pixon 0.8.0

SIMD-dispatched color conversion for the FFmpeg AVPixelFormat space: one-call decode of validated source frames to RGB / RGBA / Luma / HSV, fused resampling while converting, colorimetry-honest matrices and chroma siting, no_std-friendly.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
//! RFC #238 S1 — chroma-siting-aware 4:2:2 **resample** for `Yuv422p`.
//!
//! 4:2:2 subsamples chroma 2:1 horizontally only. Routing the centered siting
//! (`Center` / `Top` / `Bottom`, [`chroma_422_center_sited_h`]) through the
//! resample makes a downscale keep the correct chroma phase:
//!  - the **native fast tier** folds the #302 `1/4`–`3/4` triangle into the
//!    chroma area weights ([`ResamplePlan::area_chroma_422`]) — one SINGLE-
//!    rounding phased box-average on the subsampled grid;
//!  - the **encoded row-stage tier** (`with_native(false)`) reconstructs
//!    full-width chroma then bins in RGB.
//!
//! The co-sited / unspecified group stays phase 0, byte-identical to the
//! pre-siting resample (the folded form at phase 0 = the plain box overlaps).
//!
//! ★ Oracle (native tier): the EXACT code-domain box-average of the UNROUNDED
//! triangle-reconstructed chroma — a SINGLE rounding. This is the
//! user-approved, more-correct form; it differs from the #302
//! reconstruct-to-u8-then-bin (TWO roundings) by ≤ 1 LSB. The native path is
//! pinned against THIS (a YUV-domain oracle), never the RGB-domain oracle
//! (which would prove the wrong averaging domain). The encoded row-stage tier
//! IS the RGB-domain reconstruct-then-bin and is pinned against that.

use crate::{
  ChromaLocation, ColorInfo, ColorSpec, DynamicRange, KernelMatrix, PixelFormat, PixelSink,
  Primaries, Transfer,
  resample::{AreaResampler, AveragingDomain, LinearMode},
  sinker::MixedSinker,
  source::{Yuv422p, Yuv422pRow, Yuv444p, yuv422p_to, yuv444p_to},
};
use mediaframe::frame::{Yuv422pFrame, Yuv444pFrame};

const M: KernelMatrix = KernelMatrix::Bt601;
const FR: bool = true;

/// Round `a / d` half-up (ties toward `+∞`) — the production
/// `round_div_half_up`, replicated here so the oracle is independent.
fn rdhu(a: u64, d: u64) -> u64 {
  let q = a / d;
  let r = a % d;
  q + u64::from(r >= d - d / 2)
}

/// Exact box-overlap area weights for `src -> out`, mirroring
/// `resample::AxisSpans::area`: output cell `o` covers `[o·src, (o+1)·src)` in
/// `(src·out)` units, source cell `i` covers `[i·out, (i+1)·out)`; the weight
/// is their overlap. Returns per output `(first source cell, overlaps)`.
fn area_weights(src: usize, out: usize) -> Vec<(usize, Vec<u64>)> {
  let (src64, out64) = (src as u64, out as u64);
  (0..out)
    .map(|o| {
      let lo = o as u64 * src64;
      let hi = lo + src64;
      let start = (lo / out64) as usize;
      let mut w = Vec::new();
      let mut i = start as u64;
      loop {
        let clo = i * out64;
        if clo >= hi {
          break;
        }
        let chi = clo + out64;
        let ov = chi.min(hi) - clo.max(lo);
        if ov == 0 {
          break;
        }
        w.push(ov);
        if chi >= hi {
          break;
        }
        i += 1;
      }
      (start, w)
    })
    .collect()
}

/// Co-sited box-average of a full-resolution `sw x sh` u8 plane to
/// `ow x oh` (round-half-up) — the reference for a phase-free plane (luma, and
/// the co-sited chroma).
fn bin_cosited(plane: &[u8], sw: usize, sh: usize, ow: usize, oh: usize) -> Vec<u8> {
  let hw = area_weights(sw, ow);
  let vw = area_weights(sh, oh);
  let denom = (sw * sh) as u64;
  let mut out = vec![0u8; ow * oh];
  for (oy, (vs, vwin)) in vw.iter().enumerate() {
    for (ox, (hs, hwin)) in hw.iter().enumerate() {
      let mut s = 0u64;
      for (dy, &vwt) in vwin.iter().enumerate() {
        let mut hsum = 0u64;
        for (dx, &hwt) in hwin.iter().enumerate() {
          hsum += hwt * u64::from(plane[(vs + dy) * sw + hs + dx]);
        }
        s += vwt * hsum;
      }
      out[oy * ow + ox] = rdhu(s, denom) as u8;
    }
  }
  out
}

/// The EXACT centered chroma oracle: reconstruct the half-width `cw x ch`
/// chroma to full width with the #302 `1/4`–`3/4` triangle kept UNROUNDED
/// (scaled ×4 to stay integral: `r ∈ {1, 3, 4}`), then box-average to
/// `ow x oh` with a SINGLE round-half-up over `4·(2·cw)·ch`. This is the
/// code-domain twin the folded [`ResamplePlan::area_chroma_422`] weights
/// realize.
fn bin_chroma_centered(c: &[u8], cw: usize, ch: usize, ow: usize, oh: usize) -> Vec<u8> {
  let full = 2 * cw;
  // ×4 reconstruction plane (`full x ch`), independent of the production kernel.
  let mut r4 = vec![0u32; full * ch];
  for r in 0..ch {
    let row = &c[r * cw..r * cw + cw];
    for j in 0..cw {
      let l = u32::from(row[j.saturating_sub(1)]);
      let m = u32::from(row[j]);
      let rt = u32::from(row[if j + 1 < cw { j + 1 } else { j }]);
      r4[r * full + 2 * j] = l + 3 * m; // even col: (c[j-1] + 3·c[j])
      r4[r * full + 2 * j + 1] = 3 * m + rt; // odd col: (3·c[j] + c[j+1])
    }
  }
  let hw = area_weights(full, ow);
  let vw = area_weights(ch, oh);
  let denom = (4 * full * ch) as u64; // ×4 triangle × the box normalization
  let mut out = vec![0u8; ow * oh];
  for (oy, (vs, vwin)) in vw.iter().enumerate() {
    for (ox, (hs, hwin)) in hw.iter().enumerate() {
      let mut s = 0u64;
      for (dy, &vwt) in vwin.iter().enumerate() {
        let mut hsum = 0u64;
        for (dx, &hwt) in hwin.iter().enumerate() {
          hsum += hwt * u64::from(r4[(vs + dy) * full + hs + dx]);
        }
        s += vwt * hsum;
      }
      out[oy * ow + ox] = rdhu(s, denom) as u8;
    }
  }
  out
}

/// Independent #302 centered horizontal upsample (`1/4`–`3/4`, edge clamp,
/// round-half-up to u8) — the RGB-domain oracle's reconstruction step.
fn recon_full_row(c: &[u8], cw: usize) -> Vec<u8> {
  let mut out = vec![0u8; 2 * cw];
  for j in 0..cw {
    let l = u32::from(c[j.saturating_sub(1)]);
    let m = u32::from(c[j]);
    let r = u32::from(c[if j + 1 < cw { j + 1 } else { j }]);
    out[2 * j] = ((l + 3 * m + 2) >> 2) as u8;
    out[2 * j + 1] = ((3 * m + r + 2) >> 2) as u8;
  }
  out
}

/// A `Yuv422p` fixture with a strong HORIZONTAL chroma ramp (so the centered
/// triangle, which pulls neighbours, genuinely differs from the co-sited
/// nearest decode) plus a per-row tilt (a vertical mistake would show).
fn ramp(sw: usize, sh: usize) -> (Vec<u8>, Vec<u8>, Vec<u8>) {
  let cw = sw / 2;
  let mut y = vec![0u8; sw * sh];
  let mut u = vec![0u8; cw * sh];
  let mut v = vec![0u8; cw * sh];
  for (i, p) in y.iter_mut().enumerate() {
    *p = 40 + ((i as u32 * 3) % 160) as u8;
  }
  for r in 0..sh {
    for c in 0..cw {
      u[r * cw + c] = (30 + c * 44 + r * 4).min(240) as u8;
      v[r * cw + c] = (230u32.saturating_sub((c * 44 + r * 4) as u32)).max(16) as u8;
    }
  }
  (y, u, v)
}

/// A flat-chroma fixture: the centered phase is a no-op on constant chroma
/// (the triangle of a constant is that constant), so centered must equal
/// co-sited. Luma still varies.
fn flat_chroma(sw: usize, sh: usize) -> (Vec<u8>, Vec<u8>, Vec<u8>) {
  let cw = sw / 2;
  let mut y = vec![0u8; sw * sh];
  for (i, p) in y.iter_mut().enumerate() {
    *p = 40 + ((i as u32 * 7) % 170) as u8;
  }
  (y, vec![110u8; cw * sh], vec![140u8; cw * sh])
}

type Outs = (
  Vec<u8>,
  Vec<u8>,
  (Vec<u8>, Vec<u8>, Vec<u8>),
  Vec<u8>,
  Vec<u16>,
);

