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
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
//! Sinker impls for source-side YUVA 4:2:0 formats — Yuva420p (8-bit),
//! Yuva420p9, Yuva420p10, Yuva420p16.
//!
//! Tranche 8b‑2a wires:
//! - **RGBA output paths (u8 + native-depth u16)** — alpha is sourced
//!   from the source alpha plane via the new
//!   `yuv_420*_to_rgba*_with_alpha_src_row` scalar kernels. SIMD lands
//!   in 8b‑2b (u8) / 8b‑2c (u16). The 8-bit `Yuva420p` has no native
//!   `u16` RGBA output.
//! - **RGB / RGB-u16 / luma / HSV alpha-drop paths** — these reuse the
//!   existing non-alpha 4:2:0 row dispatchers verbatim. The alpha plane
//!   is simply ignored; output bytes/elements match what the
//!   corresponding `Yuv420p*` source would produce given the same
//!   Y/U/V data. Without these paths `MixedSinker::with_rgb` /
//!   `with_luma` / `with_hsv` (declared on the generic impl) would
//!   silently accept a buffer and never write it.

use super::{
  ChromaSitingChanged,
  GeometryOverflow,
  HsvFrameMut,
  InsufficientBuffer,
  MixedSinker,
  MixedSinkerError,
  NativeRouteChanged,
  RowIndexOutOfRange,
  RowShapeMismatch,
  RowSlice,
  WidthAlignment,
  check_dimensions_match,
  check_frozen_alpha_mode,
  chroma_420_bottom_sited_v,
  chroma_420_center_sited_h,
  chroma_420_top_sited_v,
  deinterleave_y_high_bit_masked,
  packed_yuva444_filter_resample,
  packed_yuva444_resample,
  planar_8bit::{
    YUV422P_CENTERED_H_PHASE, reserve_420_chroma_full, reserve_420_chroma_prev,
    stage_420_chroma_prev, upsample_420_chroma_sited, yuva420p_process_native,
  },
  reset_high_bit_yuva_streams,
  rgb_row_buf_or_scratch,
  rgba_plane_row_slice,
  rgba_u16_plane_row_slice,
  // RFC #238 S6f — the planar `Yuv420pN` u16 `Bottom` (v = 1) reconstruction +
  // one-row chroma lookback, REUSED here (YUVA is planar, so the same separate-
  // plane helpers apply); NOT the p0xx interleaved kernel.
  subsampled_4_2_0_high_bit::{
    reserve_420_chroma_prev_u16, stage_420_chroma_prev_u16, upsample_420_chroma_sited_u16,
  },
};
use crate::{
  PixelSink,
  resample::{
    AveragingDomain, InsertionContext, InsertionPoint, PlanGeometry, ResampleError,
    select_insertion_point,
  },
  row::*,
  source::*,
};

/// `Yuva420p` ships the RFC #238 Phase 5 STRAIGHT-alpha native fast tier
/// ([`yuva420p_process_native`]), so it is statically eligible to splice an
/// area downscale at the native Y / U / V / A codes — but ONLY under
/// [`AlphaMode::Straight`](super::AlphaMode::Straight). Premultiplied alpha
/// is bin-then-convert-incompatible and stays on the packed-YUVA area tail.
const YUVA420P_STRAIGHT_NATIVE_ELIGIBLE: bool = true;

/// **Fallible preflight** for the centered-siting high-bit YUVA 4:2:0 chroma
/// scratch (#302) — the `u16` twin of
/// [`reserve_420_chroma_full`](super::reserve_420_chroma_full), identical to
/// the high-bit planar `Yuv420p9` … `Yuv420p16` sinks' private helper. Grows
/// `chroma_full` to the checked `2 * width` `u16` so the later infallible
/// [`upsample_420_chroma_sited_u16`] reuses an already-sized buffer. Split
/// out from the upsample so it can run **before any output row is written**
/// (luma included) — the crate's preflight-ordering atomicity contract (cf.
/// the #180 resample fix): an allocator refusal must leave the output frame
/// *untouched*, never partially mutated. The full-resolution source alpha
/// plane is never subsampled, so it is siting-independent and consumes no
/// scratch. `height` feeds the error payloads.
fn reserve_420_chroma_full_u16(
  chroma_full: &mut std::vec::Vec<u16>,
  width: usize,
  height: usize,
) -> Result<(), MixedSinkerError> {
  // Test-only failpoint shared with the planar / semi-planar centered paths
  // (`take()` fires the armed flag exactly once); the non-test build compiles
  // it away. Reuses the parent module's thread-local so the YUVA atomicity
  // test arms it the same way the planar suite does.
  #[cfg(all(test, feature = "std", feature = "yuv-planar"))]
  if super::FORCE_CHROMA_FULL_ALLOC_FAILURE.with(|f| f.take()) {
    return Err(MixedSinkerError::Resample(ResampleError::AllocationFailed(
      PlanGeometry::new(width, height, width, height),
    )));
  }
  let needed = width
    .checked_mul(2)
    .ok_or(MixedSinkerError::GeometryOverflow(GeometryOverflow::new(
      width, height, 2,
    )))?;
  if chroma_full.len() < needed {
    chroma_full
      .try_reserve_exact(needed - chroma_full.len())
      .map_err(|_| {
        MixedSinkerError::Resample(ResampleError::AllocationFailed(PlanGeometry::new(
          width, height, width, height,
        )))
      })?;
    chroma_full.resize(needed, 0);
  }
  Ok(())
}

/// Reserves an RFC #238 **Top** forward-delay held-row `u16` buffer
/// (`chroma_top_y_u16` for the deferred odd row's luma, `chroma_top_a_u16` for its
/// full-resolution alpha) to `width` `u16` up front, so the deferred-row copy and
/// the following even row's two-row flush are infallible (the #180
/// reserve-before-write contract). Idempotent once sized. The alpha companion of
/// the planar `Yuv420pN` held-Y reservation; YUVA additionally buffers the odd
/// row's alpha because the Top delay defers its whole decode.
fn reserve_420_chroma_top_u16(
  buf: &mut std::vec::Vec<u16>,
  width: usize,
  height: usize,
) -> Result<(), MixedSinkerError> {
  if buf.len() < width {
    buf.try_reserve_exact(width - buf.len()).map_err(|_| {
      MixedSinkerError::Resample(ResampleError::AllocationFailed(PlanGeometry::new(
        width, height, width, height,
      )))
    })?;
    buf.resize(width, 0);
  }
  Ok(())
}

/// The 8-bit twin of [`reserve_420_chroma_top_u16`] — reserves an RFC #238
/// **Top** forward-delay held-row `u8` buffer (`chroma_top_y` for the deferred
/// odd row's luma, `chroma_top_a` for its full-resolution alpha) to `width` `u8`.
fn reserve_420_chroma_top_u8(
  buf: &mut std::vec::Vec<u8>,
  width: usize,
  height: usize,
) -> Result<(), MixedSinkerError> {
  if buf.len() < width {
    buf.try_reserve_exact(width - buf.len()).map_err(|_| {
      MixedSinkerError::Resample(ResampleError::AllocationFailed(PlanGeometry::new(
        width, height, width, height,
      )))
    })?;
    buf.resize(width, 0);
  }
  Ok(())
}

/// The HIGH-BIT `Yuva420pN` identity 4:4:4 colour decode from an
/// ALREADY-reconstructed full-width chroma pair `(u_full, v_full)` and the row's
/// full-resolution alpha (all wire endianness), writing the requested colour
/// outputs (`rgb_u16` / `rgba_u16` / `rgb` / `rgba` / `hsv`) for output row
/// `[one_plane_start, one_plane_end)`. The alpha-threaded twin of the planar
/// `yuv444p_top_identity_color_row`: the RFC #238 **Top** (`v = 0`) forward
/// one-row delay invokes it for BOTH the deferred odd row (buffered wire Y + A +
/// forward-blended chroma) and the current even row (co-sited chroma) — the SAME
/// kernel sequence the inline centered decode runs, so a delayed decode is
/// bit-identical to an in-order one. The full-resolution α plane is never
/// chroma-subsampled, so it flows into the 4:4:4 (+ source-α) kernels UNCHANGED.
/// `rgb_scratch` must already be grown by the caller's atomicity preflight.
#[allow(clippy::too_many_arguments, clippy::type_complexity)]
fn yuva444p_high_bit_top_identity_color_row<const BITS: u32, const BE: bool>(
  rgb: &mut Option<&mut [u8]>,
  rgba: &mut Option<&mut [u8]>,
  rgb_u16: &mut Option<&mut [u16]>,
  rgba_u16: &mut Option<&mut [u16]>,
  hsv: &mut Option<HsvFrameMut<'_>>,
  rgb_scratch: &mut std::vec::Vec<u8>,
  y: &[u16],
  a: &[u16],
  u_full: &[u16],
  v_full: &[u16],
  one_plane_start: usize,
  one_plane_end: usize,
  w: usize,
  h: usize,
  matrix: crate::KernelMatrix,
  full_range: bool,
  use_simd: bool,
  rgb_444_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &mut [u8],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
  rgb_u16_444_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &mut [u16],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
  rgba_444_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &[u16],
    &mut [u8],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
  hsv_444_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &mut [u8],
    &mut [u8],
    &mut [u8],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
  rgba_u16_444_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &[u16],
    &mut [u16],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
) -> Result<(), MixedSinkerError> {
  let want_rgb = rgb.is_some();
  let want_rgba = rgba.is_some();
  let want_hsv = hsv.is_some();
  let want_rgb_u16 = rgb_u16.is_some();
  let want_rgba_u16 = rgba_u16.is_some();

  // u16 RGB / RGBA (Strategy A): rgb_u16 (alpha-drop) once, fan out to rgba_u16
  // via the cheap per-pixel pad then overwrite α from the source plane;
  // standalone rgba_u16 goes straight through the source-α dispatcher.
  if want_rgb_u16 {
    let buf = rgb_u16.as_deref_mut().unwrap();
    let rgb_plane_end = one_plane_end
      .checked_mul(3)
      .ok_or(MixedSinkerError::GeometryOverflow(GeometryOverflow::new(
        w, h, 3,
      )))?;
    let rgb_plane_start = one_plane_start * 3;
    let rgb_u16_row = &mut buf[rgb_plane_start..rgb_plane_end];
    rgb_u16_444_dispatch(
      y,
      u_full,
      v_full,
      rgb_u16_row,
      w,
      matrix,
      full_range,
      use_simd,
      BE,
    );
    if want_rgba_u16 {
      let rgba_buf = rgba_u16.as_deref_mut().unwrap();
      let rgba_u16_row = rgba_u16_plane_row_slice(rgba_buf, one_plane_start, one_plane_end, w, h)?;
      expand_rgb_u16_to_rgba_u16_row::<BITS>(rgb_u16_row, rgba_u16_row, w);
      crate::row::alpha_extract::copy_alpha_plane_u16::<BITS, BE>(a, rgba_u16_row, w, use_simd);
    }
  } else if want_rgba_u16 {
    let rgba_buf = rgba_u16.as_deref_mut().unwrap();
    let rgba_u16_row = rgba_u16_plane_row_slice(rgba_buf, one_plane_start, one_plane_end, w, h)?;
    rgba_u16_444_dispatch(
      y,
      u_full,
      v_full,
      a,
      rgba_u16_row,
      w,
      matrix,
      full_range,
      use_simd,
      BE,
    );
  }

  // u8 RGB / RGBA / HSV (Strategy A). HSV-without-RGB-or-RGBA takes the direct
  // HSV kernel; when RGB or RGBA is attached the RGB kernel runs and HSV derives
  // off that 8-bit buffer.
  let want_hsv_direct = want_hsv && !want_rgb && !want_rgba;
  let need_rgb_kernel = want_rgb || (want_hsv && want_rgba);
  if want_hsv_direct {
    let hsv = hsv.as_mut().expect("want_hsv_direct implies hsv attached");
    let (hh, ss, vv) = hsv.hsv();
    hsv_444_dispatch(
      y,
      u_full,
      v_full,
      &mut hh[one_plane_start..one_plane_end],
      &mut ss[one_plane_start..one_plane_end],
      &mut vv[one_plane_start..one_plane_end],
      w,
      matrix,
      full_range,
      use_simd,
      BE,
    );
    return Ok(());
  }
  if want_rgba && !need_rgb_kernel {
    let rgba_buf = rgba.as_deref_mut().unwrap();
    let rgba_row = rgba_plane_row_slice(rgba_buf, one_plane_start, one_plane_end, w, h)?;
    rgba_444_dispatch(
      y, u_full, v_full, a, rgba_row, w, matrix, full_range, use_simd, BE,
    );
    return Ok(());
  }
  if !need_rgb_kernel {
    return Ok(());
  }
  let rgb_row = rgb_row_buf_or_scratch(
    rgb.as_deref_mut(),
    rgb_scratch,
    one_plane_start,
    one_plane_end,
    w,
    h,
  )?;
  rgb_444_dispatch(
    y, u_full, v_full, rgb_row, w, matrix, full_range, use_simd, BE,
  );
  if let Some(hsv) = hsv.as_mut() {
    let (hh, ss, vv) = hsv.hsv();
    rgb_to_hsv_row(
      rgb_row,
      &mut hh[one_plane_start..one_plane_end],
      &mut ss[one_plane_start..one_plane_end],
      &mut vv[one_plane_start..one_plane_end],
      w,
      use_simd,
    );
  }
  if want_rgba {
    let rgba_buf = rgba.as_deref_mut().unwrap();
    let rgba_row = rgba_plane_row_slice(rgba_buf, one_plane_start, one_plane_end, w, h)?;
    expand_rgb_to_rgba_row(rgb_row, rgba_row, w);
    crate::row::alpha_extract::copy_alpha_plane_u16_to_u8::<BITS, BE>(a, rgba_row, w, use_simd);
  }
  Ok(())
}

/// The 8-bit `Yuva420p` identity 4:4:4 colour decode from an ALREADY-reconstructed
/// full-width chroma pair `(u_full, v_full)` and the row's full-resolution `u8`
/// alpha, writing the requested colour outputs (`rgb` / `rgba` / `hsv`) for output
/// row `[one_plane_start, one_plane_end)`. The 8-bit twin of
/// [`yuva444p_high_bit_top_identity_color_row`] (the 8-bit `Yuva420p` exposes no
/// u16 colour outputs). Invoked by the RFC #238 **Top** (`v = 0`) forward one-row
/// delay for BOTH the deferred odd row (buffered Y + A + forward-blended chroma)
/// and the current even row (co-sited chroma). α flows into the 4:4:4 kernels
/// UNCHANGED. `rgb_scratch` must already be grown by the caller's preflight.
#[allow(clippy::too_many_arguments)]
fn yuva444_top_identity_color_row(
  rgb: &mut Option<&mut [u8]>,
  rgba: &mut Option<&mut [u8]>,
  hsv: &mut Option<HsvFrameMut<'_>>,
  rgb_scratch: &mut std::vec::Vec<u8>,
  y: &[u8],
  a: &[u8],
  u_full: &[u8],
  v_full: &[u8],
  one_plane_start: usize,
  one_plane_end: usize,
  w: usize,
  h: usize,
  matrix: crate::KernelMatrix,
  full_range: bool,
  use_simd: bool,
) -> Result<(), MixedSinkerError> {
  let want_rgb = rgb.is_some();
  let want_rgba = rgba.is_some();
  let want_hsv = hsv.is_some();
  let want_hsv_direct = want_hsv && !want_rgb && !want_rgba;
  let need_rgb_kernel = want_rgb || (want_hsv && want_rgba);
  if want_hsv_direct {
    let hsv = hsv.as_mut().expect("want_hsv_direct implies hsv attached");
    let (hh, ss, vv) = hsv.hsv();
    yuv_444_to_hsv_row(
      y,
      u_full,
      v_full,
      &mut hh[one_plane_start..one_plane_end],
      &mut ss[one_plane_start..one_plane_end],
      &mut vv[one_plane_start..one_plane_end],
      w,
      matrix,
      full_range,
      use_simd,
    );
    return Ok(());
  }
  if want_rgba && !need_rgb_kernel {
    let rgba_buf = rgba.as_deref_mut().unwrap();
    let rgba_row = rgba_plane_row_slice(rgba_buf, one_plane_start, one_plane_end, w, h)?;
    yuva444p_to_rgba_row(
      y, u_full, v_full, a, rgba_row, w, matrix, full_range, use_simd,
    );
    return Ok(());
  }
  if !need_rgb_kernel {
    return Ok(());
  }
  let rgb_row = rgb_row_buf_or_scratch(
    rgb.as_deref_mut(),
    rgb_scratch,
    one_plane_start,
    one_plane_end,
    w,
    h,
  )?;
  yuv_444_to_rgb_row(y, u_full, v_full, rgb_row, w, matrix, full_range, use_simd);
  if let Some(hsv) = hsv.as_mut() {
    let (hh, ss, vv) = hsv.hsv();
    rgb_to_hsv_row(
      rgb_row,
      &mut hh[one_plane_start..one_plane_end],
      &mut ss[one_plane_start..one_plane_end],
      &mut vv[one_plane_start..one_plane_end],
      w,
      use_simd,
    );
  }
  if want_rgba {
    let rgba_buf = rgba.as_deref_mut().unwrap();
    let rgba_row = rgba_plane_row_slice(rgba_buf, one_plane_start, one_plane_end, w, h)?;
    expand_rgb_to_rgba_row(rgb_row, rgba_row, w);
    crate::row::alpha_extract::copy_alpha_plane_u8(a, rgba_row, w, use_simd);
  }
  Ok(())
}

