy);
                        ss += c;
                        sq += c * c;
                    }
                }
                let aa = (sq + ((1 << (2 * bd8)) >> 1)) >> (2 * bd8);
                let bb = (ss + ((1 << bd8) >> 1)) >> bd8;
                let p = (aa * n - bb * bb).max(0) as u32;
                let z = (p.wrapping_mul(s).wrapping_add(1 << 19)) >> 20;
                let x = SGR_X_BY_X[z.min(255) as usize] as u32;
                let o = (i + 1) as usize * cols + (j + 1) as usize;
                a[o] = ((x.wrapping_mul(ss as u32).wrapping_mul(one_by_x)).wrapping_add(1 << 11) >> 12) as i32;
                b[o] = x as i32;
            }
        }
        (a, b)
    };
    let at = |v: &[i32], i: isize, j: isize| v[(i + 1) as usize * cols + (j + 1) as usize];
    let mut t5 = vec![0i32; w * h];
    let mut t3 = vec![0i32; w * h];
    if s0 != 0 {
        let (a, b) = ab(2, 25, s0, 164, true);
        for y in 0..h as isize {
            for x in 0..w as isize {
                let px = src.at(x, y);
                let v = if y & 1 == 0 {
                    let six = |p: &[i32]| (at(p, y - 1, x) + at(p, y + 1, x)) * 6 + (at(p, y - 1, x - 1) + at(p, y + 1, x - 1) + at(p, y - 1, x + 1) + at(p, y + 1, x + 1)) * 5;
                    (six(&a) - six(&b) * px + (1 << 8)) >> 9
                } else {
                    let three = |p: &[i32]| at(p, y, x) * 6 + (at(p, y, x - 1) + at(p, y, x + 1)) * 5;
                    (three(&a) - three(&b) * px + (1 << 7)) >> 8
                };
                t5[y as usize * w + x as usize] = v;
            }
        }
    }
    if s1 != 0 {
        let (a, b) = ab(1, 9, s1, 455, false);
        for y in 0..h as isize {
            for x in 0..w as isize {
                let px = src.at(x, y);
                let eight = |p: &[i32]| {
                    (at(p, y, x) + at(p, y, x - 1) + at(p, y, x + 1) + at(p, y - 1, x) + at(p, y + 1, x)) * 4
                        + (at(p, y - 1, x - 1) + at(p, y + 1, x - 1) + at(p, y - 1, x + 1) + at(p, y + 1, x + 1)) * 3
                };
                t3[y as usize * w + x as usize] = (eight(&a) - eight(&b) * px + (1 << 8)) >> 9;
            }
        }
    }
    for y in 0..h {
        for x in 0..w {
            let d = &mut dst[(y0 + y) * stride + x0 + x];
            let v = if s0 != 0 && s1 != 0 {
                w0 * t5[y * w + x] + w1 * t3[y * w + x]
            } else if s0 != 0 {
                w0 * t5[y * w + x]
            } else {
                w1 * t3[y * w + x]
            };
            *d = (*d as i32 + ((v + (1 << 10)) >> 11)).clamp(0, bitdepth_max) as u16;
        }
    }
}

impl Fd {
    fn deblock_sbrow_cols(&mut self, sby: i32) {
        let seq = self.seq.clone();
        let hdr = self.hdr.clone();
        let is_sb64 = !seq.sb128 as i32;
        let starty4 = (sby & is_sb64) << 4;
        let sbsz = 32 >> is_sb64;
        let sbl2 = 5 - is_sb64;
        let halign = ((self.bh + 31) & !31) as usize;
        let ss_ver = self.ss_ver;
        let ss_hor = self.ss_hor;
        let vmask_bits = 16 >> ss_ver;
        let hmask_bits = 16 >> ss_hor;
        let vmax = 1u32 << vmask_bits;
        let hmax = 1u32 << hmask_bits;
        let endy4 = (starty4 + (self.h4 - sby * sbsz).min(sbsz)) as u32;
        let uv_endy4 = (endy4 + ss_ver as u32) >> ss_ver;
        let mask_base = ((sby >> (!seq.sb128) as i32) * self.sb128w) as usize;