/// Drive a `Yuv422p` area resample (`sw x sh -> ow x oh`) for the full output
/// set, at `loc` siting and `native` tier.
#[allow(clippy::too_many_arguments)]
fn run(
  y: &[u8],
  u: &[u8],
  v: &[u8],
  sw: usize,
  sh: usize,
  ow: usize,
  oh: usize,
  loc: ChromaLocation,
  native: bool,
  simd: bool,
) -> Outs {
  let cw = sw / 2;
  let mut rgb = vec![0u8; ow * oh * 3];
  let mut rgba = vec![0u8; ow * oh * 4];
  let (mut hh, mut ss, mut vv) = (vec![0u8; ow * oh], vec![0u8; ow * oh], vec![0u8; ow * oh]);
  let mut luma = vec![0u8; ow * oh];
  let mut luma_u16 = vec![0u16; ow * oh];
  {
    let mut sink =
      MixedSinker::<Yuv422p, AreaResampler>::with_resampler(sw, sh, AreaResampler::to(ow, oh))
        .unwrap()
        .with_native(native)
        .with_chroma_location(loc.clone())
        .with_simd(simd)
        .with_rgb(&mut rgb)
        .unwrap()
        .with_rgba(&mut rgba)
        .unwrap()
        .with_hsv(&mut hh, &mut ss, &mut vv)
        .unwrap()
        .with_luma(&mut luma)
        .unwrap()
        .with_luma_u16(&mut luma_u16)
        .unwrap();
    let f = Yuv422pFrame::new(
      y, u, v, sw as u32, sh as u32, sw as u32, cw as u32, cw as u32,
    );
    yuv422p_to(&f, FR, sink.set_kernel_matrix(M)).unwrap();
  }
  (rgb, rgba, (hh, ss, vv), luma, luma_u16)
}

/// The centered NATIVE oracle: bin Y co-sited and U / V through the exact
/// centered chroma oracle to `ow x oh`, then convert ONCE at output width via
/// an identity `Yuv444p` sink — the exact ground truth the native tier
/// reproduces byte-for-byte.
#[allow(clippy::too_many_arguments)]
fn native_oracle(
  y: &[u8],
  u: &[u8],
  v: &[u8],
  sw: usize,
  sh: usize,
  ow: usize,
  oh: usize,
  simd: bool,
) -> Outs {
  let cw = sw / 2;
  let yb = bin_cosited(y, sw, sh, ow, oh);
  let ub = bin_chroma_centered(u, cw, sh, ow, oh);
  let vb = bin_chroma_centered(v, cw, sh, ow, oh);
  let mut rgb = vec![0u8; ow * oh * 3];
  let mut rgba = vec![0u8; ow * oh * 4];
  let (mut hh, mut ss, mut vv) = (vec![0u8; ow * oh], vec![0u8; ow * oh], vec![0u8; ow * oh]);
  let mut luma = vec![0u8; ow * oh];
  let mut luma_u16 = vec![0u16; ow * oh];
  {
    let mut sink = MixedSinker::<Yuv444p>::new(ow, oh)
      .with_simd(simd)
      .with_rgb(&mut rgb)
      .unwrap()
      .with_rgba(&mut rgba)
      .unwrap()
      .with_hsv(&mut hh, &mut ss, &mut vv)
      .unwrap()
      .with_luma(&mut luma)
      .unwrap()
      .with_luma_u16(&mut luma_u16)
      .unwrap();
    let f = Yuv444pFrame::new(
      &yb, &ub, &vb, ow as u32, oh as u32, ow as u32, ow as u32, ow as u32,
    );
    yuv444p_to(&f, FR, sink.set_kernel_matrix(M)).unwrap();
  }
  (rgb, rgba, (hh, ss, vv), luma, luma_u16)
}

/// The centered ENCODED row-stage oracle: reconstruct U / V to full width with
/// the #302 kernel (u8), then run that `Yuv444p` frame through a `with_native
/// (false)` RGB-domain resample — i.e. convert-each-row-then-bin-RGB, exactly
/// what the `Yuv422p` encoded arm does.
#[allow(clippy::too_many_arguments)]
fn encoded_oracle_rgb(
  y: &[u8],
  u: &[u8],
  v: &[u8],
  sw: usize,
  sh: usize,
  ow: usize,
  oh: usize,
  simd: bool,
) -> Vec<u8> {
  let cw = sw / 2;
  let mut uf = vec![0u8; sw * sh];
  let mut vf = vec![0u8; sw * sh];
  for r in 0..sh {
    uf[r * sw..r * sw + sw].copy_from_slice(&recon_full_row(&u[r * cw..r * cw + cw], cw));
    vf[r * sw..r * sw + sw].copy_from_slice(&recon_full_row(&v[r * cw..r * cw + cw], cw));
  }
  let mut rgb = vec![0u8; ow * oh * 3];
  {
    let mut sink =
      MixedSinker::<Yuv444p, AreaResampler>::with_resampler(sw, sh, AreaResampler::to(ow, oh))
        .unwrap()
        .with_native(false)
        .with_simd(simd)
        .with_rgb(&mut rgb)
        .unwrap();
    let f = Yuv444pFrame::new(
      y, &uf, &vf, sw as u32, sh as u32, sw as u32, sw as u32, sw as u32,
    );
    yuv444p_to(&f, FR, sink.set_kernel_matrix(M)).unwrap();
  }
  rgb
}

// ---- co-sited byte-identity (the regression contract) ----------------------

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn cosited_group_is_byte_identical_across_tiers() {
  // Every co-sited / unspecified siting must produce the byte-identical
  // pre-siting resample (phase 0 → the folded plan is never built), on BOTH
  // tiers. `Unspecified` is the baseline.
  let (y, u, v) = ramp(8, 8);
  for native in [true, false] {
    let base = run(
      &y,
      &u,
      &v,
      8,
      8,
      4,
      4,
      ChromaLocation::Unspecified,
      native,
      true,
    );
    for loc in [
      ChromaLocation::Left,
      ChromaLocation::TopLeft,
      ChromaLocation::BottomLeft,
      ChromaLocation::other("unassigned-7"),
    ] {
      let got = run(&y, &u, &v, 8, 8, 4, 4, loc.clone(), native, true);
      assert_eq!(got.0, base.0, "rgb {loc:?} native={native}");
      assert_eq!(got.1, base.1, "rgba {loc:?} native={native}");
      assert_eq!(got.2, base.2, "hsv {loc:?} native={native}");
      assert_eq!(got.3, base.3, "luma {loc:?} native={native}");
      assert_eq!(got.4, base.4, "luma_u16 {loc:?} native={native}");
    }
  }
}

// ---- centered native == the exact code-domain oracle -----------------------

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn centered_native_equals_code_domain_oracle() {
  // Clean 2:1 and a fractional ratio (both axes), for the whole centered group.
  for (sw, sh, ow, oh) in [(8, 8, 4, 4), (8, 8, 5, 3), (12, 6, 4, 4)] {
    let (y, u, v) = ramp(sw, sh);
    let o = native_oracle(&y, &u, &v, sw, sh, ow, oh, true);
    for loc in [
      ChromaLocation::Center,
      ChromaLocation::Top,
      ChromaLocation::Bottom,
    ] {
      let n = run(&y, &u, &v, sw, sh, ow, oh, loc.clone(), true, true);
      assert_eq!(n.0, o.0, "rgb {loc:?} {sw}x{sh}->{ow}x{oh}");
      assert_eq!(n.1, o.1, "rgba {loc:?} {sw}x{sh}->{ow}x{oh}");
      assert_eq!(n.2, o.2, "hsv {loc:?} {sw}x{sh}->{ow}x{oh}");
      assert_eq!(n.3, o.3, "luma {loc:?} {sw}x{sh}->{ow}x{oh}");
      assert_eq!(n.4, o.4, "luma_u16 {loc:?} {sw}x{sh}->{ow}x{oh}");
    }
  }
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn centered_native_simd_matches_scalar() {
  // Weights are precomputed integers, so the SIMD H/V passes must be 0-ULP.
  let (y, u, v) = ramp(8, 8);
  let s = run(&y, &u, &v, 8, 8, 4, 4, ChromaLocation::Center, true, false);
  let d = run(&y, &u, &v, 8, 8, 4, 4, ChromaLocation::Center, true, true);
  assert_eq!(s.0, d.0, "rgb scalar vs simd");
  assert_eq!(s.2, d.2, "hsv scalar vs simd");
  assert_eq!(s.3, d.3, "luma scalar vs simd");
}

// ---- centered encoded row-stage == RGB-domain reconstruct-then-bin ---------

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn centered_encoded_output_equals_rgb_reconstruct_then_bin() {
  for (sw, sh, ow, oh) in [(8, 8, 4, 4), (8, 8, 5, 3)] {
    let (y, u, v) = ramp(sw, sh);
    let oracle = encoded_oracle_rgb(&y, &u, &v, sw, sh, ow, oh, true);
    for loc in [
      ChromaLocation::Center,
      ChromaLocation::Top,
      ChromaLocation::Bottom,
    ] {
      let got = run(&y, &u, &v, sw, sh, ow, oh, loc.clone(), false, true);
      assert_eq!(got.0, oracle, "rgb {loc:?} {sw}x{sh}->{ow}x{oh}");
    }
  }
}

// ---- non-vacuous + flat-chroma sanity --------------------------------------

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn centered_differs_from_cosited_on_a_chroma_ramp() {
  // The phase must actually DO something: on a horizontal chroma ramp the
  // centered decode diverges from co-sited on both tiers.
  let (y, u, v) = ramp(8, 8);
  for native in [true, false] {
    let cos = run(
      &y,
      &u,
      &v,
      8,
      8,
      4,
      4,
      ChromaLocation::Unspecified,
      native,
      true,
    );
    let cen = run(&y, &u, &v, 8, 8, 4, 4, ChromaLocation::Center, native, true);
    assert_ne!(
      cen.0, cos.0,
      "centered rgb must differ from co-sited (native={native})"
    );
  }
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn centered_equals_cosited_on_flat_chroma() {
  // Sanity: on constant chroma the centered triangle is a no-op, so centered
  // and co-sited agree byte-for-byte (the phase machinery corrupts nothing).
  let (y, u, v) = flat_chroma(8, 8);
  for native in [true, false] {
    let cos = run(&y, &u, &v, 8, 8, 4, 4, ChromaLocation::Left, native, true);
    let cen = run(&y, &u, &v, 8, 8, 4, 4, ChromaLocation::Center, native, true);
    assert_eq!(cen.0, cos.0, "flat-chroma rgb (native={native})");
    assert_eq!(cen.2, cos.2, "flat-chroma hsv (native={native})");
  }
}