// ---- Yuva420p impl (8-bit) ---------------------------------------------

impl<'a, R> MixedSinker<'a, Yuva420p, R> {
  /// Attaches a packed **8‑bit** RGBA output buffer. The 8‑bit YUVA
  /// source is converted to 8‑bit RGBA via the same Q15 i32 8‑bit
  /// kernel that backs [`MixedSinker<Yuv420p>::with_rgba`]; the
  /// per-pixel alpha byte is **sourced from the alpha plane** — not
  /// constant `0xFF`.
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_rgba(mut self, buf: &'a mut [u8]) -> Result<Self, MixedSinkerError> {
    self.set_rgba(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_rgba`](Self::with_rgba).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_rgba(&mut self, buf: &'a mut [u8]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_elems(4)?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientRgbaBuffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.rgba = Some(buf);
    Ok(self)
  }

  /// Attaches a **`u16`** luma output buffer. The 8-bit Y plane is
  /// zero-extended to u16 (`out[x] = Y_byte as u16`). Length in u16
  /// **elements** (`width x height`).
  ///
  /// Returns `Err(InsufficientLumaU16Buffer)` if `buf.len() < width x height`,
  /// or `Err(GeometryOverflow)` on 32-bit targets.
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_luma_u16(mut self, buf: &'a mut [u16]) -> Result<Self, MixedSinkerError> {
    self.set_luma_u16(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_luma_u16`](Self::with_luma_u16).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_luma_u16(&mut self, buf: &'a mut [u16]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_pixels()?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientLumaU16Buffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.luma_u16 = Some(buf);
    Ok(self)
  }
}

impl<R> Yuva420pSink for MixedSinker<'_, Yuva420p, R> {}

impl<R> PixelSink for MixedSinker<'_, Yuva420p, R> {
  type Input<'r> = Yuva420pRow<'r>;
  type Error = MixedSinkerError;

  #[cfg_attr(not(tarpaulin), inline(always))]
  fn kernel_matrix(&self) -> crate::KernelMatrix {
    self.kernel_matrix
  }
  fn begin_frame(&mut self, width: u32, height: u32) -> Result<(), Self::Error> {
    if self.width & 1 != 0 {
      return Err(MixedSinkerError::WidthAlignment(WidthAlignment::odd(
        self.width,
      )));
    }
    check_dimensions_match(self.width, self.height, width, height)?;
    // New frame: restart the 4-channel u8 RGBA colour stream and the
    // native-Y luma stream (both lazily created in `process`) and re-arm the
    // alpha-mode snapshot, mirroring the alpha-aware packed-YUVA (`Vuya`)
    // sink. The luma stream kind depends on the plan: the area path bins the
    // native Y at u16 (`luma_stream_u16`), the filter path resamples the
    // contiguous native Y at u8 (`luma_filter_stream`, parity with the
    // no-alpha `Yuv420p`). The 8-bit `Yuva420p` exposes no u16 colour
    // outputs, so its u16 RGBA streams are never created.
    if let Some(stream) = self.rgba_stream.as_mut() {
      stream.reset();
    }
    if let Some(stream) = self.luma_stream_u16.as_mut() {
      stream.reset();
    }
    if let Some(stream) = self.rgba_filter_stream.as_mut() {
      stream.reset();
    }
    if let Some(stream) = self.luma_filter_stream.as_mut() {
      stream.reset();
    }
    // Restart the RFC #238 Phase 5 straight-alpha native join (Y / U / V + α
    // streams), lazily created in `process`, so the next frame reuses it.
    if let Some(native) = self.native_yuva_420.as_mut() {
      native.reset();
    }
    // Clear the per-frame native/row-stage route freeze so the next frame may
    // pick either tier (the dispatch re-freezes on its first output-bearing
    // resampled row); a flip WITHIN a frame stays rejected. Mirrors the
    // high-bit families' `reset_high_bit_yuv_streams`.
    self.frozen_native_route = None;
    self.resample_outputs = None;
    // Clear the RFC #238 frozen 4:2:0 chroma siting so the next frame may pick
    // either phase; a mid-frame flip stays rejected. (The siting invalidation of
    // the cached native join is done at point-of-use in `process`, since
    // `chroma_location` may change AFTER `begin_frame` but before row 0.)
    self.frozen_chroma_centered = None;
    // RFC #238 S4-D companion: clear the frozen vertical phase so the next frame
    // may pick Center or Bottom, and invalidate the bottom-sited vertical-phase
    // chroma lookback (#302) so frame N+1's first even row can never box-blend
    // frame N's last chroma row (only the validity tag is cleared; the buffer
    // bytes are re-overwritten before any trusted read). A mid-frame Center ⇆
    // Bottom flip stays rejected.
    self.frozen_chroma_bottom_v = None;
    self.chroma_prev_row = None;
    // RFC #238 Top companion: clear the frozen vertical `Top` (`v = 0`) phase so
    // the next frame may pick Top while a mid-frame flip stays rejected, and drop
    // any held Top forward-delay odd row so frame N's last deferred colour row
    // never emits into frame N+1 (the held `chroma_top_y` / `chroma_top_a` bytes
    // are re-overwritten before any trusted read). REQUIRED — a leaked held row
    // is the Nv21 cross-frame-corruption regression class.
    self.frozen_chroma_top_v = None;
    self.chroma_top_pending = None;
    self.frozen_alpha_mode = Some(self.alpha_mode);
    Ok(())
  }

  fn process(&mut self, row: Yuva420pRow<'_>) -> Result<(), Self::Error> {
    let w = self.width;
    let h = self.height;
    let idx = row.row();
    let use_simd = self.simd;
    // Chroma siting (#302): drives the identity-plan horizontal chroma phase
    // for the Y/U/V colour decode. The full-resolution source alpha plane is
    // siting-independent (it is never subsampled), so it passes through
    // unchanged on every path. `Copy`, so read it out before the field
    // split-borrow below.
    let chroma_location = self.chroma_location.clone();

    if w & 1 != 0 {
      return Err(MixedSinkerError::WidthAlignment(WidthAlignment::odd(w)));
    }
    if row.y().len() != w {
      return Err(MixedSinkerError::RowShapeMismatch(RowShapeMismatch::new(
        RowSlice::Y,
        idx,
        w,
        row.y().len(),
      )));
    }
    if row.u_half().len() != w / 2 {
      return Err(MixedSinkerError::RowShapeMismatch(RowShapeMismatch::new(
        RowSlice::UHalf,
        idx,
        w / 2,
        row.u_half().len(),
      )));
    }
    if row.v_half().len() != w / 2 {
      return Err(MixedSinkerError::RowShapeMismatch(RowShapeMismatch::new(
        RowSlice::VHalf,
        idx,
        w / 2,
        row.v_half().len(),
      )));
    }
    if row.a().len() != w {
      return Err(MixedSinkerError::RowShapeMismatch(RowShapeMismatch::new(
        RowSlice::AFull,
        idx,
        w,
        row.a().len(),
      )));
    }
    if idx >= self.height {
      return Err(MixedSinkerError::RowIndexOutOfRange(
        RowIndexOutOfRange::new(idx, self.height),
      ));
    }