        // Tile column boundaries.
        let mut lpf_y = (sby << sbl2) as usize;
        let mut lpf_uv = (sby << (sbl2 - ss_ver)) as usize;
        let mut tile_col = 1;
        loop {
            let mut x = hdr.tiling.col_start_sb[tile_col] as i32;
            if (x << sbl2) >= self.bw {
                break;
            }
            let bx4 = if x & is_sb64 != 0 { 16 } else { 0 };
            let cbx4 = bx4 >> ss_hor;
            x >>= is_sb64;
            let lf = &mut self.lf;
            let m = &mut lf.mask[mask_base + x as usize];
            let mut mask = 1u32 << starty4;
            for y in starty4 as u32..endy4 {
                let sidx = (mask >= 0x10000) as usize;
                let smask = (mask >> (sidx << 4)) as u16;
                let yh = &mut m.filter_y[0][bx4 as usize];
                let idx = 2 * (yh[2][sidx] & smask != 0) as usize + (yh[1][sidx] & smask != 0) as usize;
                yh[2][sidx] &= !smask;
                yh[1][sidx] &= !smask;
                yh[0][sidx] &= !smask;
                let e = lf.tx_lpf_right_edge[0][lpf_y + (y - starty4 as u32) as usize] as usize;
                yh[idx.min(e)][sidx] |= smask;
                mask <<= 1;
            }
            if self.layout != Layout::I400 {
                let mut uv_mask = 1u32 << (starty4 >> ss_ver);
                for y in (starty4 >> ss_ver) as u32..uv_endy4 {
                    let sidx = (uv_mask >= vmax) as usize;
                    let smask = (uv_mask >> (sidx << (4 - ss_ver))) as u16;
                    let uh = &mut m.filter_uv[0][cbx4 as usize];
                    let idx = (uh[1][sidx] & smask != 0) as usize;
                    uh[1][sidx] &= !smask;
                    uh[0][sidx] &= !smask;
                    let e = lf.tx_lpf_right_edge[1][lpf_uv + (y - (starty4 >> ss_ver) as u32) as usize] as usize;
                    uh[idx.min(e)][sidx] |= smask;
                    uv_mask <<= 1;
                }
            }
            lpf_y += halign;
            lpf_uv += halign >> ss_ver;
            tile_col += 1;
        }

        // Tile row boundaries.
        let start_of_tile_row = self.lf.start_of_tile_row[sby as usize] as usize;
        if start_of_tile_row != 0 {
            for x in 0..self.sb128w as usize {
                let a = &self.a[self.sb128w as usize * (start_of_tile_row - 1) + x];
                let m = &mut self.lf.mask[mask_base + x];
                let w = 32.min(self.w4 - ((x as i32) << 5)) as u32;
                let mut mask = 1u32;
                for i in 0..w {
                    let sidx = (mask >= 0x10000) as usize;
                    let smask = (mask >> (sidx << 4)) as u16;
                    let yv = &mut m.filter_y[1][starty4 as usize];
                    let idx = 2 * (yv[2][sidx] & smask != 0) as usize + (yv[1][sidx] & smask != 0) as usize;
                    yv[2][sidx] &= !smask;
                    yv[1][sidx] &= !smask;
                    yv[0][sidx] &= !smask;
                    yv[idx.min(a.tx_lpf_y[i as usize] as usize)][sidx] |= smask;
                    mask <<= 1;
                }
                if self.layout != Layout::I400 {
                    let cw = (w + ss_hor as u32) >> ss_hor;
                    let mut uv_mask = 1u32;
                    for i in 0..cw {
                        let sidx = (uv_mask >= hmax) as usize;
                        let smask = (uv_mask >> (sidx << (4 - ss_hor))) as u16;
                        let uv = &mut m.filter_uv[1][(starty4 >> ss_ver) as usize];
                        let idx = (uv[1][sidx] & smask != 0) as usize;
                        uv[1][sidx] &= !smask;
                        uv[0][sidx] &= !smask;
                        uv[idx.min(a.tx_lpf_uv[i as usize] as usize)][sidx] |= smask;
                        uv_mask <<= 1;
                    }
                }
            }
        }