// ---- the ≤1 LSB single-rounding note, pinned -------------------------------

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn centered_native_is_within_1_lsb_of_reconstruct_then_bin() {
  // The folded single-rounding native output and the #302 reconstruct-to-u8-
  // then-bin (TWO roundings) agree to ≤ 1 LSB per chroma sample — the
  // user-approved more-correct form. Compared in the chroma CODE domain so the
  // convert/clamp cannot mask or amplify the gap. `[0, 2, 0, 2]` chroma is a
  // crafted case that provably exercises the divergence: the odd-column
  // reconstruction lands on an exact `.5` (`(3·0 + 2)/4`), which the folded
  // single rounding averages down to 0 while the intermediate `>>2` rounds it
  // up to 1 first (so reconstruct-then-bin yields 1) — a genuine 1-LSB gap.
  let (cw, sh, ow, oh) = (4usize, 2usize, 4usize, 2usize);
  let u: Vec<u8> = (0..cw * sh)
    .map(|i| if i.is_multiple_of(2) { 0 } else { 2 })
    .collect();
  let folded = bin_chroma_centered(&u, cw, sh, ow, oh);
  // reconstruct-then-bin: #302 to u8 per row, then co-sited box-average.
  let mut recon = vec![0u8; 2 * cw * sh];
  for r in 0..sh {
    recon[r * 2 * cw..r * 2 * cw + 2 * cw]
      .copy_from_slice(&recon_full_row(&u[r * cw..r * cw + cw], cw));
  }
  let double = bin_cosited(&recon, 2 * cw, sh, ow, oh);
  let maxd = folded
    .iter()
    .zip(&double)
    .map(|(&a, &b)| a.abs_diff(b))
    .max()
    .unwrap();
  assert!(
    maxd <= 1,
    "folded vs reconstruct-then-bin max delta {maxd} must be ≤ 1 LSB"
  );
  // Non-vacuous: the two rounding orders DO diverge here (proving the folded
  // single-rounding form is a genuine, more-correct distinction from #302).
  assert_ne!(folded, double, "the ≤1 LSB gap must be exercised");
}

// ---- cross-frame sink reuse rebuilds the phased join (RFC #238) -------------
//
// The native / HSV-only joins cache a chroma plan built for ONE frame's siting
// and are only `reset` between frames; a reused sink whose `chroma_location`
// changed to a different phase must REBUILD the join, else frame 2 inherits
// frame 1's (folded centered ⇄ unscaled co-sited) weights.

/// Reuse ONE full-output native-tier sink across two frames of the SAME
/// content, siting `loc1` then `loc2` (via `set_chroma_location` between
/// walks), returning frame 2's outputs.
#[allow(clippy::too_many_arguments)]
fn run_reuse_native(
  y: &[u8],
  u: &[u8],
  v: &[u8],
  sw: usize,
  sh: usize,
  ow: usize,
  oh: usize,
  loc1: ChromaLocation,
  loc2: ChromaLocation,
  simd: bool,
) -> Outs {
  let cw = sw / 2;
  let mut rgb = vec![0u8; ow * oh * 3];
  let mut rgba = vec![0u8; ow * oh * 4];
  let (mut hh, mut ss, mut vv) = (vec![0u8; ow * oh], vec![0u8; ow * oh], vec![0u8; ow * oh]);
  let mut luma = vec![0u8; ow * oh];
  let mut luma_u16 = vec![0u16; ow * oh];
  {
    let mut sink =
      MixedSinker::<Yuv422p, AreaResampler>::with_resampler(sw, sh, AreaResampler::to(ow, oh))
        .unwrap()
        .with_native(true)
        .with_simd(simd)
        .with_rgb(&mut rgb)
        .unwrap()
        .with_rgba(&mut rgba)
        .unwrap()
        .with_hsv(&mut hh, &mut ss, &mut vv)
        .unwrap()
        .with_luma(&mut luma)
        .unwrap()
        .with_luma_u16(&mut luma_u16)
        .unwrap();
    let f = Yuv422pFrame::new(
      y, u, v, sw as u32, sh as u32, sw as u32, cw as u32, cw as u32,
    );
    sink.set_chroma_location(loc1.clone());
    yuv422p_to(&f, FR, sink.set_kernel_matrix(M)).unwrap();
    sink.set_chroma_location(loc2.clone());
    yuv422p_to(&f, FR, sink.set_kernel_matrix(M)).unwrap();
  }
  (rgb, rgba, (hh, ss, vv), luma, luma_u16)
}

/// One HSV-only (`with_native(false)` → the `HsvDirectPlanarYuv` join) frame.
#[allow(clippy::too_many_arguments)]
fn run_hsv_only(
  y: &[u8],
  u: &[u8],
  v: &[u8],
  sw: usize,
  sh: usize,
  ow: usize,
  oh: usize,
  loc: ChromaLocation,
  simd: bool,
) -> (Vec<u8>, Vec<u8>, Vec<u8>) {
  let cw = sw / 2;
  let (mut hh, mut ss, mut vv) = (vec![0u8; ow * oh], vec![0u8; ow * oh], vec![0u8; ow * oh]);
  {
    let mut sink =
      MixedSinker::<Yuv422p, AreaResampler>::with_resampler(sw, sh, AreaResampler::to(ow, oh))
        .unwrap()
        .with_native(false)
        .with_chroma_location(loc.clone())
        .with_simd(simd)
        .with_hsv(&mut hh, &mut ss, &mut vv)
        .unwrap();
    let f = Yuv422pFrame::new(
      y, u, v, sw as u32, sh as u32, sw as u32, cw as u32, cw as u32,
    );
    yuv422p_to(&f, FR, sink.set_kernel_matrix(M)).unwrap();
  }
  (hh, ss, vv)
}

/// Reuse ONE HSV-only sink across two frames, siting `loc1` then `loc2`.
#[allow(clippy::too_many_arguments)]
fn run_reuse_hsv_only(
  y: &[u8],
  u: &[u8],
  v: &[u8],
  sw: usize,
  sh: usize,
  ow: usize,
  oh: usize,
  loc1: ChromaLocation,
  loc2: ChromaLocation,
  simd: bool,
) -> (Vec<u8>, Vec<u8>, Vec<u8>) {
  let cw = sw / 2;
  let (mut hh, mut ss, mut vv) = (vec![0u8; ow * oh], vec![0u8; ow * oh], vec![0u8; ow * oh]);
  {
    let mut sink =
      MixedSinker::<Yuv422p, AreaResampler>::with_resampler(sw, sh, AreaResampler::to(ow, oh))
        .unwrap()
        .with_native(false)
        .with_simd(simd)
        .with_hsv(&mut hh, &mut ss, &mut vv)
        .unwrap();
    let f = Yuv422pFrame::new(
      y, u, v, sw as u32, sh as u32, sw as u32, cw as u32, cw as u32,
    );
    sink.set_chroma_location(loc1.clone());
    yuv422p_to(&f, FR, sink.set_kernel_matrix(M)).unwrap();
    sink.set_chroma_location(loc2.clone());
    yuv422p_to(&f, FR, sink.set_kernel_matrix(M)).unwrap();
  }
  (hh, ss, vv)
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn native_join_rebuilds_on_siting_change_across_frames() {
  // Reuse one native-tier sink flipping Left ⇄ Center (both directions): frame
  // 2 must match a FRESH sink for frame 2's siting — no stale-phase carryover.
  let (y, u, v) = ramp(8, 8);
  for (a, b) in [
    (ChromaLocation::Left, ChromaLocation::Center),
    (ChromaLocation::Center, ChromaLocation::Left),
  ] {
    let reused = run_reuse_native(&y, &u, &v, 8, 8, 4, 4, a.clone(), b.clone(), true);
    let fresh = run(&y, &u, &v, 8, 8, 4, 4, b.clone(), true, true);
    assert_eq!(
      reused.0, fresh.0,
      "native rgb {a:?}->{b:?} stale-phase carryover"
    );
    assert_eq!(reused.1, fresh.1, "native rgba {a:?}->{b:?}");
    assert_eq!(reused.2, fresh.2, "native hsv {a:?}->{b:?}");
    // Non-vacuous: the two sitings genuinely differ on this ramp, so matching
    // frame 2's siting (not frame 1's) is a real distinction.
    let stale = run(&y, &u, &v, 8, 8, 4, 4, a.clone(), true, true);
    assert_ne!(
      fresh.0, stale.0,
      "sitings {a:?} vs {b:?} must differ (non-vacuous)"
    );
  }
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn hsv_only_join_rebuilds_on_siting_change_across_frames() {
  // The `HsvDirectPlanarYuv` twin: reuse one HSV-only sink flipping Left ⇄
  // Center; frame 2 must match a fresh sink for its siting.
  let (y, u, v) = ramp(8, 8);
  for (a, b) in [
    (ChromaLocation::Left, ChromaLocation::Center),
    (ChromaLocation::Center, ChromaLocation::Left),
  ] {
    let reused = run_reuse_hsv_only(&y, &u, &v, 8, 8, 4, 4, a.clone(), b.clone(), true);
    let fresh = run_hsv_only(&y, &u, &v, 8, 8, 4, 4, b.clone(), true);
    assert_eq!(reused, fresh, "hsv-only {a:?}->{b:?} stale-phase carryover");
    let stale = run_hsv_only(&y, &u, &v, 8, 8, 4, 4, a.clone(), true);
    assert_ne!(
      fresh, stale,
      "sitings {a:?} vs {b:?} must differ (non-vacuous)"
    );
  }
}

// ---- siting changed AFTER begin_frame (point-of-use invalidation) -----------
//
// A begin_frame-only invalidation misses a `set_chroma_location` /
// `set_color_spec` between `begin_frame` and row 0; the point-of-use re-check
// in `process` catches it. These drive the sink MANUALLY so the setter lands
// after `begin_frame`.