    // Non-identity plan: `Yuva420p` is 8-bit planar 4:2:0 YUV **with a
    // real full-resolution source alpha plane**. Route through the
    // packed-YUVA tail at `SRC_BITS = 8`: the u8 colour stream resamples the
    // converted u8 RGBA row (`yuva420p_to_rgba_row` — chroma upsampled per
    // 4:2:0, real source α from the alpha plane, NOT forced opaque) so
    // resampled alpha is a real mean and, under `AlphaMode::Premultiplied`,
    // colour is binned premultiplied; the native-Y luma stream resamples the
    // (zero-extended) Y plane directly so luma / luma_u16 are the
    // downscaled native Y, alpha- and range-independent (never derived from
    // the colour). The 8-bit `Yuva420p` exposes no u16 colour outputs, so
    // the tail's u16 colour resampling is never active (`rgb_u16` /
    // `rgba_u16` stay `None`) and its `convert_rgba_u16` closure is never
    // invoked.
    //
    // The span kind picks the engine: `Area` bins (the alpha-aware tail —
    // premultiplied colour is binned premultiplied then un-premultiplied);
    // `Filter` runs the signed-coefficient filter on the same converted RGBA
    // (PIL RGBA semantics — all four channels filtered independently,
    // straight alpha only). Premultiplied alpha has no filter analogue (the
    // engine cannot un-premultiply), so a premultiplied `Filter` plan is
    // routed to the area tail, which surfaces the typed `UnsupportedFilter`
    // rather than emitting straight-filtered premultiplied colour.
    if self.plan.is_some() {
      let alpha_mode = self.alpha_mode;
      let matrix = row.matrix();
      let full_range = row.full_range();
      let y = row.y();
      let u_half = row.u_half();
      let v_half = row.v_half();
      let a = row.a();
      let Self {
        rgb,
        rgba,
        rgb_u16,
        rgba_u16,
        luma,
        luma_u16,
        hsv,
        rgba_scratch,
        rgb_scratch,
        rgba_scratch_u16,
        rgba_color_scratch_u16,
        luma_scratch_u16,
        chroma_full,
        plan,
        rgba_stream,
        rgba_stream_u16,
        luma_stream_u16,
        rgba_filter_stream,
        rgba_filter_stream_u16,
        luma_filter_stream,
        luma_filter_stream_u16,
        resample_outputs,
        frozen_alpha_mode,
        native,
        native_yuva_420,
        frozen_native_route,
        frozen_chroma_centered,
        frozen_chroma_bottom_v,
        frozen_chroma_top_v,
        chroma_prev,
        chroma_prev_row,
        chroma_top_pending,
        chroma_top_y,
        chroma_top_a,
        ..
      } = self;
      let plan = plan.as_ref().expect("plan.is_some() checked above");
      check_frozen_alpha_mode(*frozen_alpha_mode, alpha_mode, idx)?;
      // RFC #238 S3c — 4:2:0 horizontal chroma siting for the alpha-bearing
      // source. The centered group (`Center` / `Top` / `Bottom`,
      // [`chroma_420_center_sited_h`]) samples chroma at `+0.5` luma horizontally;
      // the co-sited / unspecified group is phase 0 (today's decode,
      // byte-identical). Siting enters the chroma RECONSTRUCTION only: the native
      // fast tier folds the phase into the chroma area weights (`area_chroma_420`),
      // while the row-stage / filter converters reconstruct full-width chroma. The
      // full-resolution α plane is NEVER subsampled, so it is siting-independent
      // and passes through UNCHANGED on every tier. RFC #238 S4-D adds the
      // VERTICAL `Bottom` (`v = 1`) fold on top ([`chroma_420_bottom_sited_v`], a
      // strict sub-case of `center_sited` — `Bottom` is `h = 0.5, v = 1`): the
      // binning tiers (native / HSV-via-packed) fold the `v = 1` triangle into
      // `area_chroma_420`'s vertical weights, and the RGB-domain reconstruction
      // tiers box-blend the even output row's chroma with the previous chroma row
      // via the `chroma_prev` lookback. RFC #238 Top adds the DISJOINT FORWARD
      // `v = 0` fold ([`chroma_420_top_sited_v`]): the native tier folds it via
      // `area_chroma_420`'s `top_v`, and the RGB-domain tiers hold the odd row a
      // FORWARD one-row delay (see below). `Center` keeps `v_phase = 0` (co-sited
      // vertical, byte-identical to S3c). α is untouched on every path.
      let center_sited = chroma_420_center_sited_h(&chroma_location);
      let bottom_v = chroma_420_bottom_sited_v(&chroma_location);
      // RFC #238 Top (`v = 0`, FORWARD fold): `Top` shares `Center`'s centered
      // horizontal phase, `TopLeft` the co-sited one; both fold the DISJOINT
      // forward vertical triangle. The native binning tier folds it into
      // `area_chroma_420`'s vertical weights via `top_v` (no delay — the area
      // window absorbs the forward reach); the RGB-domain reconstruction tiers
      // (row-stage / filter) reconstruct each row's chroma through a FORWARD
      // one-row delay (the mirror of `Bottom`'s backward `chroma_prev` lookback).
      let top_v = chroma_420_top_sited_v(&chroma_location);
      let chroma_h_phase = if center_sited {
        YUV422P_CENTERED_H_PHASE
      } else {
        0.0
      };
      let chroma_v_phase = if bottom_v { 1.0 } else { 0.0 };
      // Whether this call carries any output — the EXACT set both tiers' no-output
      // short-circuit tests. Gates BOTH the siting freeze here and the
      // native/row-stage route freeze below, so a no-output call (which consumes
      // no stream state) freezes neither.
      let need_output =
        rgb.is_some() || rgba.is_some() || hsv.is_some() || luma.is_some() || luma_u16.is_some();
      // Colour outputs drive the centered chroma reconstruction; a luma-only
      // centered row bins the native Y (siting-independent) and never
      // reconstructs chroma.
      let want_color = rgb.is_some() || rgba.is_some() || hsv.is_some();
      // Freeze the effective 4:2:0 chroma siting on the first output-bearing row,
      // mirroring the no-alpha `Yuv420p` arm. This always-compiled CHECK sits at
      // the choke point every tier passes through; the matching SET rides each
      // tier's accept-time freeze below (never before dispatch, so a rejected row
      // leaves it unset for a corrected retry). A later row observing a different
      // phase — in sequence or not — would bin a mixture of co-sited and centered
      // chroma, so it is rejected here before any reconstruction. RFC #238 S4-D
      // also freezes the VERTICAL phase: `Center` and `Bottom` are both
      // horizontally centered, so a mid-frame Center ⇆ Bottom flip is invisible to
      // `center_sited` alone and is caught by the `frozen_chroma_bottom_v`
      // companion.
      if need_output
        && let Some(frozen) = *frozen_chroma_centered
        && (frozen != center_sited
          || *frozen_chroma_bottom_v != Some(bottom_v)
          || *frozen_chroma_top_v != Some(top_v))
      {
        return Err(MixedSinkerError::ChromaSitingChanged(
          ChromaSitingChanged::new(idx),
        ));
      }
      return match plan.kind() {
        crate::resample::SpanKind::Area => {
          // RFC #238 Phase 5 splice-stage selection for the area downscale.
          // STRAIGHT alpha is the #235 native resolution: bin Y / U / V / A
          // codes, convert once, attach the binned α → straight RGBA. The
          // framework selector reproduces the native-vs-row-stage choice — it
          // returns `NativeCodes` exactly when this format is straight-native
          // eligible, the sink enabled the native tier, AND the alpha mode is
          // straight (a premultiplied frame must convert-at-source then
          // premultiply then bin then un-premultiply late, so it is NOT
          // native-eligible and keeps the packed-YUVA area tail BYTE-IDENTICAL).
          // `area_plan` is always true on this arm (the filter plan is handled
          // by the arms below).
          let insertion = select_insertion_point(
            AveragingDomain::Encoded,
            InsertionContext {
              native_eligible: YUVA420P_STRAIGHT_NATIVE_ELIGIBLE && !alpha_mode.is_premultiplied(),
              with_native: *native,
              area_plan: true,
            },
          );
          let take_native = matches!(insertion, InsertionPoint::NativeCodes);
          // Reject a mid-frame native/row-stage route flip (a caller toggling
          // `set_native` between rows — native row 0, row-stage row 1)
          // BEFORE either tier's dispatch: the two tiers carry independent,
          // in-order, once-only stream state, so splitting a frame across them
          // would yield a mixed/partial frame rather than a deterministic
          // rejection. CHECKED here and frozen below (the SET), both gated on
          // `need_output`, so a no-output call is route-invisible. A
          // preflight-rejected output-bearing call returns Err before the SET,
          // leaving `frozen_native_route` untouched so a later retry on the
          // same (or other) route is not falsely rejected.
          if need_output
            && let Some(frozen) = *frozen_native_route
            && frozen != take_native
          {
            return Err(MixedSinkerError::NativeRouteChanged(
              NativeRouteChanged::new(idx),
            ));
          }
          match insertion {
            // Straight-alpha native tier. The AlphaMode freeze
            // (`check_frozen_alpha_mode` above) plus the inherited native
            // preflight (out-of-sequence / frozen-output / alloc, all before
            // any feed) keep the call atomic. Dispatch first; freeze the route
            // to native ONLY after the call returns Ok on an output-bearing
            // row — a no-output call returns Ok(()) with `need_output` false
            // (no freeze); an out-of-sequence / frozen / alloc-refused row
            // returns Err via `?` (no freeze), so only an accepted
            // output-bearing row commits the route (the Phase 2 set-after-accept
            // lesson — never freeze before the row is accepted).
            InsertionPoint::NativeCodes => {
              // RFC #238 S3c point-of-use siting invalidation, mirroring the
              // no-alpha `Yuv420p` native arm: `chroma_location` can change at ANY
              // point before this row (including AFTER `begin_frame`, before row
              // 0), so re-check the cached join HERE and drop it when its folded
              // chroma plan was built for a different phase; `yuva420p_process_native`
              // then rebuilds it with the current siting. Retry-atomic: drop ONLY on
              // the IN-SEQUENCE fresh-frame first row (`idx == 0`, `next_y() == 0`);
              // an out-of-sequence first row is left for the delegate to reject
              // against the INTACT join. A luma-only join carries no chroma plan
              // (siting-independent), a no-output sink built no join. Transactional:
              // move the stale join OUT, let the delegate build the replacement into
              // `native_yuva_420`, and restore the intact prior-phase join on a
              // rejected rebuild so the REJECTED row mutates nothing. The α stream is
              // siting-independent, rebuilt in lockstep and carrying no phase.
              let stale_native = idx == 0
                && native_yuva_420.as_ref().is_some_and(|join| {
                  join.has_chroma()
                    && (join.chroma_centered() != center_sited
                      || join.chroma_bottom() != bottom_v
                      || join.chroma_top() != top_v)
                    && join.next_y() == 0
                });
              let prev_native = if stale_native {
                native_yuva_420.take()
              } else {
                None
              };
              let native_result = yuva420p_process_native(
                plan,
                native_yuva_420,
                resample_outputs,
                rgb,
                rgba,
                luma,
                luma_u16,
                hsv,
                rgb_scratch,
                y,
                u_half,
                v_half,
                a,
                matrix,
                full_range,
                idx,
                w,
                h,
                chroma_h_phase,
                chroma_v_phase,
                top_v,
                use_simd,
              );
              if stale_native && native_result.is_err() {
                *native_yuva_420 = prev_native;
              }
              native_result?;
              if frozen_native_route.is_none() && need_output {
                *frozen_native_route = Some(true);
              }
              // RFC #238 S3c: freeze the horizontal siting on the same accepted
              // output row; S4-D / Top freeze the vertical `Bottom` / `Top` phase
              // alongside it.
              if frozen_chroma_centered.is_none() && need_output {
                *frozen_chroma_centered = Some(center_sited);
                *frozen_chroma_bottom_v = Some(bottom_v);
                *frozen_chroma_top_v = Some(top_v);
              }
              Ok(())
            }
            // Row-stage tier (the `with_native(false)` path, and EVERY
            // premultiplied frame): convert each source row to RGBA then bin —
            // the packed-YUVA area tail. Premultiplied colour is binned
            // premultiplied then un-premultiplied, byte-identical to today.
            // Same CHECK-before / SET-after split: freeze to row-stage only
            // when the call accepts an output-bearing row.
            InsertionPoint::EncodedOutput => {
              if top_v && want_color {
                // RFC #238 Top (`v = 0`) FORWARD one-row delay for the row-stage
                // tier — the mirror of `Bottom`'s backward `chroma_prev` lookback.
                // A `Top` chroma sample is co-sited with the TOP luma row of its
                // pair, so an EVEN source row decodes co-sited chroma while an ODD
                // source row needs the vertical box-average of its chroma row and
                // the NEXT one, which the row-at-a-time stream has not been fed when
                // the odd row arrives. The whole odd row is HELD (`chroma_top_pending`
                // + the buffered Y `chroma_top_y` AND its full-resolution alpha
                // `chroma_top_a`) and the following even row feeds TWO source rows:
                // the held odd row (forward-blended through the SAME `Bottom`-EVEN
                // `upsample_420_chroma_sited(bottom_v = true)` kernels at THIS even
                // index) then the current even row (co-sited). The trailing odd row
                // of an even-height frame clamps to a co-sited decode; an ODD
                // height's final even row is the two-row flush. Retry-atomic (#180):
                // the outer preflight + every reserve precede either feed, and the
                // packed streams size on row 0, so both feeds are infallible.
                let cursor = rgba_stream.as_ref().map_or(0, |s| s.next_y());
                let expected_source = cursor + usize::from(chroma_top_pending.is_some());
                if let core::ops::ControlFlow::Break(()) = super::resample_preflight_check_only(
                  resample_outputs,
                  luma,
                  luma_u16,
                  rgb,
                  rgba,
                  &None,
                  &None,
                  &None,
                  &None,
                  &None,
                  &None,
                  &None,
                  hsv,
                  &None,
                  Some(expected_source),
                  idx,
                )? {
                  return Ok(());
                }
                reserve_420_chroma_full(chroma_full, w, h)?;
                reserve_420_chroma_prev(chroma_prev, w, h)?;
                reserve_420_chroma_top_u8(chroma_top_y, w, h)?;
                reserve_420_chroma_top_u8(chroma_top_a, w, h)?;
                let is_last = idx + 1 == h;
                if idx & 1 == 0 {
                  // Flush the held odd predecessor FIRST, forward-blending the
                  // previous chroma row (`chroma_prev`) with the current one at
                  // this even index.
                  if let Some((p_idx, p_matrix, p_full_range)) = chroma_top_pending.take() {
                    let (u_full, v_full) = upsample_420_chroma_sited(
                      chroma_full,
                      chroma_prev,
                      chroma_prev_row,
                      u_half,
                      v_half,
                      idx,
                      true,
                      center_sited,
                      false,
                      w,
                      use_simd,
                    );
                    let held_y: &[u8] = &chroma_top_y[..w];
                    let held_a: &[u8] = &chroma_top_a[..w];
                    packed_yuva444_resample::<8>(
                      rgba_stream,
                      rgba_stream_u16,
                      luma_stream_u16,
                      resample_outputs,
                      rgb,
                      rgba,
                      rgb_u16,
                      rgba_u16,
                      luma,
                      luma_u16,
                      hsv,
                      rgba_scratch,
                      rgb_scratch,
                      rgba_scratch_u16,
                      rgba_color_scratch_u16,
                      luma_scratch_u16,
                      w,
                      plan,
                      p_idx,
                      use_simd,
                      alpha_mode,
                      |dst| {
                        yuva444p_to_rgba_row(
                          held_y,
                          u_full,
                          v_full,
                          held_a,
                          dst,
                          w,
                          p_matrix,
                          p_full_range,
                          use_simd,
                        )
                      },
                      |_dst: &mut [u16]| {},
                      |dst| {
                        for (d, &s) in dst.iter_mut().zip(held_y) {
                          *d = s as u16;
                        }
                      },
                    )?;
                  }
                  // Then feed the current EVEN row (co-sited chroma).
                  let (u_full, v_full) = upsample_420_chroma_sited(
                    chroma_full,
                    chroma_prev,
                    chroma_prev_row,
                    u_half,
                    v_half,
                    idx,
                    false,
                    center_sited,
                    false,
                    w,
                    use_simd,
                  );
                  packed_yuva444_resample::<8>(
                    rgba_stream,
                    rgba_stream_u16,
                    luma_stream_u16,
                    resample_outputs,
                    rgb,
                    rgba,
                    rgb_u16,
                    rgba_u16,
                    luma,
                    luma_u16,
                    hsv,
                    rgba_scratch,
                    rgb_scratch,
                    rgba_scratch_u16,
                    rgba_color_scratch_u16,
                    luma_scratch_u16,
                    w,
                    plan,
                    idx,
                    use_simd,
                    alpha_mode,
                    |dst| {
                      yuva444p_to_rgba_row(
                        y, u_full, v_full, a, dst, w, matrix, full_range, use_simd,
                      )
                    },
                    |_dst: &mut [u16]| {},
                    |dst| {
                      for (d, &s) in dst.iter_mut().zip(y) {
                        *d = s as u16;
                      }
                    },
                  )?;
                } else if !is_last {
                  // Defer this odd row: copy its Y + full-resolution alpha (the
                  // borrows expire after `process`) and record its decode params;
                  // it emits at the next even row.
                  chroma_top_y[..w].copy_from_slice(&y[..w]);
                  chroma_top_a[..w].copy_from_slice(&a[..w]);
                  *chroma_top_pending = Some((idx, matrix, full_range));
                } else {
                  // Trailing odd row (bottom-edge clamp → co-sited): feed it now.
                  let (u_full, v_full) = upsample_420_chroma_sited(
                    chroma_full,
                    chroma_prev,
                    chroma_prev_row,
                    u_half,
                    v_half,
                    idx,
                    false,
                    center_sited,
                    false,
                    w,
                    use_simd,
                  );
                  packed_yuva444_resample::<8>(
                    rgba_stream,
                    rgba_stream_u16,
                    luma_stream_u16,
                    resample_outputs,
                    rgb,
                    rgba,
                    rgb_u16,
                    rgba_u16,
                    luma,
                    luma_u16,
                    hsv,
                    rgba_scratch,
                    rgb_scratch,
                    rgba_scratch_u16,
                    rgba_color_scratch_u16,
                    luma_scratch_u16,
                    w,
                    plan,
                    idx,
                    use_simd,
                    alpha_mode,
                    |dst| {
                      yuva444p_to_rgba_row(
                        y, u_full, v_full, a, dst, w, matrix, full_range, use_simd,
                      )
                    },
                    |_dst: &mut [u16]| {},
                    |dst| {
                      for (d, &s) in dst.iter_mut().zip(y) {
                        *d = s as u16;
                      }
                    },
                  )?;
                }
                // Refresh the lookback with the current chroma row (after the reads
                // above) so the NEXT even row's flush forward-blends this row as its
                // predecessor.
                stage_420_chroma_prev(chroma_prev, chroma_prev_row, u_half, v_half, idx, w);
              } else if (center_sited || bottom_v) && want_color {
                // Sited row-stage: reconstruct full-width chroma (centered
                // `Center` / `Bottom`, or co-sited-H + `v = 1`
                // `BottomLeft`) and decode 4:4:4 — but ONLY after the resample
                // preflight (frozen-output + sequence), so an out-of-sequence /
                // rejected row is caught before the chroma reservation (#180).
                // `packed_yuva444_resample` re-runs the idempotent preflight.
                // Gated by `want_color` — a luma-only sited row bins the native Y
                // and never reconstructs chroma, so it stays on the co-sited arm
                // below. The α plane is full-resolution and siting-independent: it
                // flows into the 4:4:4 converter UNCHANGED (never chroma-upsampled).
                let expected = rgba_stream.as_ref().map_or(0, |s| s.next_y());
                if let core::ops::ControlFlow::Break(()) = super::resample_preflight_check_only(
                  resample_outputs,
                  luma,
                  luma_u16,
                  rgb,
                  rgba,
                  &None,
                  &None,
                  &None,
                  &None,
                  &None,
                  &None,
                  &None,
                  hsv,
                  &None,
                  Some(expected),
                  idx,
                )? {
                  return Ok(());
                }
                reserve_420_chroma_full(chroma_full, w, h)?;
                // RFC #238 S4-D: the `Bottom` / `BottomLeft` even output row
                // box-blends its chroma with the previous chroma row, so reserve
                // the lookback up front — AFTER the preflight, BEFORE the commit
                // (#180).
                if bottom_v {
                  reserve_420_chroma_prev(chroma_prev, w, h)?;
                }
                let (u_full, v_full) = upsample_420_chroma_sited(
                  chroma_full,
                  chroma_prev,
                  chroma_prev_row,
                  u_half,
                  v_half,
                  idx,
                  bottom_v,
                  center_sited,
                  false,
                  w,
                  use_simd,
                );
                packed_yuva444_resample::<8>(
                  rgba_stream,
                  rgba_stream_u16,
                  luma_stream_u16,
                  resample_outputs,
                  rgb,
                  rgba,
                  rgb_u16,
                  rgba_u16,
                  luma,
                  luma_u16,
                  hsv,
                  rgba_scratch,
                  rgb_scratch,
                  rgba_scratch_u16,
                  rgba_color_scratch_u16,
                  luma_scratch_u16,
                  w,
                  plan,
                  idx,
                  use_simd,
                  alpha_mode,
                  |dst| {
                    yuva444p_to_rgba_row(y, u_full, v_full, a, dst, w, matrix, full_range, use_simd)
                  },
                  // `Yuva420p` has no u16 colour outputs, so this is never called.
                  |_dst: &mut [u16]| {},
                  |dst| {
                    for (d, &s) in dst.iter_mut().zip(y) {
                      *d = s as u16;
                    }
                  },
                )?;
                // RFC #238 S4-D bottom lookback: advance ONLY after the row-stage
                // packed resample accepts the row (its `?` above already returned
                // any reject), so a rejected row leaves the predecessor for a clean
                // retry — the sited reconstruction read the lookback but did not
                // stage (deferred, #180).
                if bottom_v {
                  stage_420_chroma_prev(chroma_prev, chroma_prev_row, u_half, v_half, idx, w);
                }
              } else {
                packed_yuva444_resample::<8>(
                  rgba_stream,
                  rgba_stream_u16,
                  luma_stream_u16,
                  resample_outputs,
                  rgb,
                  rgba,
                  rgb_u16,
                  rgba_u16,
                  luma,
                  luma_u16,
                  hsv,
                  rgba_scratch,
                  rgb_scratch,
                  rgba_scratch_u16,
                  rgba_color_scratch_u16,
                  luma_scratch_u16,
                  w,
                  plan,
                  idx,
                  use_simd,
                  alpha_mode,
                  |dst| {
                    yuva420p_to_rgba_row(y, u_half, v_half, a, dst, w, matrix, full_range, use_simd)
                  },
                  // `Yuva420p` has no u16 colour outputs, so this is never called.
                  |_dst: &mut [u16]| {},
                  |dst| {
                    for (d, &s) in dst.iter_mut().zip(y) {
                      *d = s as u16;
                    }
                  },
                )?;
              }
              if frozen_native_route.is_none() && need_output {
                *frozen_native_route = Some(false);
              }
              // RFC #238 S3c: freeze the horizontal siting on the same accepted
              // output row; S4-D / Top freeze the vertical `Bottom` / `Top` phase
              // alongside it.
              if frozen_chroma_centered.is_none() && need_output {
                *frozen_chroma_centered = Some(center_sited);
                *frozen_chroma_bottom_v = Some(bottom_v);
                *frozen_chroma_top_v = Some(top_v);
              }
              Ok(())
            }
            // The encoded domain only resolves to native-codes / encoded-output;
            // the linear-light splice is reached via the Linear averaging domain,
            // which this 4:2:0 YUVA sink does not expose.
            InsertionPoint::LinearLight => {
              unreachable!("Yuva420p area downscale never selects the linear-light splice")
            }
          }
        }
        crate::resample::SpanKind::Filter if alpha_mode.is_premultiplied() => {
          // Premultiplied + filter has no analogue: route to the area tail
          // with the filter plan so it returns the typed `UnsupportedFilter`.
          packed_yuva444_resample::<8>(
            rgba_stream,
            rgba_stream_u16,
            luma_stream_u16,
            resample_outputs,
            rgb,
            rgba,
            rgb_u16,
            rgba_u16,
            luma,
            luma_u16,
            hsv,
            rgba_scratch,
            rgb_scratch,
            rgba_scratch_u16,
            rgba_color_scratch_u16,
            luma_scratch_u16,
            w,
            plan,
            idx,
            use_simd,
            alpha_mode,
            |dst| yuva420p_to_rgba_row(y, u_half, v_half, a, dst, w, matrix, full_range, use_simd),
            |_dst: &mut [u16]| {},
            |dst| {
              for (d, &s) in dst.iter_mut().zip(y) {
                *d = s as u16;
              }
            },
          )
        }
        crate::resample::SpanKind::Filter => {
          // Reject a multi-kernel (BICUBLIN) filter plan BEFORE the centered
          // preflight/reserve below: `packed_yuva444_filter_resample`'s FIRST act
          // is this same check, so without hoisting it a BICUBLIN plan would
          // freeze the output set and reserve/reconstruct chroma before being
          // rejected as `UnsupportedFilter` (the #180 reject-before-allocation
          // invariant). Idempotent — the delegate re-runs it.
          plan.ensure_single_kernel_filter()?;
          // RFC #238 Top (`v = 0`) FORWARD one-row delay for the single-kernel
          // filter tier — the exact structure of the row-stage Top branch above,
          // only the resampler kind differs (`packed_yuva444_filter_resample`). The
          // whole odd row is HELD and the following even row filter-feeds TWO source
          // rows (held forward-blend + current co-sited); the trailing odd row
          // clamps to a co-sited decode. A filter plan carries one kind for the
          // frame, so only the siting is frozen (on the accepted row); retry-atomic
          // (#180) — the outer preflight + every reserve precede either feed.
          if top_v && want_color {
            let cursor = rgba_filter_stream.as_ref().map_or(0, |s| s.next_y());
            let expected_source = cursor + usize::from(chroma_top_pending.is_some());
            if let core::ops::ControlFlow::Break(()) = super::resample_preflight_check_only(
              resample_outputs,
              luma,
              luma_u16,
              rgb,
              rgba,
              &None,
              &None,
              &None,
              &None,
              &None,
              &None,
              &None,
              hsv,
              &None,
              Some(expected_source),
              idx,
            )? {
              return Ok(());
            }
            reserve_420_chroma_full(chroma_full, w, h)?;
            reserve_420_chroma_prev(chroma_prev, w, h)?;
            reserve_420_chroma_top_u8(chroma_top_y, w, h)?;
            reserve_420_chroma_top_u8(chroma_top_a, w, h)?;
            let is_last = idx + 1 == h;
            if idx & 1 == 0 {
              if let Some((p_idx, p_matrix, p_full_range)) = chroma_top_pending.take() {
                let (u_full, v_full) = upsample_420_chroma_sited(
                  chroma_full,
                  chroma_prev,
                  chroma_prev_row,
                  u_half,
                  v_half,
                  idx,
                  true,
                  center_sited,
                  false,
                  w,
                  use_simd,
                );
                let held_y: &[u8] = &chroma_top_y[..w];
                let held_a: &[u8] = &chroma_top_a[..w];
                packed_yuva444_filter_resample::<8, true, false>(
                  rgba_filter_stream,
                  rgba_filter_stream_u16,
                  luma_filter_stream,
                  luma_filter_stream_u16,
                  resample_outputs,
                  rgb,
                  rgba,
                  rgb_u16,
                  rgba_u16,
                  luma,
                  luma_u16,
                  hsv,
                  rgba_scratch,
                  rgb_scratch,
                  rgba_scratch_u16,
                  rgba_color_scratch_u16,
                  luma_scratch_u16,
                  w,
                  plan,
                  p_idx,
                  use_simd,
                  held_y,
                  None,
                  |dst| {
                    yuva444p_to_rgba_row(
                      held_y,
                      u_full,
                      v_full,
                      held_a,
                      dst,
                      w,
                      p_matrix,
                      p_full_range,
                      use_simd,
                    )
                  },
                  |_dst: &mut [u16]| {},
                  |_dst: &mut [u16]| {},
                  |_dst: &mut [u8]| {},
                )?;
              }
              let (u_full, v_full) = upsample_420_chroma_sited(
                chroma_full,
                chroma_prev,
                chroma_prev_row,
                u_half,
                v_half,
                idx,
                false,
                center_sited,
                false,
                w,
                use_simd,
              );
              packed_yuva444_filter_resample::<8, true, false>(
                rgba_filter_stream,
                rgba_filter_stream_u16,
                luma_filter_stream,
                luma_filter_stream_u16,
                resample_outputs,
                rgb,
                rgba,
                rgb_u16,
                rgba_u16,
                luma,
                luma_u16,
                hsv,
                rgba_scratch,
                rgb_scratch,
                rgba_scratch_u16,
                rgba_color_scratch_u16,
                luma_scratch_u16,
                w,
                plan,
                idx,
                use_simd,
                y,
                None,
                |dst| {
                  yuva444p_to_rgba_row(y, u_full, v_full, a, dst, w, matrix, full_range, use_simd)
                },
                |_dst: &mut [u16]| {},
                |_dst: &mut [u16]| {},
                |_dst: &mut [u8]| {},
              )?;
            } else if !is_last {
              chroma_top_y[..w].copy_from_slice(&y[..w]);
              chroma_top_a[..w].copy_from_slice(&a[..w]);
              *chroma_top_pending = Some((idx, matrix, full_range));
            } else {
              let (u_full, v_full) = upsample_420_chroma_sited(
                chroma_full,
                chroma_prev,
                chroma_prev_row,
                u_half,
                v_half,
                idx,
                false,
                center_sited,
                false,
                w,
                use_simd,
              );
              packed_yuva444_filter_resample::<8, true, false>(
                rgba_filter_stream,
                rgba_filter_stream_u16,
                luma_filter_stream,
                luma_filter_stream_u16,
                resample_outputs,
                rgb,
                rgba,
                rgb_u16,
                rgba_u16,
                luma,
                luma_u16,
                hsv,
                rgba_scratch,
                rgb_scratch,
                rgba_scratch_u16,
                rgba_color_scratch_u16,
                luma_scratch_u16,
                w,
                plan,
                idx,
                use_simd,
                y,
                None,
                |dst| {
                  yuva444p_to_rgba_row(y, u_full, v_full, a, dst, w, matrix, full_range, use_simd)
                },
                |_dst: &mut [u16]| {},
                |_dst: &mut [u16]| {},
                |_dst: &mut [u8]| {},
              )?;
            }
            stage_420_chroma_prev(chroma_prev, chroma_prev_row, u_half, v_half, idx, w);
            if need_output && frozen_chroma_centered.is_none() {
              *frozen_chroma_centered = Some(center_sited);
              *frozen_chroma_bottom_v = Some(bottom_v);
              *frozen_chroma_top_v = Some(top_v);
            }
            return Ok(());
          }
          let r = if (center_sited || bottom_v) && want_color {
            // Sited filter: reconstruct full-width chroma (centered `Center` /
            // `Bottom`, or co-sited-H + `v = 1` `BottomLeft`) and decode
            // 4:4:4 — but ONLY after the filter preflight (frozen-output +
            // sequence), so an out-of-sequence / rejected row is caught before the
            // chroma reservation (#180). `packed_yuva444_filter_resample` re-runs
            // the idempotent preflight. The α plane is full-resolution and
            // siting-independent — it flows into the 4:4:4 converter UNCHANGED.
            let expected = rgba_filter_stream.as_ref().map_or(0, |s| s.next_y());
            if let core::ops::ControlFlow::Break(()) = super::resample_preflight_check_only(
              resample_outputs,
              luma,
              luma_u16,
              rgb,
              rgba,
              &None,
              &None,
              &None,
              &None,
              &None,
              &None,
              &None,
              hsv,
              &None,
              Some(expected),
              idx,
            )? {
              return Ok(());
            }
            reserve_420_chroma_full(chroma_full, w, h)?;
            // RFC #238 S4-D: reserve the bottom-sited lookback the even output row
            // box-blends against — AFTER the preflight, BEFORE the commit (#180).
            if bottom_v {
              reserve_420_chroma_prev(chroma_prev, w, h)?;
            }
            let (u_full, v_full) = upsample_420_chroma_sited(
              chroma_full,
              chroma_prev,
              chroma_prev_row,
              u_half,
              v_half,
              idx,
              bottom_v,
              center_sited,
              false,
              w,
              use_simd,
            );
            packed_yuva444_filter_resample::<8, true, false>(
              rgba_filter_stream,
              rgba_filter_stream_u16,
              luma_filter_stream,
              luma_filter_stream_u16,
              resample_outputs,
              rgb,
              rgba,
              rgb_u16,
              rgba_u16,
              luma,
              luma_u16,
              hsv,
              rgba_scratch,
              rgb_scratch,
              rgba_scratch_u16,
              rgba_color_scratch_u16,
              luma_scratch_u16,
              w,
              plan,
              idx,
              use_simd,
              y,
              None,
              |dst| {
                yuva444p_to_rgba_row(y, u_full, v_full, a, dst, w, matrix, full_range, use_simd)
              },
              |_dst: &mut [u16]| {},
              |_dst: &mut [u16]| {},
              |_dst: &mut [u8]| {},
            )
          } else {
            packed_yuva444_filter_resample::<8, true, false>(
              rgba_filter_stream,
              rgba_filter_stream_u16,
              luma_filter_stream,
              luma_filter_stream_u16,
              resample_outputs,
              rgb,
              rgba,
              rgb_u16,
              rgba_u16,
              luma,
              luma_u16,
              hsv,
              rgba_scratch,
              rgb_scratch,
              rgba_scratch_u16,
              rgba_color_scratch_u16,
              luma_scratch_u16,
              w,
              plan,
              idx,
              use_simd,
              // 8-bit native-Y luma rides the u8 stream (parity with `Yuv420p`):
              // the contiguous Y plane is fed directly, so no de-interleave scratch.
              y,
              None,
              |dst| {
                yuva420p_to_rgba_row(y, u_half, v_half, a, dst, w, matrix, full_range, use_simd)
              },
              // `Yuva420p` has no u16 colour outputs, so this closure is never called.
              |_dst: &mut [u16]| {},
              // u8-luma path: the u16 luma stream is detached, so this is never
              // called.
              |_dst: &mut [u16]| {},
              // Contiguous Y plane fed directly, so this u8 de-interleave is unused.
              |_dst: &mut [u8]| {},
            )
          };
          // RFC #238 S4-D bottom lookback: advance ONLY after the filter resample
          // accepts the row, so a rejected row leaves the predecessor for a clean
          // retry — the sited reconstruction read it but did not stage (deferred,
          // #180). Gated on `want_color`: a luma-only Bottom row took the co-sited
          // else arm (no reconstruction, no lookback read/stage).
          if r.is_ok() && bottom_v && want_color {
            stage_420_chroma_prev(chroma_prev, chroma_prev_row, u_half, v_half, idx, w);
          }
          // RFC #238 S3c: freeze the horizontal siting on the same accepted output
          // row (a rejected row leaves it unset for a corrected retry); S4-D / Top
          // freeze the vertical `Bottom` / `Top` phase alongside it. The filter plan
          // kind is fixed at construction, so there is no native/row-stage route
          // here.
          if r.is_ok() && need_output && frozen_chroma_centered.is_none() {
            *frozen_chroma_centered = Some(center_sited);
            *frozen_chroma_bottom_v = Some(bottom_v);
            *frozen_chroma_top_v = Some(top_v);
          }
          r
        }
      };
    }