        let bd = Bd::new(self.bitdepth_max);
        let ys = self.cur.stride[0];
        let b4s = self.b4_stride as usize;
        let row0 = (sby * self.sb_step * 4) as usize;
        let lvl_row = (self.b4_stride * sby * sbsz) as usize;
        for x in 0..self.sb128w as usize {
            let have_left = x > 0;
            let w = 32.min(self.w4 - x as i32 * 32) as usize;
            let m = self.lf.mask[mask_base + x];
            for xx in 0..w {
                if !have_left && xx == 0 {
                    continue;
                }
                let mut hm = [0u32; 3];
                for k in 0..3 {
                    hm[k] = if starty4 == 0 {
                        let mut v = m.filter_y[0][xx][k][0] as u32;
                        if endy4 > 16 {
                            v |= (m.filter_y[0][xx][k][1] as u32) << 16;
                        }
                        v
                    } else {
                        m.filter_y[0][xx][k][1] as u32
                    };
                }
                let off = row0 * ys + x * 128 + xx * 4;
                lf_sb(&mut self.cur.planes[0], off, ys, hm, &self.lf.level, lvl_row + x * 32 + xx, b4s, -1, 0, &self.lf.lim_lut, false, 0, bd);
            }
        }
        if hdr.loopfilter.level_u == 0 && hdr.loopfilter.level_v == 0 {
            return;
        }
        let uvs = self.cur.stride[1];
        let crow0 = row0 >> ss_ver;
        let lvl_row = (self.b4_stride * (sby * sbsz >> ss_ver)) as usize;
        let sy = starty4 >> ss_ver;
        for x in 0..self.sb128w as usize {
            let have_left = x > 0;
            let w = ((32.min(self.w4 - x as i32 * 32) + ss_hor) >> ss_hor) as usize;
            let m = self.lf.mask[mask_base + x];
            for xx in 0..w {
                if !have_left && xx == 0 {
                    continue;
                }
                let mut hm = [0u32; 3];
                for k in 0..2 {
                    hm[k] = if sy == 0 {
                        let mut v = m.filter_uv[0][xx][k][0] as u32;
                        if uv_endy4 > (16 >> ss_ver) as u32 {
                            v |= (m.filter_uv[0][xx][k][1] as u32) << (16 >> ss_ver);
                        }
                        v
                    } else {
                        m.filter_uv[0][xx][k][1] as u32
                    };
                }
                let off = crow0 * uvs + x * (128 >> ss_hor) + xx * 4;
                let l0 = lvl_row + x * (32 >> ss_hor) + xx;
                for (pl, comp) in [(1usize, 2usize), (2, 3)] {
                    lf_sb(&mut self.cur.planes[pl], off, uvs, hm, &self.lf.level, l0, b4s, -1, comp, &self.lf.lim_lut, true, 0, bd);
                }
            }
        }
    }

    fn deblock_sbrow_rows(&mut self, sby: i32) {
        let seq = self.seq.clone();
        let hdr = self.hdr.clone();
        let have_top = sby > 0;
        let is_sb64 = !seq.sb128 as i32;
        let starty4 = (sby & is_sb64) << 4;
        let sbsz = 32 >> is_sb64;
        let ss_ver = self.ss_ver;
        let ss_hor = self.ss_hor;
        let endy4 = starty4 + (self.h4 - sby * sbsz).min(sbsz);
        let uv_endy4 = (endy4 + ss_ver) >> ss_ver;
        let mask_base = ((sby >> (!seq.sb128) as i32) * self.sb128w) as usize;
        let bd = Bd::new(self.bitdepth_max);
        let ys = self.cur.stride[0];
        let b4s = self.b4_stride as usize;
        let row0 = (sby * self.sb_step * 4) as usize;
        let lvl_row = (self.b4_stride * sby * sbsz) as usize;
        for x in 0..self.sb128w as usize {
            let m = self.lf.mask[mask_base + x];
            for (k, y) in (starty4..endy4).enumerate() {
                if !have_top && y == 0 {
                    continue;
                }
                let yy = y as usize;
                let vm = [0, 1, 2].map(|t| m.filter_y[1][yy][t][0] as u32 | ((m.filter_y[1][yy][t][1] as u32) << 16));
                let off = (row0 + k * 4) * ys + x * 128;
                let l0 = lvl_row + k * b4s + x * 32;
                lf_sb(&mut self.cur.planes[0], off, ys, vm, &self.lf.level, l0, 1, -(b4s as isize), 1, &self.lf.lim_lut, false, 1, bd);
            }
        }
        if hdr.loopfilter.level_u == 0 && hdr.loopfilter.level_v == 0 {
            return;
        }
        let uvs = self.cur.stride[1];
        let crow0 = row0 >> ss_ver;
        let lvl_row = (self.b4_stride * (sby * sbsz >> ss_ver)) as usize;
        for x in 0..self.sb128w as usize {
            let m = self.lf.mask[mask_base + x];
            for (k, y) in ((starty4 >> ss_ver)..uv_endy4).enumerate() {
                if !have_top && y == 0 {
                    continue;
                }
                let yy = y as usize;
                let sh = 16 >> ss_hor;
                let vm = [m.filter_uv[1][yy][0][0] as u32 | ((m.filter_uv[1][yy][0][1] as u32) << sh), m.filter_uv[1][yy][1][0] as u32 | ((m.filter_uv[1][yy][1][1] as u32) << sh), 0];
                let off = (crow0 + k * 4) * uvs + x * (128 >> ss_hor);
                let l0 = lvl_row + k * b4s + x * (32 >> ss_hor);
                for (pl, comp) in [(1usize, 2usize), (2, 3)] {
                    lf_sb(&mut self.cur.planes[pl], off, uvs, vm, &self.lf.level, l0, 1, -(b4s as isize), comp, &self.lf.lim_lut, true, 1, bd);
                }
            }
        }
    }