/// Apply the new siting via one of the two setters (both funnel to
/// `self.chroma_location`, the field the point-of-use check reads).
fn apply_siting<R>(
  sink: &mut MixedSinker<'_, Yuv422p, R>,
  loc: ChromaLocation,
  via_color_spec: bool,
) {
  if via_color_spec {
    let spec = ColorSpec::from_info(
      PixelFormat::Yuv422p,
      ColorInfo::new(
        Primaries::Unspecified,
        Transfer::Unspecified,
        crate::ColorMatrix::from(M),
        DynamicRange::Limited,
        loc,
      ),
    );
    sink.set_color_spec(&spec).unwrap();
  } else {
    sink.set_chroma_location(loc.clone());
  }
}

/// Reuse a native-tier sink: frame 1 at `loc1` (walker), then MANUALLY drive
/// frame 2 — `begin_frame` while still `loc1`, THEN switch to `loc2`, THEN feed
/// rows — returning frame 2's outputs.
#[allow(clippy::too_many_arguments)]
fn run_reuse_native_setter_after(
  y: &[u8],
  u: &[u8],
  v: &[u8],
  sw: usize,
  sh: usize,
  ow: usize,
  oh: usize,
  loc1: ChromaLocation,
  loc2: ChromaLocation,
  via_color_spec: bool,
  simd: bool,
) -> Outs {
  let cw = sw / 2;
  let mut rgb = vec![0u8; ow * oh * 3];
  let mut rgba = vec![0u8; ow * oh * 4];
  let (mut hh, mut ss, mut vv) = (vec![0u8; ow * oh], vec![0u8; ow * oh], vec![0u8; ow * oh]);
  let mut luma = vec![0u8; ow * oh];
  let mut luma_u16 = vec![0u16; ow * oh];
  {
    let mut sink =
      MixedSinker::<Yuv422p, AreaResampler>::with_resampler(sw, sh, AreaResampler::to(ow, oh))
        .unwrap()
        .with_native(true)
        .with_simd(simd)
        .with_rgb(&mut rgb)
        .unwrap()
        .with_rgba(&mut rgba)
        .unwrap()
        .with_hsv(&mut hh, &mut ss, &mut vv)
        .unwrap()
        .with_luma(&mut luma)
        .unwrap()
        .with_luma_u16(&mut luma_u16)
        .unwrap();
    let f = Yuv422pFrame::new(
      y, u, v, sw as u32, sh as u32, sw as u32, cw as u32, cw as u32,
    );
    sink.set_chroma_location(loc1.clone());
    yuv422p_to(&f, FR, sink.set_kernel_matrix(M)).unwrap();
    PixelSink::begin_frame(&mut sink, sw as u32, sh as u32).unwrap();
    apply_siting(&mut sink, loc2, via_color_spec); // AFTER begin_frame, before row 0
    for r in 0..sh {
      let row = Yuv422pRow::for_tests(
        &y[r * sw..r * sw + sw],
        &u[r * cw..r * cw + cw],
        &v[r * cw..r * cw + cw],
        r,
        M,
        FR,
      );
      PixelSink::process(&mut sink, row).unwrap();
    }
  }
  (rgb, rgba, (hh, ss, vv), luma, luma_u16)
}

/// The HSV-only (`with_native(false)` → `HsvDirectPlanarYuv`) twin of
/// [`run_reuse_native_setter_after`].
#[allow(clippy::too_many_arguments)]
fn run_reuse_hsv_setter_after(
  y: &[u8],
  u: &[u8],
  v: &[u8],
  sw: usize,
  sh: usize,
  ow: usize,
  oh: usize,
  loc1: ChromaLocation,
  loc2: ChromaLocation,
  via_color_spec: bool,
  simd: bool,
) -> (Vec<u8>, Vec<u8>, Vec<u8>) {
  let cw = sw / 2;
  let (mut hh, mut ss, mut vv) = (vec![0u8; ow * oh], vec![0u8; ow * oh], vec![0u8; ow * oh]);
  {
    let mut sink =
      MixedSinker::<Yuv422p, AreaResampler>::with_resampler(sw, sh, AreaResampler::to(ow, oh))
        .unwrap()
        .with_native(false)
        .with_simd(simd)
        .with_hsv(&mut hh, &mut ss, &mut vv)
        .unwrap();
    let f = Yuv422pFrame::new(
      y, u, v, sw as u32, sh as u32, sw as u32, cw as u32, cw as u32,
    );
    sink.set_chroma_location(loc1.clone());
    yuv422p_to(&f, FR, sink.set_kernel_matrix(M)).unwrap();
    PixelSink::begin_frame(&mut sink, sw as u32, sh as u32).unwrap();
    apply_siting(&mut sink, loc2, via_color_spec);
    for r in 0..sh {
      let row = Yuv422pRow::for_tests(
        &y[r * sw..r * sw + sw],
        &u[r * cw..r * cw + cw],
        &v[r * cw..r * cw + cw],
        r,
        M,
        FR,
      );
      PixelSink::process(&mut sink, row).unwrap();
    }
  }
  (hh, ss, vv)
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn native_join_rebuilds_on_siting_change_after_begin_frame() {
  // set_chroma_location AND set_color_spec, both Left ⇄ Center, applied AFTER
  // begin_frame: frame 2 must still match a FRESH sink for the new siting.
  let (y, u, v) = ramp(8, 8);
  for via_color_spec in [false, true] {
    for (a, b) in [
      (ChromaLocation::Left, ChromaLocation::Center),
      (ChromaLocation::Center, ChromaLocation::Left),
    ] {
      let reused = run_reuse_native_setter_after(
        &y,
        &u,
        &v,
        8,
        8,
        4,
        4,
        a.clone(),
        b.clone(),
        via_color_spec,
        true,
      );
      let fresh = run(&y, &u, &v, 8, 8, 4, 4, b.clone(), true, true);
      assert_eq!(
        reused.0, fresh.0,
        "native rgb {a:?}->{b:?} color_spec={via_color_spec}: stale after begin_frame"
      );
      assert_eq!(
        reused.2, fresh.2,
        "native hsv {a:?}->{b:?} color_spec={via_color_spec}"
      );
      let stale = run(&y, &u, &v, 8, 8, 4, 4, a, true, true);
      assert_ne!(fresh.0, stale.0, "sitings must differ (non-vacuous)");
    }
  }
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn hsv_only_join_rebuilds_on_siting_change_after_begin_frame() {
  let (y, u, v) = ramp(8, 8);
  for via_color_spec in [false, true] {
    for (a, b) in [
      (ChromaLocation::Left, ChromaLocation::Center),
      (ChromaLocation::Center, ChromaLocation::Left),
    ] {
      let reused = run_reuse_hsv_setter_after(
        &y,
        &u,
        &v,
        8,
        8,
        4,
        4,
        a.clone(),
        b.clone(),
        via_color_spec,
        true,
      );
      let fresh = run_hsv_only(&y, &u, &v, 8, 8, 4, 4, b.clone(), true);
      assert_eq!(
        reused, fresh,
        "hsv-only {a:?}->{b:?} color_spec={via_color_spec}: stale after begin_frame"
      );
      let stale = run_hsv_only(&y, &u, &v, 8, 8, 4, 4, a, true);
      assert_ne!(fresh, stale, "sitings must differ (non-vacuous)");
    }
  }
}

// ---- atomicity: the centered reserve sits BEHIND the resample preflight -----

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn out_of_sequence_centered_first_row_is_rejected_before_the_chroma_reserve() {
  use super::super::MixedSinkerError;
  use crate::resample::ResampleError;
  // The centered chroma reservation must run AFTER the resample preflight, so an
  // out-of-sequence FIRST row is rejected BEFORE any allocation (#180) — a
  // primed allocator refusal is never reached (OutOfSequenceRow, not
  // AllocationFailed). `with_native(false)` forces the encoded convert path (the
  // native fast tier bins the folded chroma plan and never reserves).
  let (y, u, v) = ramp(8, 8);
  let mut rgb = vec![0u8; 4 * 4 * 3];
  let mut sink =
    MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_native(false)
      .with_chroma_location(ChromaLocation::Center)
      .with_rgb(&mut rgb)
      .unwrap();
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  super::super::arm_chroma_full_alloc_failure();
  // First process call is row 5 — the stream expects row 0.
  let bad = Yuv422pRow::for_tests(
    &y[5 * 8..6 * 8],
    &u[5 * 4..6 * 4],
    &v[5 * 4..6 * 4],
    5,
    M,
    FR,
  );
  let err = PixelSink::process(&mut sink, bad).unwrap_err();
  assert!(
    matches!(
      err,
      MixedSinkerError::Resample(ResampleError::OutOfSequenceRow(_))
    ),
    "out-of-sequence centered first row must be OutOfSequenceRow (reserve unreached), got {err:?}"
  );
  assert_eq!(
    sink.chroma_full.len(),
    0,
    "a rejected row must allocate no chroma scratch"
  );
  // Non-vacuous: the failpoint is still armed, so a VALID first row now REACHES
  // the reserve (proving the guard is ordering, not a disabled reserve) — the
  // primed refusal surfaces as a recoverable AllocationFailed.
  let good = Yuv422pRow::for_tests(&y[0..8], &u[0..4], &v[0..4], 0, M, FR);
  let err0 = PixelSink::process(&mut sink, good).unwrap_err();
  assert!(
    matches!(
      err0,
      MixedSinkerError::Resample(ResampleError::AllocationFailed(_))
    ),
    "a valid centered row reaches the reserve (failpoint fires), got {err0:?}"
  );
}