    let Self {
      rgb,
      rgba,
      luma,
      luma_u16,
      hsv,
      rgb_scratch,
      chroma_full,
      chroma_prev,
      chroma_prev_row,
      // RFC #238 S4-D: the 4:2:0 horizontal + vertical (`Bottom`) chroma phase
      // frozen on the first output row; Top adds the disjoint `Top` (`v = 0`)
      // vertical phase + the forward one-row-delay held-odd-row state.
      frozen_chroma_centered,
      frozen_chroma_bottom_v,
      frozen_chroma_top_v,
      chroma_top_pending,
      chroma_top_y,
      chroma_top_a,
      ..
    } = self;

    let want_rgb = rgb.is_some();
    let want_rgba = rgba.is_some();
    let want_hsv = hsv.is_some();
    // A colour output that drives the centered / bottom chroma reconstruction;
    // the full-resolution α plane is siting-independent and never gates it.
    let want_color = want_rgb || want_rgba || want_hsv;
    // A row with no attached output is a no-op — return BEFORE the bottom-sited
    // lookback is primed (reserved / staged below), so a genuinely no-output row
    // never advances the vertical-phase lookback (mirrors the no-alpha `Yuv420p`
    // direct decode); it also guarantees ≥ 1 output was attached, so the
    // single-plane row ranges below cannot overflow.
    let need_output = want_color || luma.is_some() || luma_u16.is_some();
    if !need_output {
      return Ok(());
    }
    let one_plane_start = idx * w;
    let one_plane_end = one_plane_start + w;

    // Chroma siting (#302): the centered horizontal sitings reconstruct chroma
    // at the phase-0.5 position then feed the 4:4:4 (+ source-alpha) kernels;
    // the default / co-sited path keeps the byte-identical fused 4:2:0 decode.
    // RFC #238 S4-D: `Bottom` (`v = 1`) additionally box-blends the even output
    // row's chroma with the previous chroma row via the `chroma_prev` lookback.
    let center_sited = chroma_420_center_sited_h(&chroma_location);
    let bottom_v = chroma_420_bottom_sited_v(&chroma_location);
    // RFC #238 Top (`v = 0`, FORWARD fold): `Top` / `TopLeft` box-blend the ODD
    // output row's chroma with the NEXT chroma row — unavailable to a row-at-a-time
    // walk when the odd row arrives — so the identity colour decode holds the odd
    // row's whole output (its Y, its full-resolution alpha, and its decode params)
    // in a FORWARD one-row delay and emits it once the following even row supplies
    // the next chroma. Luma is siting-independent and written in order regardless.
    let top_v = chroma_420_top_sited_v(&chroma_location);

    // Per-frame chroma-siting freeze (RFC #238, mirroring the resample-path guard
    // above): the first output-bearing row pins the effective 4:2:0 phase — the
    // horizontal centered flag and BOTH vertical (`Bottom` / `Top`) flags. A later
    // row whose siting flipped would decode a mixture of phases into ONE frame, box-
    // blend against a STALE `chroma_prev` lookback, or leak a held odd row, so reject
    // it here BEFORE any scratch reserve, lookback priming, or output write. This
    // CHECK precedes the `stage_420_chroma_prev` / reconstruct staging below so a
    // rejected flip leaves `chroma_prev` / `chroma_prev_row` untouched (retry-atomic,
    // #180). `begin_frame` clears the freeze so the next frame may pick either phase.
    if need_output
      && let Some(frozen) = *frozen_chroma_centered
      && (frozen != center_sited
        || *frozen_chroma_bottom_v != Some(bottom_v)
        || *frozen_chroma_top_v != Some(top_v))
    {
      return Err(MixedSinkerError::ChromaSitingChanged(
        ChromaSitingChanged::new(idx),
      ));
    }

    // HSV-without-RGB-or-RGBA goes through the direct `yuv_420_to_hsv_row`
    // kernel (no source-width RGB scratch). HSV is colour-only — the
    // source alpha plane is dropped — so a YUVA HSV is byte-identical to
    // the no-alpha `Yuv420p` HSV on the same Y/U/V; the alpha-drop RGB
    // path already reuses `yuv_420_to_rgb_row`, so HSV reuses the matching
    // planar `yuv_420_to_hsv_row`. RGB or RGBA also attached keeps the
    // convert-once-then-derive path alive via `need_rgb_kernel` (the RGB
    // kernel runs anyway, so HSV derives off that buffer; an RGBA sink
    // still needs the alpha plane, so it stays on the RGB path).
    let want_hsv_direct = want_hsv && !want_rgb && !want_rgba;
    let need_rgb_kernel = want_rgb || (want_hsv && want_rgba);

    let matrix = row.matrix();
    let full_range = row.full_range();