    fn cdef_frame(&mut self, src: &Picture) {
        let hdr = self.hdr.clone();
        let bd8 = self.bpc as i32 - 8;
        let damping = hdr.cdef.damping as i32 + bd8;
        let layout = self.layout;
        let uv_dirs: [[i32; 8]; 2] = [[0, 1, 2, 3, 4, 5, 6, 7], [7, 0, 2, 4, 5, 6, 6, 6]];
        let uv_dir = uv_dirs[(layout == Layout::I422) as usize];
        let (ss_hor, ss_ver) = (self.ss_hor as usize, self.ss_ver as usize);
        let sb64w = (self.sb128w << 1) as usize;
        let mut tmp = vec![0i32; 12 * 12];
        let mut by = 0;
        while by < self.bh {
            let have_top = by > 0;
            let have_bottom = by + 2 < self.bh;
            for sbx in 0..sb64w {
                let sb128x = sbx >> 1;
                let sb64_idx = (((by & 16) >> 3) as usize) + (sbx & 1);
                let mask_i = (by >> 5) as usize * self.sb128w as usize + sb128x;
                let cdef_idx = self.lf.mask[mask_i].cdef_idx[sb64_idx];
                if cdef_idx == -1 {
                    continue;
                }
                let ci = cdef_idx as usize;
                let y_lvl = hdr.cdef.y_strength[ci] as i32;
                let uv_lvl = hdr.cdef.uv_strength[ci] as i32;
                if y_lvl == 0 && uv_lvl == 0 {
                    continue;
                }
                let by_idx = ((by & 30) >> 1) as usize;
                let nr = self.lf.mask[mask_i].noskip_mask[by_idx];
                let noskip = (nr[1] as u32) << 16 | nr[0] as u32;
                let y_pri = (y_lvl >> 2) << bd8;
                let mut y_sec = y_lvl & 3;
                y_sec += (y_sec == 3) as i32;
                y_sec <<= bd8;
                let uv_pri = (uv_lvl >> 2) << bd8;
                let mut uv_sec = uv_lvl & 3;
                uv_sec += (uv_sec == 3) as i32;
                uv_sec <<= bd8;
                let mut bx = sbx as i32 * 16;
                while bx < ((sbx as i32 + 1) * 16).min(self.bw) {
                    let have_left = bx > 0;
                    let have_right = bx + 2 < self.bw;
                    let bx_mask = 3u32 << (bx & 30);
                    if noskip & bx_mask == 0 {
                        bx += 2;
                        continue;
                    }
                    let ys = src.stride[0];
                    let yoff = (by as usize * 4) * ys + bx as usize * 4;
                    let (dir, var) = if y_pri != 0 || uv_pri != 0 { cdef_find_dir(&src.planes[0], yoff, ys, bd8) } else { (0, 0) };
                    let edges = (have_left, have_right, have_top, have_bottom);
                    if y_pri != 0 {
                        let adj = adjust_strength(y_pri, var);
                        if adj != 0 || y_sec != 0 {
                            fill_cdef_tmp(&mut tmp, &src.planes[0], ys, yoff, 8, 8, edges);
                            cdef_filter_block(&mut self.cur.planes[0], yoff, ys, &tmp, adj, y_sec, dir, damping, 8, 8, bd8);
                        }
                    } else if y_sec != 0 {
                        fill_cdef_tmp(&mut tmp, &src.planes[0], ys, yoff, 8, 8, edges);
                        cdef_filter_block(&mut self.cur.planes[0], yoff, ys, &tmp, 0, y_sec, 0, damping, 8, 8, bd8);
                    }
                    if uv_lvl != 0 && layout != Layout::I400 {
                        let uvdir = if uv_pri != 0 { uv_dir[dir as usize] } else { 0 };
                        let uvs = src.stride[1];
                        let (cw, ch) = (8 >> ss_hor, 8 >> ss_ver);
                        let coff = ((by as usize * 4) >> ss_ver) * uvs + ((bx as usize * 4) >> ss_hor);
                        for pl in 1..3 {
                            fill_cdef_tmp(&mut tmp, &src.planes[pl], uvs, coff, cw, ch, edges);
                            cdef_filter_block(&mut self.cur.planes[pl], coff, uvs, &tmp, uv_pri, uv_sec, uvdir, damping - 1, cw, ch, bd8);
                        }
                    }
                    bx += 2;
                }
            }
            by += 2;
        }
    }