/// A centered COLOUR area sink whose first row DOES reach the chroma reserve,
/// firing (and thereby consuming) the thread-local chroma failpoint. Asserts it
/// was still armed — confirming a preceding path never reserved — and leaves it
/// disarmed for the next test.
fn assert_chroma_failpoint_armed_then_consume(y: &[u8], u: &[u8], v: &[u8]) {
  use super::super::MixedSinkerError;
  use crate::resample::ResampleError;
  let mut rgb = vec![0u8; 4 * 4 * 3];
  let mut sink =
    MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_native(false)
      .with_chroma_location(ChromaLocation::Center)
      .with_rgb(&mut rgb)
      .unwrap();
  let f = Yuv422pFrame::new(y, u, v, 8, 8, 8, 4, 4);
  let err = yuv422p_to(&f, FR, sink.set_kernel_matrix(M)).unwrap_err();
  assert!(
    matches!(
      err,
      MixedSinkerError::Resample(ResampleError::AllocationFailed(_))
    ),
    "the chroma failpoint must still be armed (the prior path never reserved), got {err:?}"
  );
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn centered_filter_bicublin_rejected_before_the_chroma_reserve() {
  use super::super::MixedSinkerError;
  use crate::resample::{Bicublin, ResampleError};
  // A BICUBLIN (multi-kernel) filter plan on Yuv422p must be rejected as
  // UnsupportedFilter BEFORE the centered chroma reserve. With the failpoint
  // armed, an unfixed reserve-first path would surface the wrong
  // AllocationFailed instead.
  let (y, u, v) = ramp(8, 8);
  let mut rgb = vec![0u8; 4 * 4 * 3];
  {
    let mut sink = MixedSinker::<Yuv422p, Bicublin>::with_resampler(8, 8, Bicublin::to(4, 4))
      .unwrap()
      .with_chroma_location(ChromaLocation::Center)
      .with_rgb(&mut rgb)
      .unwrap();
    PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
    super::super::arm_chroma_full_alloc_failure();
    let row = Yuv422pRow::for_tests(&y[0..8], &u[0..4], &v[0..4], 0, M, FR);
    let err = PixelSink::process(&mut sink, row).unwrap_err();
    assert!(
      matches!(
        err,
        MixedSinkerError::Resample(ResampleError::UnsupportedFilter(_))
      ),
      "a BICUBLIN filter plan must be UnsupportedFilter (reserve unreached), got {err:?}"
    );
    assert_eq!(
      sink.chroma_full.len(),
      0,
      "a rejected filter plan allocates no chroma scratch"
    );
  }
  // The reject never reached the reserve, so the failpoint is still armed.
  assert_chroma_failpoint_armed_then_consume(&y, &u, &v);
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn luma_only_centered_area_does_not_reserve_chroma() {
  // A luma-only centered area resample (with_native(false), only luma attached)
  // never calls the RGB converter, so it must NOT reserve/reconstruct chroma:
  // with the failpoint armed it still succeeds (an unfixed path would reserve
  // and mask the luma output with a spurious chroma AllocationFailed).
  let (y, u, v) = ramp(8, 8);
  let mut luma = vec![0u8; 4 * 4];
  {
    let mut sink =
      MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
        .unwrap()
        .with_native(false)
        .with_chroma_location(ChromaLocation::Center)
        .with_luma(&mut luma)
        .unwrap();
    PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
    super::super::arm_chroma_full_alloc_failure();
    for r in 0..8 {
      let row = Yuv422pRow::for_tests(
        &y[r * 8..r * 8 + 8],
        &u[r * 4..r * 4 + 4],
        &v[r * 4..r * 4 + 4],
        r,
        M,
        FR,
      );
      PixelSink::process(&mut sink, row).unwrap();
    }
    assert_eq!(
      sink.chroma_full.len(),
      0,
      "luma-only centered resample must never reserve chroma scratch"
    );
  }
  // The luma-only path never reserved, so the failpoint is still armed.
  assert_chroma_failpoint_armed_then_consume(&y, &u, &v);
}

// ---- centered LINEAR folds the chroma phase (both decodes) ------------------

/// A `Yuv422p` linear-light area resample (`with_native(false)`) to RGB, at
/// `loc` siting and `mode` (display-referred clamped vs scene-referred f32).
fn run_linear_422(
  y: &[u8],
  u: &[u8],
  v: &[u8],
  loc: ChromaLocation,
  mode: LinearMode,
  simd: bool,
) -> Vec<u8> {
  let (sw, sh, ow, oh, cw) = (8usize, 8usize, 4usize, 4usize, 4usize);
  let mut rgb = vec![0u8; ow * oh * 3];
  {
    let mut sink =
      MixedSinker::<Yuv422p, AreaResampler>::with_resampler(sw, sh, AreaResampler::to(ow, oh))
        .unwrap()
        .with_averaging_domain(AveragingDomain::Linear)
        .with_native(false)
        .with_linear_mode(mode)
        .with_chroma_location(loc.clone())
        .with_simd(simd)
        .with_rgb(&mut rgb)
        .unwrap();
    let f = Yuv422pFrame::new(
      y, u, v, sw as u32, sh as u32, sw as u32, cw as u32, cw as u32,
    );
    yuv422p_to(&f, FR, sink.set_kernel_matrix(M)).unwrap();
  }
  rgb
}

/// The centered-Linear oracle: reconstruct U / V to full width with the #302
/// kernel (u8), then run that `Yuv444p` frame through the SAME linear-light
/// resample — i.e. reconstruct-then-linear-average, the exact operation the
/// centered `Yuv422p` Linear arm performs (decode over full-width chroma,
/// linearise, area-average, re-encode).
fn oracle_linear_reconstruct(
  y: &[u8],
  u: &[u8],
  v: &[u8],
  mode: LinearMode,
  simd: bool,
) -> Vec<u8> {
  let (sw, sh, ow, oh, cw) = (8usize, 8usize, 4usize, 4usize, 4usize);
  let mut uf = vec![0u8; sw * sh];
  let mut vf = vec![0u8; sw * sh];
  for r in 0..sh {
    uf[r * sw..r * sw + sw].copy_from_slice(&recon_full_row(&u[r * cw..r * cw + cw], cw));
    vf[r * sw..r * sw + sw].copy_from_slice(&recon_full_row(&v[r * cw..r * cw + cw], cw));
  }
  let mut rgb = vec![0u8; ow * oh * 3];
  {
    let mut sink =
      MixedSinker::<Yuv444p, AreaResampler>::with_resampler(sw, sh, AreaResampler::to(ow, oh))
        .unwrap()
        .with_averaging_domain(AveragingDomain::Linear)
        .with_native(false)
        .with_linear_mode(mode)
        .with_simd(simd)
        .with_rgb(&mut rgb)
        .unwrap();
    let f = Yuv444pFrame::new(
      y, &uf, &vf, sw as u32, sh as u32, sw as u32, sw as u32, sw as u32,
    );
    yuv444p_to(&f, FR, sink.set_kernel_matrix(M)).unwrap();
  }
  rgb
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn centered_linear_folds_the_phase_for_both_decodes() {
  // Centered Linear must reconstruct full-width chroma and decode 4:4:4 (not
  // silently co-site) for BOTH the display-referred (clamped u8) and
  // scene-referred (f32 unclamped) decodes: it equals reconstruct-then-linear,
  // and differs from co-sited on a chroma ramp.
  let (y, u, v) = ramp(8, 8);
  for mode in [LinearMode::DisplayReferred, LinearMode::SceneReferred] {
    let centered = run_linear_422(&y, &u, &v, ChromaLocation::Center, mode, true);
    let oracle = oracle_linear_reconstruct(&y, &u, &v, mode, true);
    assert_eq!(
      centered, oracle,
      "centered Linear ({mode:?}) must equal reconstruct-then-linear-average"
    );
    let cosited = run_linear_422(&y, &u, &v, ChromaLocation::Left, mode, true);
    assert_ne!(
      centered, cosited,
      "centered Linear ({mode:?}) must differ from co-sited (non-vacuous)"
    );
  }
}

// ---- mid-frame phase change rejects WITHOUT dropping the cached stream ------

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn native_mid_frame_phase_change_rejection_keeps_the_stream_retryable() {
  use super::super::MixedSinkerError;
  // Advance rows 0,1 (Center), then flip siting mid-frame (Left). FIX 1 (R6):
  // the frozen-phase CHECK rejects it with ChromaSitingChanged at the choke
  // point, ahead of the out-of-sequence check — a mixed-phase frame is never
  // emitted; the frame must be restarted (a rejected row mutates no state).
  let (y, u, v) = ramp(8, 8);
  let cw = 4usize;
  let mut rgb = vec![0u8; 4 * 4 * 3];
  let mut sink =
    MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_native(true)
      .with_chroma_location(ChromaLocation::Center)
      .with_rgb(&mut rgb)
      .unwrap();
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  for r in 0..2 {
    let row = Yuv422pRow::for_tests(
      &y[r * 8..r * 8 + 8],
      &u[r * cw..r * cw + cw],
      &v[r * cw..r * cw + cw],
      r,
      M,
      FR,
    );
    PixelSink::process(&mut sink, row).unwrap();
  }
  sink.set_chroma_location(ChromaLocation::Left);
  let bad = Yuv422pRow::for_tests(
    &y[5 * 8..6 * 8],
    &u[5 * cw..6 * cw],
    &v[5 * cw..6 * cw],
    5,
    M,
    FR,
  );
  let err = PixelSink::process(&mut sink, bad).unwrap_err();
  assert!(
    matches!(err, MixedSinkerError::ChromaSitingChanged(_)),
    "mid-frame siting change must be ChromaSitingChanged, got {err:?}"
  );
  // The rejected row mutated no stream state: begin_frame restarts cleanly and
  // a fresh frame at the new siting processes without error.
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  for r in 0..8 {
    let row = Yuv422pRow::for_tests(
      &y[r * 8..r * 8 + 8],
      &u[r * cw..r * cw + cw],
      &v[r * cw..r * cw + cw],
      r,
      M,
      FR,
    );
    PixelSink::process(&mut sink, row).unwrap();
  }
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn hsv_mid_frame_phase_change_rejection_keeps_the_stream_retryable() {
  use super::super::MixedSinkerError;
  // The HSV-only (`hsv_direct`) twin: a mid-frame siting flip is rejected with
  // ChromaSitingChanged (FIX 1), and begin_frame restarts cleanly.
  let (y, u, v) = ramp(8, 8);
  let cw = 4usize;
  let (mut hh, mut ss, mut vv) = (vec![0u8; 4 * 4], vec![0u8; 4 * 4], vec![0u8; 4 * 4]);
  let mut sink =
    MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_native(false)
      .with_chroma_location(ChromaLocation::Center)
      .with_hsv(&mut hh, &mut ss, &mut vv)
      .unwrap();
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  for r in 0..2 {
    let row = Yuv422pRow::for_tests(
      &y[r * 8..r * 8 + 8],
      &u[r * cw..r * cw + cw],
      &v[r * cw..r * cw + cw],
      r,
      M,
      FR,
    );
    PixelSink::process(&mut sink, row).unwrap();
  }
  sink.set_chroma_location(ChromaLocation::Left);
  let bad = Yuv422pRow::for_tests(
    &y[5 * 8..6 * 8],
    &u[5 * cw..6 * cw],
    &v[5 * cw..6 * cw],
    5,
    M,
    FR,
  );
  let err = PixelSink::process(&mut sink, bad).unwrap_err();
  assert!(
    matches!(err, MixedSinkerError::ChromaSitingChanged(_)),
    "mid-frame siting change (HSV) must be ChromaSitingChanged, got {err:?}"
  );
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  for r in 0..8 {
    let row = Yuv422pRow::for_tests(
      &y[r * 8..r * 8 + 8],
      &u[r * cw..r * cw + cw],
      &v[r * cw..r * cw + cw],
      r,
      M,
      FR,
    );
    PixelSink::process(&mut sink, row).unwrap();
  }
}