    // Atomicity preflight (#302, cf. the crate's #180 resample fix): reserve
    // EVERY fallible row scratch this identity row needs BEFORE any output row
    // (luma / luma_u16 included) is written, so an allocator refusal returns a
    // typed error leaving the output frame untouched, never partially mutated.
    // Scratches that can grow:
    //  1. the centered-siting full-width chroma (`chroma_full`), on any colour
    //     decode with a non-co-sited phase — `Center` / `Bottom` / `Top` /
    //     `TopLeft` (the alpha plane never subsamples, so it is siting-independent);
    //  2. (RFC #238 S4-D) the bottom-sited (`Bottom`, `v = 1`) one-row chroma
    //     lookback (`chroma_prev`), on EVERY Bottom OUTPUT row — colour OR
    //     luma-only. RFC #238 Top rides the SAME lookback: its even-row flush of the
    //     held odd row forward-blends the previous chroma row, so it reserves and
    //     primes it too;
    //  3. (RFC #238 Top) the forward-delay held-odd-row buffers (`chroma_top_y` +
    //     its full-resolution alpha `chroma_top_a`), reserved so the deferred-row
    //     copy is infallible; and
    //  4. the u8 RGB row scratch, reached exactly when a colour decode needs an RGB
    //     row but no caller RGB buffer is borrowable (`want_hsv && want_rgba &&
    //     !want_rgb`).
    // The later `upsample_420_chroma_sited` / `rgb_row_buf_or_scratch` reuse the
    // already-sized buffers. `need_centered_chroma` drives ONLY the shared
    // Center / Bottom full-width recon below; Top takes a SEPARATE forward-delay
    // path (returns early), so it is EXCLUDED here (`!top_v`) to keep that shared
    // upsample — which would double-stage the lookback — from running for it.
    let need_centered_chroma = (center_sited || bottom_v) && want_color && !top_v;
    if need_centered_chroma || (top_v && want_color) {
      reserve_420_chroma_full(chroma_full, w, h)?;
    }
    if bottom_v || top_v {
      reserve_420_chroma_prev(chroma_prev, w, h)?;
    }
    if top_v && want_color {
      reserve_420_chroma_top_u8(chroma_top_y, w, h)?;
      reserve_420_chroma_top_u8(chroma_top_a, w, h)?;
    }
    // Grow the u8 RGB row scratch up front — before any caller-output write — so
    // an allocator refusal in the HSV-only scratch path (`want_hsv && want_rgba &&
    // !want_rgb`, the only fallible RGB acquisition) returns a recoverable error
    // rather than leaving a partially-written caller buffer; the later per-path
    // `rgb_row_buf_or_scratch` calls (the Top helper and the non-Top decode) then
    // reuse the already-sized scratch and are infallible.
    if want_hsv && want_rgba && !want_rgb {
      rgb_row_buf_or_scratch(
        rgb.as_deref_mut(),
        rgb_scratch,
        one_plane_start,
        one_plane_end,
        w,
        h,
      )?;
    }

    // Bottom / Top LUMA-ONLY row: the colour upsample (which refreshes the vertical
    // lookback via `stage = true`) won't run, so stage the current chroma row HERE,
    // after its reservation and BEFORE any luma write, so a LATER colour row in the
    // same frame box-blends it (a luma-only-then-colour in-order walk reconstructs
    // the same vertical phase as the all-output walk). A colour row instead stages
    // inside `upsample_420_chroma_sited` (Bottom) or the Top forward-delay branch,
    // so gate on `!want_color`; the `chroma_prev_row` tag still guards
    // out-of-sequence / cross-frame reads (→ clamp, never stale).
    if (bottom_v || top_v) && !want_color {
      stage_420_chroma_prev(
        chroma_prev,
        chroma_prev_row,
        row.u_half(),
        row.v_half(),
        idx,
        w,
      );
    }

    // Freeze the effective 4:2:0 phase on the first output-bearing row — AFTER the
    // fallible scratch reserves above have succeeded, so an `AllocationFailed` row
    // stays retryable (frozen stays unset); later rows are checked against it up
    // top. The horizontal centered flag and BOTH vertical (`Bottom` / `Top`) flags
    // are pinned together.
    if need_output && frozen_chroma_centered.is_none() {
      *frozen_chroma_centered = Some(center_sited);
      *frozen_chroma_bottom_v = Some(bottom_v);
      *frozen_chroma_top_v = Some(top_v);
    }

    // Luma — copy the Y plane verbatim (8-bit YUVA's Y is already u8). Written
    // in order for EVERY siting including Top (luma is siting-independent).
    if let Some(luma) = luma.as_deref_mut() {
      luma[one_plane_start..one_plane_end].copy_from_slice(&row.y()[..w]);
    }

    // Luma u16 — zero-extend the Y plane (`out[x] = Y_byte as u16`).
    if let Some(luma_u16) = luma_u16.as_deref_mut() {
      for (d, &s) in luma_u16[one_plane_start..one_plane_end]
        .iter_mut()
        .zip(&row.y()[..w])
      {
        *d = s as u16;
      }
    }

    // RFC #238 Top (`v = 0`) FORWARD one-row delay for the identity colour decode.
    // An EVEN output row is a plain co-sited decode; an ODD output row needs the
    // vertical box-average of its chroma row and the NEXT one, unavailable until
    // the following even row arrives, so each odd colour row is HELD
    // (`chroma_top_pending` + the buffered Y `chroma_top_y` AND its full-resolution
    // alpha `chroma_top_a`) and emitted at the following even row, forward-blended
    // through the SAME `Bottom`-EVEN `upsample_420_chroma_sited(bottom_v = true)`
    // kernels at the even index. The trailing odd row of the frame clamps to a
    // co-sited decode; an ODD height's final even row is the two-row flush. Luma
    // was already written above; only colour is delayed. α flows into the 4:4:4
    // kernels UNCHANGED (it is full-resolution, never chroma-subsampled).
    if top_v && want_color {
      let is_last = idx + 1 == h;
      if idx & 1 == 0 {
        // Flush the held odd predecessor FIRST, forward-blending the previous
        // chroma row (`chroma_prev`) with the current one at this even index.
        if let Some((p_idx, p_matrix, p_full_range)) = chroma_top_pending.take() {
          let (u_full, v_full) = upsample_420_chroma_sited(
            chroma_full,
            chroma_prev,
            chroma_prev_row,
            row.u_half(),
            row.v_half(),
            idx,
            true,
            center_sited,
            false,
            w,
            use_simd,
          );
          yuva444_top_identity_color_row(
            rgb,
            rgba,
            hsv,
            rgb_scratch,
            &chroma_top_y[..w],
            &chroma_top_a[..w],
            u_full,
            v_full,
            p_idx * w,
            p_idx * w + w,
            w,
            h,
            p_matrix,
            p_full_range,
            use_simd,
          )?;
        }
        // Then the current EVEN row (co-sited chroma).
        let (u_full, v_full) = upsample_420_chroma_sited(
          chroma_full,
          chroma_prev,
          chroma_prev_row,
          row.u_half(),
          row.v_half(),
          idx,
          false,
          center_sited,
          false,
          w,
          use_simd,
        );
        yuva444_top_identity_color_row(
          rgb,
          rgba,
          hsv,
          rgb_scratch,
          row.y(),
          row.a(),
          u_full,
          v_full,
          one_plane_start,
          one_plane_end,
          w,
          h,
          matrix,
          full_range,
          use_simd,
        )?;
      } else if !is_last {
        // Defer this odd row: copy its Y + full-resolution alpha (the borrows
        // expire after `process`) + record its decode params; its colour emits at
        // the next even row.
        chroma_top_y[..w].copy_from_slice(&row.y()[..w]);
        chroma_top_a[..w].copy_from_slice(&row.a()[..w]);
        *chroma_top_pending = Some((idx, matrix, full_range));
      } else {
        // Trailing odd row (bottom-edge clamp → co-sited): decode now.
        let (u_full, v_full) = upsample_420_chroma_sited(
          chroma_full,
          chroma_prev,
          chroma_prev_row,
          row.u_half(),
          row.v_half(),
          idx,
          false,
          center_sited,
          false,
          w,
          use_simd,
        );
        yuva444_top_identity_color_row(
          rgb,
          rgba,
          hsv,
          rgb_scratch,
          row.y(),
          row.a(),
          u_full,
          v_full,
          one_plane_start,
          one_plane_end,
          w,
          h,
          matrix,
          full_range,
          use_simd,
        )?;
      }
      // Refresh the lookback with the current chroma row so the NEXT even row's
      // odd-flush forward-blends this row as its predecessor (after the read above).
      stage_420_chroma_prev(
        chroma_prev,
        chroma_prev_row,
        row.u_half(),
        row.v_half(),
        idx,
        w,
      );
      return Ok(());
    }

    // Non-Top colour decode. Acquire the u8 RGB row buffer (infallible — the
    // scratch was grown up front) and reconstruct the shared Center / Bottom
    // full-width chroma once.
    let rgb_row = if need_rgb_kernel {
      Some(rgb_row_buf_or_scratch(
        rgb.as_deref_mut(),
        rgb_scratch,
        one_plane_start,
        one_plane_end,
        w,
        h,
      )?)
    } else {
      None
    };

    // Sited full-width chroma, reconstructed ONCE per row from the wire half-width
    // U / V and reused by every colour decode below — centered phase-0.5
    // (`Center`) or co-sited nearest-neighbor. For `Bottom` / `BottomLeft`
    // (`v = 1`, RFC #238 S4-D) the even output row additionally box-blends with the
    // previous chroma row (the `chroma_prev` lookback), and this call stages that
    // lookback for the next pair (`stage = true`: the direct decode's remaining
    // work is effectively infallible). Infallible — the scratches were reserved
    // above. The default left/unspecified siting leaves it `None`, so the fused
    // 4:2:0 kernels upsample chroma in-register and the output stays byte-identical.
    let centered = if need_centered_chroma {
      Some(upsample_420_chroma_sited(
        chroma_full,
        chroma_prev,
        chroma_prev_row,
        row.u_half(),
        row.v_half(),
        idx,
        bottom_v,
        center_sited,
        true,
        w,
        use_simd,
      ))
    } else {
      None
    };

    // HSV-only (no RGB / RGBA): convert the source Y/U/V straight to HSV
    // via the alpha-drop kernel — no source-width RGB scratch. Centered siting
    // (#302) routes through the 4:4:4 twin fed the full-width phase-0.5 chroma.
    if want_hsv_direct {
      let hsv = hsv.as_mut().expect("want_hsv_direct implies hsv attached");
      let (h, s, v) = hsv.hsv();
      if let Some((u_full, v_full)) = centered {
        yuv_444_to_hsv_row(
          row.y(),
          u_full,
          v_full,
          &mut h[one_plane_start..one_plane_end],
          &mut s[one_plane_start..one_plane_end],
          &mut v[one_plane_start..one_plane_end],
          w,
          row.matrix(),
          row.full_range(),
          use_simd,
        );
      } else {
        yuv_420_to_hsv_row(
          row.y(),
          row.u_half(),
          row.v_half(),
          &mut h[one_plane_start..one_plane_end],
          &mut s[one_plane_start..one_plane_end],
          &mut v[one_plane_start..one_plane_end],
          w,
          row.matrix(),
          row.full_range(),
          use_simd,
        );
      }
      return Ok(());
    }

    // ---- u8 RGB / RGBA / HSV path ----------------------------------
    // `need_rgb_kernel` / `rgb_row` were computed and (when needed)
    // allocated at the top, before any caller-output write.
    //
    // Direct-RGBA-only fast path routes through the alpha-source-aware
    // dispatcher (real source α, NOT forced opaque). Centered siting feeds the
    // 4:4:4 + source-alpha kernel the full-width phase-0.5 chroma; the alpha
    // plane is identical on both paths.
    if want_rgba && !need_rgb_kernel {
      let rgba_buf = rgba.as_deref_mut().unwrap();
      let rgba_row = rgba_plane_row_slice(rgba_buf, one_plane_start, one_plane_end, w, h)?;
      if let Some((u_full, v_full)) = centered {
        yuva444p_to_rgba_row(
          row.y(),
          u_full,
          v_full,
          row.a(),
          rgba_row,
          w,
          row.matrix(),
          row.full_range(),
          use_simd,
        );
      } else {
        yuva420p_to_rgba_row(
          row.y(),
          row.u_half(),
          row.v_half(),
          row.a(),
          rgba_row,
          w,
          row.matrix(),
          row.full_range(),
          use_simd,
        );
      }
      return Ok(());
    }

    // RGB kernel — alpha-drop reuses the planar dispatcher verbatim (the 4:4:4
    // twin for centered siting).
    let Some(rgb_row) = rgb_row else {
      return Ok(());
    };
    if let Some((u_full, v_full)) = centered {
      yuv_444_to_rgb_row(
        row.y(),
        u_full,
        v_full,
        rgb_row,
        w,
        row.matrix(),
        row.full_range(),
        use_simd,
      );
    } else {
      yuv_420_to_rgb_row(
        row.y(),
        row.u_half(),
        row.v_half(),
        rgb_row,
        w,
        row.matrix(),
        row.full_range(),
        use_simd,
      );
    }

    if let Some(hsv) = hsv.as_mut() {
      let (h, s, v) = hsv.hsv();
      rgb_to_hsv_row(
        rgb_row,
        &mut h[one_plane_start..one_plane_end],
        &mut s[one_plane_start..one_plane_end],
        &mut v[one_plane_start..one_plane_end],
        w,
        use_simd,
      );
    }

    // Both rgb and rgba attached (combo case): Strategy A+ — expand the
    // already-computed rgb_row → rgba_row (fills α = 0xFF), then
    // overwrite α slot from the source alpha plane. Avoids a second
    // chroma kernel — so the centered RGB row above is reused as-is, and only
    // the real source alpha is layered on, identical to the default path.
    if want_rgba {
      let rgba_buf = rgba.as_deref_mut().unwrap();
      let rgba_row = rgba_plane_row_slice(rgba_buf, one_plane_start, one_plane_end, w, h)?;
      expand_rgb_to_rgba_row(rgb_row, rgba_row, w);
      crate::row::alpha_extract::copy_alpha_plane_u8(row.a(), rgba_row, w, use_simd);
    }

    Ok(())
  }
}

// ---- Yuva420p9 impl ---------------------------------------------------

impl<'a, R, const BE: bool> MixedSinker<'a, Yuva420p9<BE>, R> {
  /// Attaches a packed **8-bit** RGBA output buffer. Source-derived
  /// alpha (depth-converted via `>> 1`).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_rgba(mut self, buf: &'a mut [u8]) -> Result<Self, MixedSinkerError> {
    self.set_rgba(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_rgba`](Self::with_rgba).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_rgba(&mut self, buf: &'a mut [u8]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_elems(4)?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientRgbaBuffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.rgba = Some(buf);
    Ok(self)
  }

  /// Attaches a packed **`u16`** RGBA output buffer (9‑bit
  /// low‑packed, `[0, 511]`). Alpha is sourced at native depth.
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_rgba_u16(mut self, buf: &'a mut [u16]) -> Result<Self, MixedSinkerError> {
    self.set_rgba_u16(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_rgba_u16`](Self::with_rgba_u16).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_rgba_u16(&mut self, buf: &'a mut [u16]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_elems(4)?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientRgbaU16Buffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.rgba_u16 = Some(buf);
    Ok(self)
  }

  /// Attaches a packed **`u16`** RGB output buffer (alpha-drop).
  /// Output is identical to [`MixedSinker<Yuv420p9>::with_rgb_u16`].
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_rgb_u16(mut self, buf: &'a mut [u16]) -> Result<Self, MixedSinkerError> {
    self.set_rgb_u16(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_rgb_u16`](Self::with_rgb_u16).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_rgb_u16(&mut self, buf: &'a mut [u16]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_elems(3)?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientRgbU16Buffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.rgb_u16 = Some(buf);
    Ok(self)
  }

  /// Attaches a **`u16`** luma output buffer. Luma is the **native Y**
  /// (the binned Y plane at native depth under a non-identity plan; the
  /// Y plane verbatim otherwise). Length in u16 **elements**
  /// (`width x height`).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_luma_u16(mut self, buf: &'a mut [u16]) -> Result<Self, MixedSinkerError> {
    self.set_luma_u16(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_luma_u16`](Self::with_luma_u16).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_luma_u16(&mut self, buf: &'a mut [u16]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_pixels()?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientLumaU16Buffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.luma_u16 = Some(buf);
    Ok(self)
  }
}

impl<R, const BE: bool> Yuva420p9Sink<BE> for MixedSinker<'_, Yuva420p9<BE>, R> {}

impl<R, const BE: bool> PixelSink for MixedSinker<'_, Yuva420p9<BE>, R> {
  type Input<'r> = Yuva420p9Row<'r>;
  type Error = MixedSinkerError;