    fn superres(&mut self) {
        let bd = Bd::new(self.bitdepth_max);
        let sr = self.sr_cur.as_mut().unwrap();
        let np = if self.layout == Layout::I400 { 1 } else { 3 };
        for pl in 0..np {
            let (ss_hor, ss_ver) = self.cur.ss(pl);
            let dst_w = ((sr.w + ss_hor) >> ss_hor) as usize;
            let src_w = (4 * self.bw + ss_hor) >> ss_hor;
            let h = ((self.cur.h + ss_ver) >> ss_ver) as usize;
            let (ds, ss) = (sr.pstride(pl), self.cur.pstride(pl));
            let step = self.resize_step[(pl != 0) as usize];
            let start = self.resize_start[(pl != 0) as usize];
            for y in 0..h {
                let (d, s) = (&mut sr.planes[pl][y * ds..y * ds + dst_w], &self.cur.planes[pl][y * ss..]);
                mc::resize_row(d, s, dst_w, src_w, step, start, bd);
            }
        }
    }

    /// Deblocked rows for loop restoration (upscaled with superres).
    fn lr_source_rows(&self, deblocked: &Picture) -> Picture {
        match &self.sr_cur {
            None => deblocked.clone(),
            Some(sr) => {
                let bd = Bd::new(self.bitdepth_max);
                let mut out = Picture::new(sr.w, sr.h, sr.layout, sr.bpc);
                let np = if self.layout == Layout::I400 { 1 } else { 3 };
                for pl in 0..np {
                    let (ss_hor, ss_ver) = deblocked.ss(pl);
                    let dst_w = ((sr.w + ss_hor) >> ss_hor) as usize;
                    let src_w = (4 * self.bw) >> ss_hor;
                    let h = ((deblocked.h + ss_ver) >> ss_ver) as usize;
                    let (ds, ss) = (out.pstride(pl), deblocked.pstride(pl));
                    let step = self.resize_step[ss_hor as usize];
                    let start = self.resize_start[ss_hor as usize];
                    for y in 0..h {
                        let (d, s) = (&mut out.planes[pl][y * ds..y * ds + dst_w], &deblocked.planes[pl][y * ss..]);
                        mc::resize_row(d, s, dst_w, src_w, step, start, bd);
                    }
                }
                out
            }
        }
    }