// ---- IN-SEQUENCE mid-frame phase change is rejected (not silently mixed) -----
//
// freezing the phase per-frame is not enough to DROP a stale plan — an
// in-sequence row after a mid-frame `set_chroma_location` passes the sequence
// preflight, so without the frozen-phase CHECK it would reconstruct the new
// phase and the frame would bin a mixture. The effective siting is frozen on
// the first output-bearing row; a later in-sequence row observing a different
// phase must be rejected with `ChromaSitingChanged`, uniformly across tiers.

/// Drive one Yuv422p resample frame: `begin_frame`, accept row 0 at `loc1`
/// (freezes the phase), flip to `loc2`, then feed the IN-SEQUENCE row 1 and
/// return its `process` result.
fn in_sequence_flip_row1<R>(
  mut sink: MixedSinker<'_, Yuv422p, R>,
  y: &[u8],
  u: &[u8],
  v: &[u8],
  loc1: ChromaLocation,
  loc2: ChromaLocation,
) -> Result<(), super::super::MixedSinkerError> {
  let cw = 4usize;
  sink.set_chroma_location(loc1.clone());
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  let row0 = Yuv422pRow::for_tests(&y[0..8], &u[0..cw], &v[0..cw], 0, M, FR);
  PixelSink::process(&mut sink, row0).unwrap();
  sink.set_chroma_location(loc2.clone());
  let row1 = Yuv422pRow::for_tests(&y[8..16], &u[cw..2 * cw], &v[cw..2 * cw], 1, M, FR);
  PixelSink::process(&mut sink, row1)
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn in_sequence_mid_frame_phase_change_rejected_across_tiers() {
  use super::super::MixedSinkerError;
  let (y, u, v) = ramp(8, 8);
  // Both flip directions: Center->Left (drop the phase) and Left->Center (add
  // it). Each must reject the in-sequence row 1 with ChromaSitingChanged.
  for (loc1, loc2) in [
    (ChromaLocation::Center, ChromaLocation::Left),
    (ChromaLocation::Left, ChromaLocation::Center),
  ] {
    // Native fast tier.
    let mut rgb = vec![0u8; 4 * 4 * 3];
    let sink = MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_native(true)
      .with_rgb(&mut rgb)
      .unwrap();
    let err = in_sequence_flip_row1(sink, &y, &u, &v, loc1.clone(), loc2.clone()).unwrap_err();
    assert!(
      matches!(err, MixedSinkerError::ChromaSitingChanged(_)),
      "native {loc1:?}->{loc2:?}: want ChromaSitingChanged, got {err:?}"
    );

    // Encoded row-stage RGB tier (`with_native(false)`).
    let mut rgb = vec![0u8; 4 * 4 * 3];
    let sink = MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_native(false)
      .with_rgb(&mut rgb)
      .unwrap();
    let err = in_sequence_flip_row1(sink, &y, &u, &v, loc1.clone(), loc2.clone()).unwrap_err();
    assert!(
      matches!(err, MixedSinkerError::ChromaSitingChanged(_)),
      "encoded-rgb {loc1:?}->{loc2:?}: want ChromaSitingChanged, got {err:?}"
    );

    // HSV-only tier (the `HsvDirectPlanarYuv` join).
    let (mut hh, mut ss, mut vv) = (vec![0u8; 4 * 4], vec![0u8; 4 * 4], vec![0u8; 4 * 4]);
    let sink = MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_native(false)
      .with_hsv(&mut hh, &mut ss, &mut vv)
      .unwrap();
    let err = in_sequence_flip_row1(sink, &y, &u, &v, loc1.clone(), loc2.clone()).unwrap_err();
    assert!(
      matches!(err, MixedSinkerError::ChromaSitingChanged(_)),
      "hsv {loc1:?}->{loc2:?}: want ChromaSitingChanged, got {err:?}"
    );

    // Linear averaging domain.
    let mut rgb = vec![0u8; 4 * 4 * 3];
    let sink = MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_averaging_domain(AveragingDomain::Linear)
      .with_native(false)
      .with_rgb(&mut rgb)
      .unwrap();
    let err = in_sequence_flip_row1(sink, &y, &u, &v, loc1.clone(), loc2.clone()).unwrap_err();
    assert!(
      matches!(err, MixedSinkerError::ChromaSitingChanged(_)),
      "linear {loc1:?}->{loc2:?}: want ChromaSitingChanged, got {err:?}"
    );

    // Filter tier (single-kernel Triangle FilteredResampler).
    let mut rgb = vec![0u8; 4 * 4 * 3];
    let sink =
      MixedSinker::<Yuv422p, crate::resample::FilteredResampler<crate::resample::Triangle>>::with_resampler(
        8,
        8,
        crate::resample::FilteredResampler::new(4, 4, crate::resample::Triangle),
      )
      .unwrap()
      .with_rgb(&mut rgb)
      .unwrap();
    let err = in_sequence_flip_row1(sink, &y, &u, &v, loc1.clone(), loc2.clone()).unwrap_err();
    assert!(
      matches!(err, MixedSinkerError::ChromaSitingChanged(_)),
      "filter {loc1:?}->{loc2:?}: want ChromaSitingChanged, got {err:?}"
    );
  }
}

// ---- centered Linear preflight runs BEFORE the chroma reserve ---------------

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn centered_linear_output_set_change_rejects_before_the_chroma_reserve() {
  use super::super::MixedSinkerError;
  // a luma-only centered Linear row 0 reserves no chroma (want_color
  // false). Attaching RGB mid-frame flips want_color true while chroma_full is
  // still empty, so the CENTERED Linear arm would reserve + reconstruct — but
  // the row is a mid-frame output-set change and must be rejected
  // (ResampleOutputsChanged) by the full Linear preflight BEFORE any allocation.
  // With the chroma failpoint armed, a reserve-first path would surface the
  // WRONG AllocationFailed instead (non-vacuous).
  let (y, u, v) = ramp(8, 8);
  let cw = 4usize;
  let mut luma = vec![0u8; 4 * 4];
  let mut rgb = vec![0u8; 4 * 4 * 3];
  {
    let mut sink =
      MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
        .unwrap()
        .with_averaging_domain(AveragingDomain::Linear)
        .with_native(false)
        .with_chroma_location(ChromaLocation::Center)
        .with_luma(&mut luma)
        .unwrap();
    PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
    let row0 = Yuv422pRow::for_tests(&y[0..8], &u[0..cw], &v[0..cw], 0, M, FR);
    PixelSink::process(&mut sink, row0).unwrap(); // luma-only: reserves no chroma
    super::super::arm_chroma_full_alloc_failure();
    sink.set_rgb(&mut rgb).unwrap(); // attach RGB mid-frame → want_color true
    let row1 = Yuv422pRow::for_tests(&y[8..16], &u[cw..2 * cw], &v[cw..2 * cw], 1, M, FR);
    let err = PixelSink::process(&mut sink, row1).unwrap_err();
    assert!(
      matches!(err, MixedSinkerError::ResampleOutputsChanged(_)),
      "mid-frame output-set change must be ResampleOutputsChanged (reserve unreached), got {err:?}"
    );
    assert_eq!(
      sink.chroma_full.len(),
      0,
      "a rejected centered-Linear row reserves no chroma scratch"
    );
  }
  // The reject never reached the reserve, so the failpoint is still armed.
  assert_chroma_failpoint_armed_then_consume(&y, &u, &v);
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn centered_rowstage_reserve_oom_leaves_outputs_unfrozen_for_retry() {
  use super::super::MixedSinkerError;
  use crate::resample::ResampleError;
  // the centered row-stage / filter arm gates its chroma reserve with a
  // CHECK-ONLY preflight (no output-set freeze). A recoverable reserve OOM on
  // the first output-bearing row must leave `resample_outputs` uncommitted, so a
  // retry of row 0 with a CHANGED output attachment is accepted — not falsely
  // rejected as ResampleOutputsChanged (a committing preflight would freeze the
  // old output set before the failing reserve).
  let (y, u, v) = ramp(8, 8);
  let cw = 4usize;
  let mut rgb = vec![0u8; 4 * 4 * 3];
  let mut luma = vec![0u8; 4 * 4];
  let mut sink =
    MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_native(false)
      .with_chroma_location(ChromaLocation::Center)
      .with_rgb(&mut rgb)
      .unwrap();
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  super::super::arm_chroma_full_alloc_failure();
  let row0 = Yuv422pRow::for_tests(&y[0..8], &u[0..cw], &v[0..cw], 0, M, FR);
  let err = PixelSink::process(&mut sink, row0).unwrap_err();
  assert!(
    matches!(
      err,
      MixedSinkerError::Resample(ResampleError::AllocationFailed(_))
    ),
    "the armed failpoint must surface AllocationFailed on the row-0 reserve, got {err:?}"
  );
  // Attach luma before retrying (changes the output set). The failpoint is
  // consumed, so the reserve now succeeds; because the OOM row never froze the
  // output set, row 0 is accepted with the new attachment.
  sink.set_luma(&mut luma).unwrap();
  let row0b = Yuv422pRow::for_tests(&y[0..8], &u[0..cw], &v[0..cw], 0, M, FR);
  PixelSink::process(&mut sink, row0b).unwrap();
}

// ---- a genuine LATER-row output-set change is ResampleOutputsChanged --------
//
// the check-only preflight's ordering — the
// output-set compare precedes the general sequence reject on a later row, so
// attaching an output for row 1 (which selects a fresh, next_y==0 stream) is
// surfaced as ResampleOutputsChanged, NOT OutOfSequenceRow.