  #[cfg_attr(not(tarpaulin), inline(always))]
  fn kernel_matrix(&self) -> crate::KernelMatrix {
    self.kernel_matrix
  }
  fn begin_frame(&mut self, width: u32, height: u32) -> Result<(), Self::Error> {
    if self.width & 1 != 0 {
      return Err(MixedSinkerError::WidthAlignment(WidthAlignment::odd(
        self.width,
      )));
    }
    check_dimensions_match(self.width, self.height, width, height)?;
    reset_high_bit_yuva_streams(self);
    Ok(())
  }

  fn process(&mut self, row: Yuva420p9Row<'_>) -> Result<(), Self::Error> {
    if self.plan.is_some() {
      return yuva420p_high_bit_resample::<9, BE>(
        self,
        row.row(),
        row.y(),
        row.u_half(),
        row.v_half(),
        row.a(),
        row.matrix(),
        row.full_range(),
        RowSlice::Y9,
        RowSlice::UHalf9,
        RowSlice::VHalf9,
        RowSlice::AFull9,
        yuva420p9_to_rgba_row_endian,
        yuva420p9_to_rgba_u16_row_endian,
        // Centered-siting (#302, RFC #238 S6c) 4:4:4 twins fed the horizontally
        // reconstructed full-width chroma; α passed through unchanged.
        yuva444p9_to_rgba_row_endian,
        yuva444p9_to_rgba_u16_row_endian,
      );
    }
    yuva420p_high_bit_process::<9, BE, _, _>(
      self,
      row.row(),
      row.y(),
      row.u_half(),
      row.v_half(),
      row.a(),
      row.matrix(),
      row.full_range(),
      RowSlice::Y9,
      RowSlice::UHalf9,
      RowSlice::VHalf9,
      RowSlice::AFull9,
      yuv420p9_to_rgb_row_endian,
      yuv420p9_to_rgb_u16_row_endian,
      yuva420p9_to_rgba_row_endian,
      yuv420p9_to_hsv_row_endian,
      yuva420p9_to_rgba_u16_row_endian,
      // Centered-siting (#302) 4:4:4 twins.
      yuv444p9_to_rgb_row_endian,
      yuv444p9_to_rgb_u16_row_endian,
      yuva444p9_to_rgba_row_endian,
      yuv444p9_to_hsv_row_endian,
      yuva444p9_to_rgba_u16_row_endian,
    )
  }
}

// ---- Yuva420p10 impl --------------------------------------------------

impl<'a, R, const BE: bool> MixedSinker<'a, Yuva420p10<BE>, R> {
  /// Attaches a packed **8-bit** RGBA output buffer. Source-derived
  /// alpha (depth-converted via `>> 2`).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_rgba(mut self, buf: &'a mut [u8]) -> Result<Self, MixedSinkerError> {
    self.set_rgba(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_rgba`](Self::with_rgba).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_rgba(&mut self, buf: &'a mut [u8]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_elems(4)?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientRgbaBuffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.rgba = Some(buf);
    Ok(self)
  }

  /// Attaches a packed **`u16`** RGBA output buffer (10‑bit
  /// low‑packed, `[0, 1023]`). Alpha is sourced at native depth.
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_rgba_u16(mut self, buf: &'a mut [u16]) -> Result<Self, MixedSinkerError> {
    self.set_rgba_u16(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_rgba_u16`](Self::with_rgba_u16).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_rgba_u16(&mut self, buf: &'a mut [u16]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_elems(4)?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientRgbaU16Buffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.rgba_u16 = Some(buf);
    Ok(self)
  }

  /// Attaches a packed **`u16`** RGB output buffer (alpha-drop).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_rgb_u16(mut self, buf: &'a mut [u16]) -> Result<Self, MixedSinkerError> {
    self.set_rgb_u16(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_rgb_u16`](Self::with_rgb_u16).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_rgb_u16(&mut self, buf: &'a mut [u16]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_elems(3)?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientRgbU16Buffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.rgb_u16 = Some(buf);
    Ok(self)
  }

  /// Attaches a **`u16`** luma output buffer. Luma is the **native Y**
  /// (the binned Y plane at native depth under a non-identity plan; the
  /// Y plane verbatim otherwise). Length in u16 **elements**
  /// (`width x height`).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_luma_u16(mut self, buf: &'a mut [u16]) -> Result<Self, MixedSinkerError> {
    self.set_luma_u16(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_luma_u16`](Self::with_luma_u16).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_luma_u16(&mut self, buf: &'a mut [u16]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_pixels()?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientLumaU16Buffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.luma_u16 = Some(buf);
    Ok(self)
  }
}

impl<R, const BE: bool> Yuva420p10Sink<BE> for MixedSinker<'_, Yuva420p10<BE>, R> {}

impl<R, const BE: bool> PixelSink for MixedSinker<'_, Yuva420p10<BE>, R> {
  type Input<'r> = Yuva420p10Row<'r>;
  type Error = MixedSinkerError;

  #[cfg_attr(not(tarpaulin), inline(always))]
  fn kernel_matrix(&self) -> crate::KernelMatrix {
    self.kernel_matrix
  }
  fn begin_frame(&mut self, width: u32, height: u32) -> Result<(), Self::Error> {
    if self.width & 1 != 0 {
      return Err(MixedSinkerError::WidthAlignment(WidthAlignment::odd(
        self.width,
      )));
    }
    check_dimensions_match(self.width, self.height, width, height)?;
    reset_high_bit_yuva_streams(self);
    Ok(())
  }

  fn process(&mut self, row: Yuva420p10Row<'_>) -> Result<(), Self::Error> {
    if self.plan.is_some() {
      return yuva420p_high_bit_resample::<10, BE>(
        self,
        row.row(),
        row.y(),
        row.u_half(),
        row.v_half(),
        row.a(),
        row.matrix(),
        row.full_range(),
        RowSlice::Y10,
        RowSlice::UHalf10,
        RowSlice::VHalf10,
        RowSlice::AFull10,
        yuva420p10_to_rgba_row_endian,
        yuva420p10_to_rgba_u16_row_endian,
        // Centered-siting (#302, RFC #238 S6c) 4:4:4 twins fed the horizontally
        // reconstructed full-width chroma; α passed through unchanged.
        yuva444p10_to_rgba_row_endian,
        yuva444p10_to_rgba_u16_row_endian,
      );
    }
    yuva420p_high_bit_process::<10, BE, _, _>(
      self,
      row.row(),
      row.y(),
      row.u_half(),
      row.v_half(),
      row.a(),
      row.matrix(),
      row.full_range(),
      RowSlice::Y10,
      RowSlice::UHalf10,
      RowSlice::VHalf10,
      RowSlice::AFull10,
      yuv420p10_to_rgb_row_endian,
      yuv420p10_to_rgb_u16_row_endian,
      yuva420p10_to_rgba_row_endian,
      yuv420p10_to_hsv_row_endian,
      yuva420p10_to_rgba_u16_row_endian,
      // Centered-siting (#302) 4:4:4 twins.
      yuv444p10_to_rgb_row_endian,
      yuv444p10_to_rgb_u16_row_endian,
      yuva444p10_to_rgba_row_endian,
      yuv444p10_to_hsv_row_endian,
      yuva444p10_to_rgba_u16_row_endian,
    )
  }
}

// ---- Yuva420p12 impl --------------------------------------------------

impl<'a, R, const BE: bool> MixedSinker<'a, Yuva420p12<BE>, R> {
  /// Attaches a packed **8-bit** RGBA output buffer. Source-derived
  /// alpha (depth-converted via `>> 4`).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_rgba(mut self, buf: &'a mut [u8]) -> Result<Self, MixedSinkerError> {
    self.set_rgba(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_rgba`](Self::with_rgba).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_rgba(&mut self, buf: &'a mut [u8]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_elems(4)?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientRgbaBuffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.rgba = Some(buf);
    Ok(self)
  }

  /// Attaches a packed **`u16`** RGBA output buffer (12‑bit
  /// low‑packed, `[0, 4095]`). Alpha is sourced at native depth.
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_rgba_u16(mut self, buf: &'a mut [u16]) -> Result<Self, MixedSinkerError> {
    self.set_rgba_u16(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_rgba_u16`](Self::with_rgba_u16).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_rgba_u16(&mut self, buf: &'a mut [u16]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_elems(4)?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientRgbaU16Buffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.rgba_u16 = Some(buf);
    Ok(self)
  }

  /// Attaches a packed **`u16`** RGB output buffer (alpha-drop).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_rgb_u16(mut self, buf: &'a mut [u16]) -> Result<Self, MixedSinkerError> {
    self.set_rgb_u16(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_rgb_u16`](Self::with_rgb_u16).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_rgb_u16(&mut self, buf: &'a mut [u16]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_elems(3)?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientRgbU16Buffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.rgb_u16 = Some(buf);
    Ok(self)
  }

  /// Attaches a **`u16`** luma output buffer. Luma is the **native Y**
  /// (the binned Y plane at native depth under a non-identity plan; the
  /// Y plane verbatim otherwise). Length in u16 **elements**
  /// (`width x height`).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_luma_u16(mut self, buf: &'a mut [u16]) -> Result<Self, MixedSinkerError> {
    self.set_luma_u16(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_luma_u16`](Self::with_luma_u16).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_luma_u16(&mut self, buf: &'a mut [u16]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_pixels()?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientLumaU16Buffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.luma_u16 = Some(buf);
    Ok(self)
  }
}

impl<R, const BE: bool> Yuva420p12Sink<BE> for MixedSinker<'_, Yuva420p12<BE>, R> {}

impl<R, const BE: bool> PixelSink for MixedSinker<'_, Yuva420p12<BE>, R> {
  type Input<'r> = Yuva420p12Row<'r>;
  type Error = MixedSinkerError;

  #[cfg_attr(not(tarpaulin), inline(always))]
  fn kernel_matrix(&self) -> crate::KernelMatrix {
    self.kernel_matrix
  }
  fn begin_frame(&mut self, width: u32, height: u32) -> Result<(), Self::Error> {
    if self.width & 1 != 0 {
      return Err(MixedSinkerError::WidthAlignment(WidthAlignment::odd(
        self.width,
      )));
    }
    check_dimensions_match(self.width, self.height, width, height)?;
    reset_high_bit_yuva_streams(self);
    Ok(())
  }

  fn process(&mut self, row: Yuva420p12Row<'_>) -> Result<(), Self::Error> {
    if self.plan.is_some() {
      return yuva420p_high_bit_resample::<12, BE>(
        self,
        row.row(),
        row.y(),
        row.u_half(),
        row.v_half(),
        row.a(),
        row.matrix(),
        row.full_range(),
        RowSlice::Y12,
        RowSlice::UHalf12,
        RowSlice::VHalf12,
        RowSlice::AFull12,
        yuva420p12_to_rgba_row_endian,
        yuva420p12_to_rgba_u16_row_endian,
        // Centered-siting (#302, RFC #238 S6c) 4:4:4 twins fed the horizontally
        // reconstructed full-width chroma; α passed through unchanged.
        yuva444p12_to_rgba_row_endian,
        yuva444p12_to_rgba_u16_row_endian,
      );
    }
    yuva420p_high_bit_process::<12, BE, _, _>(
      self,
      row.row(),
      row.y(),
      row.u_half(),
      row.v_half(),
      row.a(),
      row.matrix(),
      row.full_range(),
      RowSlice::Y12,
      RowSlice::UHalf12,
      RowSlice::VHalf12,
      RowSlice::AFull12,
      yuv420p12_to_rgb_row_endian,
      yuv420p12_to_rgb_u16_row_endian,
      yuva420p12_to_rgba_row_endian,
      yuv420p12_to_hsv_row_endian,
      yuva420p12_to_rgba_u16_row_endian,
      // Centered-siting (#302) 4:4:4 twins.
      yuv444p12_to_rgb_row_endian,
      yuv444p12_to_rgb_u16_row_endian,
      yuva444p12_to_rgba_row_endian,
      yuv444p12_to_hsv_row_endian,
      yuva444p12_to_rgba_u16_row_endian,
    )
  }
}

// ---- Yuva420p16 impl --------------------------------------------------

impl<'a, R, const BE: bool> MixedSinker<'a, Yuva420p16<BE>, R> {
  /// Attaches a packed **8-bit** RGBA output buffer. Source-derived
  /// alpha (depth-converted via `>> 8`).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_rgba(mut self, buf: &'a mut [u8]) -> Result<Self, MixedSinkerError> {
    self.set_rgba(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_rgba`](Self::with_rgba).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_rgba(&mut self, buf: &'a mut [u8]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_elems(4)?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientRgbaBuffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.rgba = Some(buf);
    Ok(self)
  }

  /// Attaches a packed **`u16`** RGBA output buffer (16‑bit, full
  /// `u16` range `[0, 65535]`). Alpha is sourced at native depth.
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_rgba_u16(mut self, buf: &'a mut [u16]) -> Result<Self, MixedSinkerError> {
    self.set_rgba_u16(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_rgba_u16`](Self::with_rgba_u16).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_rgba_u16(&mut self, buf: &'a mut [u16]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_elems(4)?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientRgbaU16Buffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.rgba_u16 = Some(buf);
    Ok(self)
  }

  /// Attaches a packed **`u16`** RGB output buffer (alpha-drop).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_rgb_u16(mut self, buf: &'a mut [u16]) -> Result<Self, MixedSinkerError> {
    self.set_rgb_u16(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_rgb_u16`](Self::with_rgb_u16).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_rgb_u16(&mut self, buf: &'a mut [u16]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_elems(3)?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientRgbU16Buffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.rgb_u16 = Some(buf);
    Ok(self)
  }

  /// Attaches a **`u16`** luma output buffer. Luma is the **native Y**
  /// (the binned Y plane at native depth under a non-identity plan; the
  /// Y plane verbatim otherwise). Length in u16 **elements**
  /// (`width x height`).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn with_luma_u16(mut self, buf: &'a mut [u16]) -> Result<Self, MixedSinkerError> {
    self.set_luma_u16(buf)?;
    Ok(self)
  }
  /// In-place variant of [`with_luma_u16`](Self::with_luma_u16).
  #[cfg_attr(not(tarpaulin), inline(always))]
  pub fn set_luma_u16(&mut self, buf: &'a mut [u16]) -> Result<&mut Self, MixedSinkerError> {
    let expected = self.frame_pixels()?;
    if buf.len() < expected {
      return Err(MixedSinkerError::InsufficientLumaU16Buffer(
        InsufficientBuffer::new(expected, buf.len()),
      ));
    }
    self.luma_u16 = Some(buf);
    Ok(self)
  }
}

impl<R, const BE: bool> Yuva420p16Sink<BE> for MixedSinker<'_, Yuva420p16<BE>, R> {}

impl<R, const BE: bool> PixelSink for MixedSinker<'_, Yuva420p16<BE>, R> {
  type Input<'r> = Yuva420p16Row<'r>;
  type Error = MixedSinkerError;

  #[cfg_attr(not(tarpaulin), inline(always))]
  fn kernel_matrix(&self) -> crate::KernelMatrix {
    self.kernel_matrix
  }
  fn begin_frame(&mut self, width: u32, height: u32) -> Result<(), Self::Error> {
    if self.width & 1 != 0 {
      return Err(MixedSinkerError::WidthAlignment(WidthAlignment::odd(
        self.width,
      )));
    }
    check_dimensions_match(self.width, self.height, width, height)?;
    reset_high_bit_yuva_streams(self);
    Ok(())
  }

  fn process(&mut self, row: Yuva420p16Row<'_>) -> Result<(), Self::Error> {
    if self.plan.is_some() {
      return yuva420p_high_bit_resample::<16, BE>(
        self,
        row.row(),
        row.y(),
        row.u_half(),
        row.v_half(),
        row.a(),
        row.matrix(),
        row.full_range(),
        RowSlice::Y16,
        RowSlice::UHalf16,
        RowSlice::VHalf16,
        RowSlice::AFull16,
        yuva420p16_to_rgba_row_endian,
        yuva420p16_to_rgba_u16_row_endian,
        // Centered-siting (#302, RFC #238 S6c) 4:4:4 twins fed the horizontally
        // reconstructed full-width chroma; α passed through unchanged.
        yuva444p16_to_rgba_row_endian,
        yuva444p16_to_rgba_u16_row_endian,
      );
    }
    yuva420p_high_bit_process::<16, BE, _, _>(
      self,
      row.row(),
      row.y(),
      row.u_half(),
      row.v_half(),
      row.a(),
      row.matrix(),
      row.full_range(),
      RowSlice::Y16,
      RowSlice::UHalf16,
      RowSlice::VHalf16,
      RowSlice::AFull16,
      yuv420p16_to_rgb_row_endian,
      yuv420p16_to_rgb_u16_row_endian,
      yuva420p16_to_rgba_row_endian,
      yuv420p16_to_hsv_row_endian,
      yuva420p16_to_rgba_u16_row_endian,
      // Centered-siting (#302) 4:4:4 twins.
      yuv444p16_to_rgb_row_endian,
      yuv444p16_to_rgb_u16_row_endian,
      yuva444p16_to_rgba_row_endian,
      yuv444p16_to_hsv_row_endian,
      yuva444p16_to_rgba_u16_row_endian,
    )
  }
}