    fn lr_frame(&mut self, lpf: &Picture) {
        let seq = self.seq.clone();
        let hdr = self.hdr.clone();
        let superres = hdr.width[0] != hdr.width[1];
        let mut pic = if superres { self.sr_cur.take().unwrap() } else { std::mem::replace(&mut self.cur, Picture::new(1, 1, self.layout, self.bpc)) };
        for plane in 0..3 {
            if (self.lf.restore_planes >> plane) & 1 == 0 {
                continue;
            }
            let (ss_hor, ss_ver) = pic.ss(plane);
            let h = (pic.h + ss_ver) >> ss_ver;
            let w = (pic.w + ss_hor) >> ss_hor;
            let stride = pic.pstride(plane);
            // Every unit reads the unrestored CDEF output.
            let cdef_out = pic.planes[plane].clone();
            let unit_size_log2 = hdr.restoration.unit_size[(plane != 0) as usize] as i32;
            let unit_size = 1 << unit_size_log2;
            let half_unit = unit_size >> 1;
            let max_unit = unit_size + half_unit;
            let shift_hor = 7 - ss_hor;
            for sby in 0..self.sbh {
                let not_last = sby + 1 < self.sbh;
                let next_row_y = (sby + 1) << ((6 - ss_ver) + seq.sb128 as i32);
                let row_h = (next_row_y - (8 >> ss_ver) * not_last as i32).min(h);
                let offset = (8 * (sby != 0) as i32) >> ss_ver;
                let y = (sby << ((6 - ss_ver) + seq.sb128 as i32)) - offset;
                let row_y = y + ((8 >> ss_ver) * (y != 0) as i32);
                let mut aligned_unit_pos = row_y & !(unit_size - 1);
                if aligned_unit_pos != 0 && aligned_unit_pos + half_unit > h {
                    aligned_unit_pos -= unit_size;
                }
                aligned_unit_pos <<= ss_ver;
                let sb_idx = ((aligned_unit_pos >> 7) * self.sr_sb128w) as usize;
                let unit_idx = (((aligned_unit_pos >> 6) & 1) << 1) as usize;
                let mut units: Vec<(i32, i32, RestorationUnit, u32)> = Vec::new();
                let mut edges = (if y > 0 { LR_HAVE_TOP } else { 0 }) | LR_HAVE_RIGHT;
                let mut x = 0;
                let mut cur_lr = self.lf.lr_mask[sb_idx].lr[plane][unit_idx];
                while x + max_unit <= w {
                    let next_x = x + unit_size;
                    let next_u_idx = unit_idx + ((next_x >> (shift_hor - 1)) & 1) as usize;
                    let nlr = self.lf.lr_mask[sb_idx + (next_x >> shift_hor) as usize].lr[plane][next_u_idx];
                    if cur_lr.typ != RESTORATION_NONE {
                        units.push((x, unit_size, cur_lr, edges));
                    }
                    edges |= LR_HAVE_LEFT;
                    x = next_x;
                    cur_lr = nlr;
                }
                if cur_lr.typ != RESTORATION_NONE {
                    edges &= !LR_HAVE_RIGHT;
                    units.push((x, w - x, cur_lr, edges));
                }
                for (ux, uw, lr, e) in units {
                    let mut sy = y;
                    let mut stripe_h = ((64 - 8 * (sy == 0) as i32) >> ss_ver).min(row_h - sy);
                    let mut e = e;
                    while sy + stripe_h <= row_h {
                        let have_bottom = sby + 1 != self.sbh || sy + stripe_h != row_h;
                        if have_bottom {
                            e |= LR_HAVE_BOTTOM;
                        } else {
                            e &= !LR_HAVE_BOTTOM;
                        }
                        let src = build_stripe(&cdef_out, &lpf.planes[plane], stride, lpf.pstride(plane), ux, sy, uw, stripe_h, w, h, e);
                        filter_unit(&mut pic.planes[plane], stride, ux as usize, sy as usize, uw as usize, stripe_h as usize, &src, &lr, self.bitdepth_max);
                        sy += stripe_h;
                        e |= LR_HAVE_TOP;
                        stripe_h = (64 >> ss_ver).min(row_h - sy);
                        if stripe_h == 0 {
                            break;
                        }
                    }
                }
            }
        }
        if superres {
            self.sr_cur = Some(pic);
        } else {
            self.cur = pic;
        }
    }

    /// Run the in-loop filters over the decoded frame.
    pub fn apply_filters(&mut self) {
        let hdr = self.hdr.clone();
        let seq = self.seq.clone();
        if hdr.loopfilter.level_y[0] != 0 || hdr.loopfilter.level_y[1] != 0 {
            for sby in 0..self.sbh {
                self.deblock_sbrow_cols(sby);
                self.deblock_sbrow_rows(sby);
            }
        }
        let deblocked = if seq.cdef || self.lf.restore_planes != 0 { Some(self.cur.clone()) } else { None };
        if seq.cdef {
            self.cdef_frame(deblocked.as_ref().unwrap());
        }
        if self.sr_cur.is_some() {
            self.superres();
        }
        if self.lf.restore_planes != 0 {
            let lpf = self.lr_source_rows(deblocked.as_ref().unwrap());
            self.lr_frame(&lpf);
        }
    }

}