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn centered_rowstage_later_output_attach_is_resample_outputs_changed() {
  use super::super::MixedSinkerError;
  let (y, u, v) = ramp(8, 8);
  let cw = 4usize;
  let mut rgb = vec![0u8; 4 * 4 * 3];
  let mut luma = vec![0u8; 4 * 4];
  let mut sink =
    MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_native(false)
      .with_chroma_location(ChromaLocation::Center)
      .with_rgb(&mut rgb)
      .unwrap();
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  let row0 = Yuv422pRow::for_tests(&y[0..8], &u[0..cw], &v[0..cw], 0, M, FR);
  PixelSink::process(&mut sink, row0).unwrap(); // RGB-only, freezes {rgb}
  sink.set_luma(&mut luma).unwrap(); // attach luma → output set changed
  let row1 = Yuv422pRow::for_tests(&y[8..16], &u[cw..2 * cw], &v[cw..2 * cw], 1, M, FR);
  let err = PixelSink::process(&mut sink, row1).unwrap_err();
  assert!(
    matches!(err, MixedSinkerError::ResampleOutputsChanged(_)),
    "later output-set change (row-stage) must be ResampleOutputsChanged, got {err:?}"
  );
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn centered_filter_later_output_attach_is_resample_outputs_changed() {
  use super::super::MixedSinkerError;
  let (y, u, v) = ramp(8, 8);
  let cw = 4usize;
  let mut rgb = vec![0u8; 4 * 4 * 3];
  let mut luma = vec![0u8; 4 * 4];
  let mut sink = MixedSinker::<
    Yuv422p,
    crate::resample::FilteredResampler<crate::resample::Triangle>,
  >::with_resampler(
    8,
    8,
    crate::resample::FilteredResampler::new(4, 4, crate::resample::Triangle),
  )
  .unwrap()
  .with_chroma_location(ChromaLocation::Center)
  .with_rgb(&mut rgb)
  .unwrap();
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  let row0 = Yuv422pRow::for_tests(&y[0..8], &u[0..cw], &v[0..cw], 0, M, FR);
  PixelSink::process(&mut sink, row0).unwrap(); // RGB-only, freezes {rgb}
  sink.set_luma(&mut luma).unwrap(); // attach luma → output set changed
  let row1 = Yuv422pRow::for_tests(&y[8..16], &u[cw..2 * cw], &v[cw..2 * cw], 1, M, FR);
  let err = PixelSink::process(&mut sink, row1).unwrap_err();
  assert!(
    matches!(err, MixedSinkerError::ResampleOutputsChanged(_)),
    "later output-set change (filter) must be ResampleOutputsChanged, got {err:?}"
  );
}

// ---- an out-of-sequence first row is state-atomic: it does NOT drop the join -
//
// the point-of-use phase invalidation must not clear a cached join before
// the row's sequence is validated. A reused sink, new frame, changed siting,
// then an out-of-sequence first process call must be rejected (OutOfSequenceRow)
// with the cached join INTACT, so a corrected row-0 retry rebuilds cleanly.

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn native_out_of_sequence_first_row_does_not_drop_the_cached_join() {
  use super::super::MixedSinkerError;
  use crate::resample::ResampleError;
  let (y, u, v) = ramp(8, 8);
  let cw = 4usize;
  let mut rgb = vec![0u8; 4 * 4 * 3];
  let mut sink =
    MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_native(true)
      .with_chroma_location(ChromaLocation::Left)
      .with_rgb(&mut rgb)
      .unwrap();
  // Frame 1 at Left builds the native join.
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  for r in 0..8 {
    let row = Yuv422pRow::for_tests(
      &y[r * 8..r * 8 + 8],
      &u[r * cw..r * cw + cw],
      &v[r * cw..r * cw + cw],
      r,
      M,
      FR,
    );
    PixelSink::process(&mut sink, row).unwrap();
  }
  assert!(
    sink.native_planar.is_some(),
    "frame 1 builds the native join"
  );
  // Frame 2: change siting to Center, then feed an OUT-OF-SEQUENCE first row.
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  sink.set_chroma_location(ChromaLocation::Center);
  let bad = Yuv422pRow::for_tests(
    &y[3 * 8..4 * 8],
    &u[3 * cw..4 * cw],
    &v[3 * cw..4 * cw],
    3,
    M,
    FR,
  );
  let err = PixelSink::process(&mut sink, bad).unwrap_err();
  assert!(
    matches!(
      err,
      MixedSinkerError::Resample(ResampleError::OutOfSequenceRow(_))
    ),
    "out-of-sequence first row must be OutOfSequenceRow, got {err:?}"
  );
  // the rejected row is state-atomic — the cached join is NOT dropped.
  assert!(
    sink.native_planar.is_some(),
    "an out-of-sequence first row must NOT drop the cached native join"
  );
  // The corrected retry (row 0, now rebuilding for Center) succeeds.
  for r in 0..8 {
    let row = Yuv422pRow::for_tests(
      &y[r * 8..r * 8 + 8],
      &u[r * cw..r * cw + cw],
      &v[r * cw..r * cw + cw],
      r,
      M,
      FR,
    );
    PixelSink::process(&mut sink, row).unwrap();
  }
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn hsv_out_of_sequence_first_row_does_not_drop_the_cached_join() {
  use super::super::MixedSinkerError;
  use crate::resample::ResampleError;
  let (y, u, v) = ramp(8, 8);
  let cw = 4usize;
  let (mut hh, mut ss, mut vv) = (vec![0u8; 4 * 4], vec![0u8; 4 * 4], vec![0u8; 4 * 4]);
  let mut sink =
    MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_native(false)
      .with_chroma_location(ChromaLocation::Left)
      .with_hsv(&mut hh, &mut ss, &mut vv)
      .unwrap();
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  for r in 0..8 {
    let row = Yuv422pRow::for_tests(
      &y[r * 8..r * 8 + 8],
      &u[r * cw..r * cw + cw],
      &v[r * cw..r * cw + cw],
      r,
      M,
      FR,
    );
    PixelSink::process(&mut sink, row).unwrap();
  }
  assert!(
    sink.hsv_planar.is_some(),
    "frame 1 builds the HSV-only join"
  );
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  sink.set_chroma_location(ChromaLocation::Center);
  let bad = Yuv422pRow::for_tests(
    &y[3 * 8..4 * 8],
    &u[3 * cw..4 * cw],
    &v[3 * cw..4 * cw],
    3,
    M,
    FR,
  );
  let err = PixelSink::process(&mut sink, bad).unwrap_err();
  assert!(
    matches!(
      err,
      MixedSinkerError::Resample(ResampleError::OutOfSequenceRow(_))
    ),
    "out-of-sequence first HSV row must be OutOfSequenceRow, got {err:?}"
  );
  assert!(
    sink.hsv_planar.is_some(),
    "an out-of-sequence first row must NOT drop the cached HSV join"
  );
  for r in 0..8 {
    let row = Yuv422pRow::for_tests(
      &y[r * 8..r * 8 + 8],
      &u[r * cw..r * cw + cw],
      &v[r * cw..r * cw + cw],
      r,
      M,
      FR,
    );
    PixelSink::process(&mut sink, row).unwrap();
  }
}

// ---- a row-0 phase rebuild that OOMs is state-atomic: the old join survives ---
//
// a valid in-sequence row-0 siting change rebuilds the cached join. If that
// rebuild's allocation is refused, the OLD (prior-phase) join must be restored
// (Some), so the rejected row mutates nothing and a corrected retry can proceed.
// (Gated to the features that compile the native-join chroma failpoint.)