// ---- Shared high-bit YUVA 4:2:0 process body --------------------------
//
// The 9 / 10 / 16-bit YUVA 4:2:0 sinker `process` bodies are
// structurally identical — only the depth-named row primitives, the
// `RowSlice` variants used in error reports, and the depth-conversion
// shift (`BITS - 8`) for luma differ. Factor the shared body into a
// generic helper to avoid 3x of the same ~70 lines.
//
// Strategy A combine for the alpha-drop paths (rgb_u16 alpha-drop +
// rgba_u16 source alpha): runs the rgb_u16 kernel into the caller
// buffer + the rgba_u16 kernel separately. Cannot reuse rgb_u16's
// output because alpha comes from the source plane; the expand
// helper would splat opaque alpha.
#[allow(clippy::too_many_arguments, clippy::type_complexity)]
#[cfg_attr(not(tarpaulin), inline(always))]
fn yuva420p_high_bit_process<const BITS: u32, const BE: bool, F: crate::SourceFormat, R>(
  sinker: &mut MixedSinker<'_, F, R>,
  idx: usize,
  y_row: &[u16],
  u_half_row: &[u16],
  v_half_row: &[u16],
  a_row: &[u16],
  matrix: crate::KernelMatrix,
  full_range: bool,
  y_slice: RowSlice,
  u_slice: RowSlice,
  v_slice: RowSlice,
  a_slice: RowSlice,
  // Each dispatch fn is the `_endian` variant — last `bool` is the runtime
  // `big_endian` flag the helper passes from `BE`. Function POINTERS (not
  // generic `Fn` bounds) so the 4:2:0 dispatcher and its 4:4:4 twin below —
  // distinct fn items of the SAME signature — coerce to one parameter type.
  rgb_dispatch: fn(&[u16], &[u16], &[u16], &mut [u8], usize, crate::KernelMatrix, bool, bool, bool),
  rgb_u16_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &mut [u16],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
  rgba_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &[u16],
    &mut [u8],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
  // The matching alpha-drop YUV→HSV `_endian` kernel — same Y/U/V inputs as
  // `rgb_dispatch`, three `&mut [u8]` H/S/V outputs.
  hsv_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &mut [u8],
    &mut [u8],
    &mut [u8],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
  rgba_u16_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &[u16],
    &mut [u16],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
  // Chroma siting (#302): the centered (phase-0.5) 4:4:4 twins of the five
  // dispatchers above. `*_444` rgb / rgb_u16 / hsv are the alpha-drop 4:4:4
  // kernels (`yuv444pN_to_rgb*` / `_to_hsv`); `*_444` rgba / rgba_u16 are the
  // source-alpha 4:4:4 kernels (`yuva444pN_to_rgba*`) — the alpha plane is
  // full-resolution, so it is fed unchanged. Each takes FULL-width U / V (the
  // reconstructed chroma); their signatures match the matching 4:2:0
  // dispatcher. They are invoked only on the centered sitings; the default
  // path never touches them.
  rgb_444_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &mut [u8],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
  rgb_u16_444_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &mut [u16],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
  rgba_444_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &[u16],
    &mut [u8],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
  hsv_444_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &mut [u8],
    &mut [u8],
    &mut [u8],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
  rgba_u16_444_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &[u16],
    &mut [u16],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
) -> Result<(), MixedSinkerError> {
  let w = sinker.width;
  let h = sinker.height;
  let use_simd = sinker.simd;
  // Chroma siting (#302): `Copy`, read before the field split-borrow below.
  let chroma_location = sinker.chroma_location.clone();

  if w & 1 != 0 {
    return Err(MixedSinkerError::WidthAlignment(WidthAlignment::odd(w)));
  }
  if y_row.len() != w {
    return Err(MixedSinkerError::RowShapeMismatch(RowShapeMismatch::new(
      y_slice,
      idx,
      w,
      y_row.len(),
    )));
  }
  if u_half_row.len() != w / 2 {
    return Err(MixedSinkerError::RowShapeMismatch(RowShapeMismatch::new(
      u_slice,
      idx,
      w / 2,
      u_half_row.len(),
    )));
  }
  if v_half_row.len() != w / 2 {
    return Err(MixedSinkerError::RowShapeMismatch(RowShapeMismatch::new(
      v_slice,
      idx,
      w / 2,
      v_half_row.len(),
    )));
  }
  if a_row.len() != w {
    return Err(MixedSinkerError::RowShapeMismatch(RowShapeMismatch::new(
      a_slice,
      idx,
      w,
      a_row.len(),
    )));
  }
  if idx >= sinker.height {
    return Err(MixedSinkerError::RowIndexOutOfRange(
      RowIndexOutOfRange::new(idx, sinker.height),
    ));
  }

  let MixedSinker {
    rgb,
    rgb_u16,
    rgba,
    rgba_u16,
    luma,
    luma_u16,
    hsv,
    rgb_scratch,
    chroma_full_u16,
    chroma_prev_u16,
    chroma_prev_row,
    // RFC #238 S6f: the 4:2:0 horizontal + vertical (`Bottom`) chroma phase frozen
    // on the first output row; Top adds the disjoint `Top` (`v = 0`) vertical phase
    // + the forward one-row-delay held-odd-row state (Y `chroma_top_y_u16` + its
    // full-resolution alpha `chroma_top_a_u16`).
    frozen_chroma_centered,
    frozen_chroma_bottom_v,
    frozen_chroma_top_v,
    chroma_top_pending,
    chroma_top_y_u16,
    chroma_top_a_u16,
    ..
  } = sinker;

  let want_rgb = rgb.is_some();
  let want_rgba = rgba.is_some();
  let want_hsv = hsv.is_some();
  let want_rgb_u16 = rgb_u16.is_some();
  let want_rgba_u16 = rgba_u16.is_some();
  // Any colour output drives the centered / bottom chroma reconstruction; the
  // full-resolution α plane is siting-independent and never gates it.
  let want_color = want_rgb || want_rgba || want_hsv || want_rgb_u16 || want_rgba_u16;
  // Repo-wide no-output invariant (RFC #238 S6f, mirroring the 8-bit YUVA /
  // `Yuv420p` direct decode): a `process` call carrying NO output — no colour, no
  // luma — runs NOTHING (no per-row offset arithmetic, no allocation, no
  // bottom-sited lookback prime). Returning HERE, before the `idx * w` offsets,
  // keeps the invariant overflow-safe (a no-output call never ran an attach-time
  // `w x h` validation, so `idx * w` could overflow on a 32-bit target) AND means
  // such a row never reserves `chroma_prev_u16` nor primes the lookback (so it can
  // never make a later colour even row box-blend through an invisible row).
  let need_output = want_color || luma.is_some() || luma_u16.is_some();
  if !need_output {
    return Ok(());
  }

  let one_plane_start = idx * w;
  let one_plane_end = one_plane_start + w;

  // Chroma siting (#302): the centered horizontal sitings reconstruct chroma
  // at the phase-0.5 position then feed the 4:4:4 (+ source-alpha) kernels; the
  // default / co-sited path keeps the byte-identical fused 4:2:0 decode.
  let center_sited = chroma_420_center_sited_h(&chroma_location);
  // RFC #238 S6f: `Bottom` (a strict sub-case of `center_sited` — h = 0.5, v = 1)
  // additionally box-blends the even output row's chroma with the previous chroma
  // row via the `chroma_prev_u16` lookback maintained below; `Center` keeps the
  // vertical-replicate (co-sited) decode, byte-identical to S6c.
  let bottom_v = chroma_420_bottom_sited_v(&chroma_location);
  // RFC #238 Top (`v = 0`, FORWARD fold): `Top` / `TopLeft` hold the ODD output
  // row's whole decode (its Y, its full-resolution alpha, and its decode params) in
  // a forward one-row delay and emit it at the following even row, forward-blended
  // through the `Bottom`-EVEN kernels at the even index; luma is siting-independent
  // and written in order regardless.
  let top_v = chroma_420_top_sited_v(&chroma_location);

  // Per-frame chroma-siting freeze (RFC #238, mirroring the resample-path guard):
  // the first output-bearing row pins the effective 4:2:0 phase — the horizontal
  // centered flag and BOTH vertical (`Bottom` / `Top`) flags. A later row whose
  // siting flipped would decode a mixture of phases into ONE frame, box-blend
  // against a STALE `chroma_prev_u16` lookback, or leak a held odd row, so reject
  // it here BEFORE any scratch reserve, lookback priming, or output write. This
  // CHECK precedes the `stage_420_chroma_prev_u16` / reconstruct staging below so a
  // rejected flip leaves `chroma_prev_u16` / `chroma_prev_row` untouched
  // (retry-atomic, #180). `begin_frame`'s `reset_high_bit_yuva_streams` clears the
  // freeze so the next frame may pick either phase.
  if need_output
    && let Some(frozen) = *frozen_chroma_centered
    && (frozen != center_sited
      || *frozen_chroma_bottom_v != Some(bottom_v)
      || *frozen_chroma_top_v != Some(top_v))
  {
    return Err(MixedSinkerError::ChromaSitingChanged(
      ChromaSitingChanged::new(idx),
    ));
  }

  // HSV-without-RGB-or-RGBA goes through the direct alpha-drop
  // `hsv_dispatch` kernel (no source-width RGB scratch). HSV is
  // colour-only — the source alpha plane is dropped — so a high-bit YUVA
  // HSV is byte-identical to the no-alpha planar HSV on the same Y/U/V
  // (the alpha-drop `rgb_dispatch` and `hsv_dispatch` share the same
  // depth-named kernel family). RGB or RGBA also attached keeps the
  // convert-once-then-derive path alive via `need_rgb_kernel` (an RGBA
  // sink still needs the alpha plane, so it stays on the RGB path).
  let want_hsv_direct = want_hsv && !want_rgb && !want_rgba;
  let need_rgb_kernel = want_rgb || (want_hsv && want_rgba);

  // Atomicity preflight (#302, cf. the crate's #180 resample fix): reserve
  // EVERY fallible row scratch this identity row needs BEFORE any output row
  // (luma included) is written, so an allocator refusal returns a typed error
  // leaving the output frame untouched, never partially mutated. Scratches that
  // can grow: the centered-siting full-width `u16` chroma (`chroma_full_u16`),
  // needed by ANY non-co-sited colour output (`Center` / `Bottom` / `Top` /
  // `TopLeft`); (RFC #238 S6f) the bottom-sited one-row chroma lookback
  // (`chroma_prev_u16`), on EVERY Bottom OUTPUT row — RFC #238 Top rides the SAME
  // lookback for its even-row forward blend; (RFC #238 Top) the forward-delay
  // held-odd-row buffers (`chroma_top_y_u16` + its full-resolution alpha
  // `chroma_top_a_u16`); and the u8 RGB row scratch (below). `need_centered_chroma`
  // drives ONLY the shared Center / Bottom full-width recon below; Top takes a
  // SEPARATE forward-delay path (returns early), so it is EXCLUDED here (`!top_v`)
  // to keep that shared upsample — which would double-stage the lookback — from
  // running for it.
  let need_centered_chroma = (center_sited || bottom_v) && want_color && !top_v;
  if need_centered_chroma || (top_v && want_color) {
    reserve_420_chroma_full_u16(chroma_full_u16, w, h)?;
  }
  // Reserve the bottom-sited lookback on EVERY bottom-sited OUTPUT row — colour OR
  // luma-only — because it is maintained so a LATER colour even row can box-blend
  // it (the luma-only staging below runs before any luma write). RFC #238 Top rides
  // the same lookback. A no-output row returned early above and never primes it.
  if bottom_v || top_v {
    reserve_420_chroma_prev_u16(chroma_prev_u16, w, h)?;
  }
  // RFC #238 Top forward-delay held-odd-row buffers, reserved up front so the
  // deferred-row copy of the odd Y + its full-resolution alpha is infallible.
  if top_v && want_color {
    reserve_420_chroma_top_u16(chroma_top_y_u16, w, h)?;
    reserve_420_chroma_top_u16(chroma_top_a_u16, w, h)?;
  }
  // Grow the u8 RGB row scratch up front — before any caller-output write — so an
  // allocator refusal in the HSV-only scratch path (`want_hsv && want_rgba &&
  // !want_rgb`, the only fallible RGB acquisition) returns a recoverable error
  // rather than leaving a partially-written caller buffer; the later per-path
  // `rgb_row_buf_or_scratch` calls (the Top helper and the non-Top decode) then
  // reuse the already-sized scratch and are infallible.
  if want_hsv && want_rgba && !want_rgb {
    rgb_row_buf_or_scratch(
      rgb.as_deref_mut(),
      rgb_scratch,
      one_plane_start,
      one_plane_end,
      w,
      h,
    )?;
  }

  // Bottom / Top LUMA-ONLY row (RFC #238 S6f / Top): the colour upsample (which
  // refreshes the vertical lookback via `stage = true`) won't run, so stage the
  // current chroma row HERE, after its reservation and BEFORE any luma write, so a
  // LATER colour row in the same frame box-blends it (a luma-only-then-colour
  // in-order walk reconstructs the same vertical phase as the all-output walk). A
  // colour row instead stages inside the sited upsample (Bottom) or the Top
  // forward-delay branch, so gate on `!want_color`; the `chroma_prev_row` tag still
  // guards out-of-sequence / cross-frame reads (→ clamp, never stale).
  if (bottom_v || top_v) && !want_color {
    stage_420_chroma_prev_u16(
      chroma_prev_u16,
      chroma_prev_row,
      u_half_row,
      v_half_row,
      idx,
      w,
    );
  }

  // Freeze the effective 4:2:0 phase on the first output-bearing row — AFTER the
  // fallible scratch reserves above have succeeded, so an `AllocationFailed` row
  // stays retryable (frozen stays unset); later rows are checked against it up top.
  // The horizontal centered flag and BOTH vertical (`Bottom` / `Top`) flags are
  // pinned together.
  if need_output && frozen_chroma_centered.is_none() {
    *frozen_chroma_centered = Some(center_sited);
    *frozen_chroma_bottom_v = Some(bottom_v);
    *frozen_chroma_top_v = Some(top_v);
  }

  // ---- luma (native Y; luma narrows `>> (BITS - 8)`, luma_u16 is the
  // host-native logical Y) ---------------------------------------------
  if luma.is_some() || luma_u16.is_some() {
    let luma_dst = luma.as_deref_mut();
    let luma_u16_dst = luma_u16.as_deref_mut();
    // Borrow each output's row slice up front so the per-pixel loop just
    // writes (the two outputs are disjoint fields).
    let mut luma_row = luma_dst.map(|b| &mut b[one_plane_start..one_plane_end]);
    let mut luma_u16_row = luma_u16_dst.map(|b| &mut b[one_plane_start..one_plane_end]);
    // Mask each decoded Y to the source's native depth `(1 << BITS) - 1`
    // (a no-op at `BITS = 16`). `Yuva420p*Frame::try_new` is geometry-only,
    // so a malformed-but-accepted frame can carry out-of-range Y (e.g.
    // `0x1000` at 12-bit); without this mask `luma_u16` would publish that
    // raw value, inconsistent with the `(1 << BITS) - 1`-masked Y the
    // RGB/RGBA row kernels decode from the same row.
    let sample_mask = ((1u32 << BITS) - 1) as u16;
    for (i, &s) in y_row.iter().enumerate().take(w) {
      // Normalize BE-encoded wire bytes to host-native before use —
      // without this, a valid BE sample like mid-gray `0x0200` (10-bit)
      // would be read as `0x0002` on a LE host and the `>> (BITS - 8)`
      // would write 0 instead of 128.
      let logical = (if BE { u16::from_be(s) } else { u16::from_le(s) }) & sample_mask;
      if let Some(row) = luma_row.as_deref_mut() {
        row[i] = (logical >> (BITS - 8)) as u8;
      }
      if let Some(row) = luma_u16_row.as_deref_mut() {
        row[i] = logical;
      }
    }
  }