#[test]
#[cfg(any(
  feature = "yuv-packed",
  feature = "yuv-444-packed",
  all(feature = "yuv-semi-planar", feature = "rgb")
))]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn native_row0_siting_change_rebuild_oom_restores_the_old_join() {
  use super::super::MixedSinkerError;
  use crate::resample::ResampleError;
  let (y, u, v) = ramp(8, 8);
  let cw = 4usize;
  let mut rgb = vec![0u8; 4 * 4 * 3];
  let mut luma = vec![0u8; 4 * 4];
  let mut sink =
    MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_native(true)
      .with_chroma_location(ChromaLocation::Left)
      .with_rgb(&mut rgb)
      .unwrap();
  // Frame 1 at Left builds the native join.
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  for r in 0..8 {
    let row = Yuv422pRow::for_tests(
      &y[r * 8..r * 8 + 8],
      &u[r * cw..r * cw + cw],
      &v[r * cw..r * cw + cw],
      r,
      M,
      FR,
    );
    PixelSink::process(&mut sink, row).unwrap();
  }
  assert!(sink.native_planar.is_some());
  // Frame 2: change siting to Center; the in-sequence row-0 rebuild hits an
  // allocation refusal building the new (Center) chroma plan.
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  sink.set_chroma_location(ChromaLocation::Center);
  crate::sinker::mixed::arm_planar_native_chroma_failure();
  let row0 = Yuv422pRow::for_tests(&y[0..8], &u[0..cw], &v[0..cw], 0, M, FR);
  let err = PixelSink::process(&mut sink, row0).unwrap_err();
  assert!(
    matches!(
      err,
      MixedSinkerError::Resample(ResampleError::AllocationFailed(_))
    ),
    "the row-0 rebuild OOM must surface AllocationFailed, got {err:?}"
  );
  // the rejected row is state-atomic — the intact prior-phase (Left) join
  // is restored (dropping the join up front then rebuilding would leave
  // native_planar None on a build failure).
  assert!(
    sink.native_planar.is_some(),
    "a row-0 rebuild OOM must restore the intact prior-phase join"
  );
  // the output-set freeze was also never committed (the delegate commits only
  // after its build succeeds) — a retry of row 0 with a CHANGED output attachment
  // (luma added) is accepted, not mis-rejected as ResampleOutputsChanged.
  sink.set_luma(&mut luma).unwrap();
  let row0b = Yuv422pRow::for_tests(&y[0..8], &u[0..cw], &v[0..cw], 0, M, FR);
  PixelSink::process(&mut sink, row0b).unwrap();
}

#[test]
#[cfg(feature = "rgb")]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn native_row0_rebuild_scratch_oom_is_state_atomic() {
  use super::super::MixedSinkerError;
  use crate::resample::ResampleError;
  // A stale-phase rebuild whose output-width RGB scratch grow OOMs must be
  // state-atomic. The delegate builds the replacement into a local and grows the
  // scratch BEFORE committing, so a scratch refusal leaves the field `None` (the
  // arm took the stale join) and `resample_outputs` uncommitted; the arm's
  // stale-gated restore brings the prior join back. Prime a native HSV-only join
  // (reserves no RGB scratch), switch to centered RGB, and refuse the (now-empty)
  // RGB-scratch grow on row 0.
  let (y, u, v) = ramp(8, 8);
  let cw = 4usize;
  let (mut hh, mut ss, mut vv) = (vec![0u8; 4 * 4], vec![0u8; 4 * 4], vec![0u8; 4 * 4]);
  let mut rgb = vec![0u8; 4 * 4 * 3];
  let mut luma = vec![0u8; 4 * 4];
  let mut sink =
    MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_native(true)
      .with_chroma_location(ChromaLocation::Left)
      .with_hsv(&mut hh, &mut ss, &mut vv)
      .unwrap();
  // Frame 1: native HSV-only join at Left (reserves no RGB scratch).
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  for r in 0..8 {
    let row = Yuv422pRow::for_tests(
      &y[r * 8..r * 8 + 8],
      &u[r * cw..r * cw + cw],
      &v[r * cw..r * cw + cw],
      r,
      M,
      FR,
    );
    PixelSink::process(&mut sink, row).unwrap();
  }
  assert!(sink.native_planar.is_some());
  // Frame 2: switch to Center AND attach RGB → the row-0 rebuild inserts the new
  // join, then the (now-needed, empty) RGB-scratch grow is refused.
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  sink.set_chroma_location(ChromaLocation::Center);
  sink.set_rgb(&mut rgb).unwrap();
  crate::sinker::mixed::arm_native_rgb_scratch_failure();
  let row0 = Yuv422pRow::for_tests(&y[0..8], &u[0..cw], &v[0..cw], 0, M, FR);
  let err = PixelSink::process(&mut sink, row0).unwrap_err();
  assert!(
    matches!(
      err,
      MixedSinkerError::Resample(ResampleError::AllocationFailed(_))
    ),
    "the post-insert RGB-scratch OOM must surface AllocationFailed, got {err:?}"
  );
  // state-atomic — the stale-gated restore brings the prior join back (the
  // delegate never committed the rebuilt join, since the scratch grow precedes
  // the insert).
  assert!(
    sink.native_planar.is_some(),
    "a rejected rebuild must restore the prior join"
  );
  // the output-set freeze was never committed (the delegate commits only after
  // the scratch grows), so a retry of row 0 with a CHANGED output set (luma
  // added) is ACCEPTED, not mis-rejected as ResampleOutputsChanged.
  sink.set_luma(&mut luma).unwrap();
  let row0b = Yuv422pRow::for_tests(&y[0..8], &u[0..cw], &v[0..cw], 0, M, FR);
  PixelSink::process(&mut sink, row0b).unwrap();
}

#[test]
#[cfg(feature = "rgb")]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn native_first_build_scratch_oom_leaves_freeze_unfrozen_for_retry() {
  use super::super::MixedSinkerError;
  use crate::resample::ResampleError;
  // An ORDINARY first-ever row-0 native build (no prior/stale join) grows the
  // output RGB scratch BEFORE the delegate commits the output-set freeze. A
  // scratch OOM on that first build therefore leaves `resample_outputs`
  // uncommitted (the delegate never freezes on a pre-feed failure), with no arm
  // rollback needed, so a retry with a changed output attachment is accepted,
  // not mis-rejected as ResampleOutputsChanged.
  let (y, u, v) = ramp(8, 8);
  let cw = 4usize;
  let mut rgb = vec![0u8; 4 * 4 * 3];
  let mut luma = vec![0u8; 4 * 4];
  let mut sink =
    MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_native(true)
      .with_chroma_location(ChromaLocation::Center)
      .with_rgb(&mut rgb)
      .unwrap();
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  crate::sinker::mixed::arm_native_rgb_scratch_failure();
  let row0 = Yuv422pRow::for_tests(&y[0..8], &u[0..cw], &v[0..cw], 0, M, FR);
  let err = PixelSink::process(&mut sink, row0).unwrap_err();
  assert!(
    matches!(
      err,
      MixedSinkerError::Resample(ResampleError::AllocationFailed(_))
    ),
    "first-build RGB-scratch OOM must surface AllocationFailed, got {err:?}"
  );
  // The freeze was never committed on the failed build — a retry of row 0 with
  // luma added (changed output set) is ACCEPTED, not ResampleOutputsChanged.
  sink.set_luma(&mut luma).unwrap();
  let row0b = Yuv422pRow::for_tests(&y[0..8], &u[0..cw], &v[0..cw], 0, M, FR);
  PixelSink::process(&mut sink, row0b).unwrap();
}

#[test]
#[cfg(feature = "rgb")]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn native_colour_capability_rebuild_scratch_oom_leaves_freeze_unfrozen_for_retry() {
  use super::super::MixedSinkerError;
  use crate::resample::ResampleError;
  // A cross-frame COLOUR-CAPABILITY rebuild: frame 1 builds a luma-only native
  // join (chroma absent, no output RGB scratch). Frame 2 attaches RGB, so the
  // delegate must rebuild the join WITH chroma and grow the (empty) output RGB
  // scratch — both before the row is accepted. A scratch OOM on that rebuild must
  // leave the cached luma-only join intact AND `resample_outputs` uncommitted, so
  // a row-0 retry with a CHANGED output set is accepted, not mis-rejected as
  // ResampleOutputsChanged. Pre-fix the committing preflight froze the luma-only
  // set and the up-front `*native = None` dropped the join before the failing
  // rebuild — both poisoning the retry.
  let (y, u, v) = ramp(8, 8);
  let cw = 4usize;
  let mut luma = vec![0u8; 4 * 4];
  let mut rgb = vec![0u8; 4 * 4 * 3];
  let (mut hh, mut ss, mut vv) = (vec![0u8; 4 * 4], vec![0u8; 4 * 4], vec![0u8; 4 * 4]);
  let mut sink =
    MixedSinker::<Yuv422p, AreaResampler>::with_resampler(8, 8, AreaResampler::to(4, 4))
      .unwrap()
      .with_native(true)
      .with_luma(&mut luma)
      .unwrap();
  // Frame 1: a luma-only native join (chroma absent, reserves no RGB scratch).
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  for r in 0..8 {
    let row = Yuv422pRow::for_tests(
      &y[r * 8..r * 8 + 8],
      &u[r * cw..r * cw + cw],
      &v[r * cw..r * cw + cw],
      r,
      M,
      FR,
    );
    PixelSink::process(&mut sink, row).unwrap();
  }
  assert!(
    sink.native_planar.is_some(),
    "frame 1 builds the luma-only native join"
  );
  // Frame 2: attach RGB → the row-0 rebuild must build the chroma half and grow
  // the (now-needed, empty) output RGB scratch; refuse that grow.
  PixelSink::begin_frame(&mut sink, 8, 8).unwrap();
  sink.set_rgb(&mut rgb).unwrap();
  crate::sinker::mixed::arm_native_rgb_scratch_failure();
  let row0 = Yuv422pRow::for_tests(&y[0..8], &u[0..cw], &v[0..cw], 0, M, FR);
  let err = PixelSink::process(&mut sink, row0).unwrap_err();
  assert!(
    matches!(
      err,
      MixedSinkerError::Resample(ResampleError::AllocationFailed(_))
    ),
    "the colour-capability rebuild scratch OOM must surface AllocationFailed, got {err:?}"
  );
  // The delegate built the replacement into a local and never cleared the field,
  // so the intact luma-only join survives the rejected rebuild.
  assert!(
    sink.native_planar.is_some(),
    "a rejected colour-capability rebuild must not drop the cached join"
  );
  // The output-set freeze was never committed (the delegate commits only after
  // the scratch grows), so a row-0 retry with ANOTHER output attached (add HSV)
  // is ACCEPTED, not mis-rejected as ResampleOutputsChanged.
  sink.set_hsv(&mut hh, &mut ss, &mut vv).unwrap();
  let row0b = Yuv422pRow::for_tests(&y[0..8], &u[0..cw], &v[0..cw], 0, M, FR);
  PixelSink::process(&mut sink, row0b).unwrap();
}