  // RFC #238 Top (`v = 0`) FORWARD one-row delay for the identity colour decode.
  // An EVEN output row is a plain co-sited decode; an ODD output row needs the
  // vertical box-average of its chroma row and the NEXT one, unavailable until the
  // following even row arrives, so each odd colour row is HELD (`chroma_top_pending`
  // + the buffered wire Y `chroma_top_y_u16` AND its full-resolution alpha
  // `chroma_top_a_u16`) and emitted at the following even row, forward-blended
  // through the SAME `Bottom`-EVEN `upsample_420_chroma_sited_u16(bottom_v = true)`
  // kernels at the even index. The trailing odd row clamps to a co-sited decode; an
  // ODD height's final even row is the two-row flush. Luma was already written above
  // (siting-independent); only colour is delayed. α flows into the 4:4:4 (+ source-α)
  // kernels UNCHANGED (it is full-resolution, never chroma-subsampled).
  if top_v && want_color {
    let is_last = idx + 1 == h;
    if idx & 1 == 0 {
      // Flush the held odd predecessor FIRST, forward-blending the previous chroma
      // row (`chroma_prev_u16`) with the current one at this even index.
      if let Some((p_idx, p_matrix, p_full_range)) = chroma_top_pending.take() {
        let (u_full, v_full) = upsample_420_chroma_sited_u16::<BITS>(
          chroma_full_u16,
          chroma_prev_u16,
          chroma_prev_row,
          u_half_row,
          v_half_row,
          idx,
          true,
          center_sited,
          false,
          w,
          BE,
          use_simd,
        );
        yuva444p_high_bit_top_identity_color_row::<BITS, BE>(
          rgb,
          rgba,
          rgb_u16,
          rgba_u16,
          hsv,
          rgb_scratch,
          &chroma_top_y_u16[..w],
          &chroma_top_a_u16[..w],
          u_full,
          v_full,
          p_idx * w,
          p_idx * w + w,
          w,
          h,
          p_matrix,
          p_full_range,
          use_simd,
          rgb_444_dispatch,
          rgb_u16_444_dispatch,
          rgba_444_dispatch,
          hsv_444_dispatch,
          rgba_u16_444_dispatch,
        )?;
      }
      // Then the current EVEN row (co-sited chroma).
      let (u_full, v_full) = upsample_420_chroma_sited_u16::<BITS>(
        chroma_full_u16,
        chroma_prev_u16,
        chroma_prev_row,
        u_half_row,
        v_half_row,
        idx,
        false,
        center_sited,
        false,
        w,
        BE,
        use_simd,
      );
      yuva444p_high_bit_top_identity_color_row::<BITS, BE>(
        rgb,
        rgba,
        rgb_u16,
        rgba_u16,
        hsv,
        rgb_scratch,
        y_row,
        a_row,
        u_full,
        v_full,
        one_plane_start,
        one_plane_end,
        w,
        h,
        matrix,
        full_range,
        use_simd,
        rgb_444_dispatch,
        rgb_u16_444_dispatch,
        rgba_444_dispatch,
        hsv_444_dispatch,
        rgba_u16_444_dispatch,
      )?;
    } else if !is_last {
      // Defer this odd row: copy its wire Y + full-resolution alpha (the borrows
      // expire after `process`) + record its decode params; its colour emits at the
      // next even row.
      chroma_top_y_u16[..w].copy_from_slice(&y_row[..w]);
      chroma_top_a_u16[..w].copy_from_slice(&a_row[..w]);
      *chroma_top_pending = Some((idx, matrix, full_range));
    } else {
      // Trailing odd row (bottom-edge clamp → co-sited): decode now.
      let (u_full, v_full) = upsample_420_chroma_sited_u16::<BITS>(
        chroma_full_u16,
        chroma_prev_u16,
        chroma_prev_row,
        u_half_row,
        v_half_row,
        idx,
        false,
        center_sited,
        false,
        w,
        BE,
        use_simd,
      );
      yuva444p_high_bit_top_identity_color_row::<BITS, BE>(
        rgb,
        rgba,
        rgb_u16,
        rgba_u16,
        hsv,
        rgb_scratch,
        y_row,
        a_row,
        u_full,
        v_full,
        one_plane_start,
        one_plane_end,
        w,
        h,
        matrix,
        full_range,
        use_simd,
        rgb_444_dispatch,
        rgb_u16_444_dispatch,
        rgba_444_dispatch,
        hsv_444_dispatch,
        rgba_u16_444_dispatch,
      )?;
    }
    // Refresh the lookback with the current chroma row so the NEXT even row's
    // odd-flush forward-blends this row as its predecessor (after the read above).
    stage_420_chroma_prev_u16(
      chroma_prev_u16,
      chroma_prev_row,
      u_half_row,
      v_half_row,
      idx,
      w,
    );
    return Ok(());
  }

  // Non-Top colour decode. Acquire the u8 RGB row buffer (infallible — the scratch
  // was grown up front) and reconstruct the shared Center / Bottom full-width
  // chroma once, reused by every colour decode (u16 and u8). The default
  // left/unspecified siting leaves `centered` `None`, so the fused 4:2:0 kernels
  // upsample chroma in-register and the output stays byte-identical. `Center` takes
  // the plain horizontal phase-0.5 fold; `Bottom` (RFC #238 S6f) additionally
  // box-blends the even output row with the previous chroma row (`stage = true` —
  // the direct decode's post-reconstruction work is infallible).
  let rgb_row = if need_rgb_kernel {
    Some(rgb_row_buf_or_scratch(
      rgb.as_deref_mut(),
      rgb_scratch,
      one_plane_start,
      one_plane_end,
      w,
      h,
    )?)
  } else {
    None
  };
  let centered = if need_centered_chroma {
    Some(upsample_420_chroma_sited_u16::<BITS>(
      chroma_full_u16,
      chroma_prev_u16,
      chroma_prev_row,
      u_half_row,
      v_half_row,
      idx,
      bottom_v,
      center_sited,
      true,
      w,
      BE,
      use_simd,
    ))
  } else {
    None
  };

  // ---- u16 RGB / RGBA path --------------------------------------
  // rgb_u16 (alpha-drop) reuses the non-alpha dispatcher.
  // rgba_u16 — when rgb_u16 is also present (combo case), Strategy A+:
  // expand the already-computed rgb_u16_row → rgba_u16_row then
  // overwrite α slot from the source plane (avoids a second chroma
  // kernel). When rgba_u16 is alone, delegate to the alpha-source-aware
  // dispatcher directly (standalone path — already optimal). Centered siting
  // (#302) routes each kernel through its 4:4:4 twin fed the full-width
  // phase-0.5 chroma reconstructed above.
  if want_rgb_u16 {
    let buf = rgb_u16.as_deref_mut().unwrap();
    let rgb_plane_end = one_plane_end
      .checked_mul(3)
      .ok_or(MixedSinkerError::GeometryOverflow(GeometryOverflow::new(
        w, h, 3,
      )))?;
    let rgb_plane_start = one_plane_start * 3;
    let rgb_u16_row = &mut buf[rgb_plane_start..rgb_plane_end];
    if let Some((u_full, v_full)) = centered {
      rgb_u16_444_dispatch(
        y_row,
        u_full,
        v_full,
        rgb_u16_row,
        w,
        matrix,
        full_range,
        use_simd,
        BE,
      );
    } else {
      rgb_u16_dispatch(
        y_row,
        u_half_row,
        v_half_row,
        rgb_u16_row,
        w,
        matrix,
        full_range,
        use_simd,
        BE,
      );
    }
    if want_rgba_u16 {
      // Combo: expand rgb_u16_row → rgba_u16_row, then overwrite α slot.
      let rgba_buf = rgba_u16.as_deref_mut().unwrap();
      let rgba_u16_row = rgba_u16_plane_row_slice(rgba_buf, one_plane_start, one_plane_end, w, h)?;
      expand_rgb_u16_to_rgba_u16_row::<BITS>(rgb_u16_row, rgba_u16_row, w);
      // `BE` is plumbed from the row type so the alpha-plane copy honors
      // the same wire endianness as the caller's high-bit Yuva*p input.
      crate::row::alpha_extract::copy_alpha_plane_u16::<BITS, BE>(a_row, rgba_u16_row, w, use_simd);
    }
  } else if want_rgba_u16 {
    // Standalone rgba_u16: delegate to the alpha-source-aware dispatcher.
    let rgba_buf = rgba_u16.as_deref_mut().unwrap();
    let rgba_u16_row = rgba_u16_plane_row_slice(rgba_buf, one_plane_start, one_plane_end, w, h)?;
    if let Some((u_full, v_full)) = centered {
      rgba_u16_444_dispatch(
        y_row,
        u_full,
        v_full,
        a_row,
        rgba_u16_row,
        w,
        matrix,
        full_range,
        use_simd,
        BE,
      );
    } else {
      rgba_u16_dispatch(
        y_row,
        u_half_row,
        v_half_row,
        a_row,
        rgba_u16_row,
        w,
        matrix,
        full_range,
        use_simd,
        BE,
      );
    }
  }

  // ---- u8 RGB / RGBA / HSV path ----------------------------------
  // `need_rgb_kernel` / `rgb_row` were computed and (when needed)
  // allocated at the top, before any caller-output write.
  //
  // HSV-only (no u8 RGB / RGBA): convert the source Y/U/V straight to HSV
  // via the alpha-drop kernel — no source-width RGB scratch. Any u16
  // RGB/RGBA outputs already ran above. Centered siting (#302) routes through
  // the 4:4:4 twin fed the full-width phase-0.5 chroma.
  if want_hsv_direct {
    let hsv = hsv.as_mut().expect("want_hsv_direct implies hsv attached");
    let (h, s, v) = hsv.hsv();
    if let Some((u_full, v_full)) = centered {
      hsv_444_dispatch(
        y_row,
        u_full,
        v_full,
        &mut h[one_plane_start..one_plane_end],
        &mut s[one_plane_start..one_plane_end],
        &mut v[one_plane_start..one_plane_end],
        w,
        matrix,
        full_range,
        use_simd,
        BE,
      );
    } else {
      hsv_dispatch(
        y_row,
        u_half_row,
        v_half_row,
        &mut h[one_plane_start..one_plane_end],
        &mut s[one_plane_start..one_plane_end],
        &mut v[one_plane_start..one_plane_end],
        w,
        matrix,
        full_range,
        use_simd,
        BE,
      );
    }
    return Ok(());
  }

  // Direct-RGBA-only (real source α, NOT forced opaque). Centered siting feeds
  // the 4:4:4 + source-alpha kernel the full-width phase-0.5 chroma; the alpha
  // plane is identical on both paths.
  if want_rgba && !need_rgb_kernel {
    let rgba_buf = rgba.as_deref_mut().unwrap();
    let rgba_row = rgba_plane_row_slice(rgba_buf, one_plane_start, one_plane_end, w, h)?;
    if let Some((u_full, v_full)) = centered {
      rgba_444_dispatch(
        y_row, u_full, v_full, a_row, rgba_row, w, matrix, full_range, use_simd, BE,
      );
    } else {
      rgba_dispatch(
        y_row, u_half_row, v_half_row, a_row, rgba_row, w, matrix, full_range, use_simd, BE,
      );
    }
    return Ok(());
  }

  // RGB kernel (alpha-drop reuses the non-alpha dispatcher verbatim; the 4:4:4
  // twin for centered siting).
  let Some(rgb_row) = rgb_row else {
    return Ok(());
  };
  if let Some((u_full, v_full)) = centered {
    rgb_444_dispatch(
      y_row, u_full, v_full, rgb_row, w, matrix, full_range, use_simd, BE,
    );
  } else {
    rgb_dispatch(
      y_row, u_half_row, v_half_row, rgb_row, w, matrix, full_range, use_simd, BE,
    );
  }

  if let Some(hsv) = hsv.as_mut() {
    let (h, s, v) = hsv.hsv();
    rgb_to_hsv_row(
      rgb_row,
      &mut h[one_plane_start..one_plane_end],
      &mut s[one_plane_start..one_plane_end],
      &mut v[one_plane_start..one_plane_end],
      w,
      use_simd,
    );
  }

  // Both rgb and rgba attached (combo case): Strategy A+ — expand the
  // already-computed rgb_row → rgba_row (fills α = 0xFF), then
  // overwrite α slot from the source alpha plane with depth-conv
  // `>> (BITS - 8)`. Avoids a second chroma kernel.
  if want_rgba {
    let rgba_buf = rgba.as_deref_mut().unwrap();
    let rgba_row = rgba_plane_row_slice(rgba_buf, one_plane_start, one_plane_end, w, h)?;
    expand_rgb_to_rgba_row(rgb_row, rgba_row, w);
    // BE = false: see the rgba_u16 branch above for rationale.
    crate::row::alpha_extract::copy_alpha_plane_u16_to_u8::<BITS, BE>(a_row, rgba_row, w, use_simd);
  }

  Ok(())
}

// ---- Shared high-bit YUVA 4:2:0 resample-routing body -----------------
//
// The non-identity plan branch for the 9 / 10 / 16-bit YUVA 4:2:0 sinks.
// Mirrors the 8-bit `Yuva420p` plan branch but at native depth `BITS`. The
// `Area` (downscale) arm routes through the shared packed-YUVA area tail with
// THREE independent binnings — the u8 colour stream (the `_to_rgba` u8 kernel
// with real source α, chroma upsampled 4:2:0), the **independent** native u16
// colour stream (the `_to_rgba_u16` kernel — never a narrowing of the u8
// bin), and the **low-packed** native-Y luma stream
// (`deinterleave_y_high_bit_masked::<BITS, BE>`, a host-native copy masked to
// the native depth — planar YUVA Y stores logical values directly, unlike the
// high-bit-packed semi-planar P-formats, so luma is `binned_Y >> (BITS - 8)`).
// The `Filter` arm routes
// the SAME converted RGBA / native-Y through the signed-coefficient filter
// tail (`NATIVE_LUMA_U8 = false`, the u16-luma branch — native Y is u16),
// straight alpha only. The 4:2:0-vs-4:2:2 difference (chroma row index `r / 2`
// vs `r`) lives in the walker, so the 4:2:2 sinks reuse this body verbatim
// with the same half-chroma kernels.
#[allow(clippy::too_many_arguments, clippy::type_complexity)]
#[cfg_attr(not(tarpaulin), inline(always))]
fn yuva420p_high_bit_resample<const BITS: u32, const BE: bool>(
  sinker: &mut MixedSinker<'_, impl crate::SourceFormat, impl Sized>,
  idx: usize,
  y_row: &[u16],
  u_half_row: &[u16],
  v_half_row: &[u16],
  a_row: &[u16],
  matrix: crate::KernelMatrix,
  full_range: bool,
  y_slice: RowSlice,
  u_slice: RowSlice,
  v_slice: RowSlice,
  a_slice: RowSlice,
  rgba_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &[u16],
    &mut [u8],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
  rgba_u16_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &[u16],
    &mut [u16],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
  // Chroma siting (#302, RFC #238 S6c): the centered (phase-0.5) 4:4:4 twins of
  // the two source-alpha dispatchers above (`yuva444pN_to_rgba*`). Each takes
  // FULL-width U / V (the horizontally reconstructed chroma) with the α plane
  // fed unchanged; their signatures match the matching 4:2:0 dispatcher. They
  // are invoked ONLY on the centered horizontal sitings; the co-sited / default
  // path keeps the byte-identical fused 4:2:0 decode.
  rgba_444_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &[u16],
    &mut [u8],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
  rgba_u16_444_dispatch: fn(
    &[u16],
    &[u16],
    &[u16],
    &[u16],
    &mut [u16],
    usize,
    crate::KernelMatrix,
    bool,
    bool,
    bool,
  ),
) -> Result<(), MixedSinkerError> {
  let w = sinker.width;
  let h = sinker.height;
  let use_simd = sinker.simd;
  // Chroma siting (#302): `Copy`, read before the field split-borrow below.
  let chroma_location = sinker.chroma_location.clone();

  if w & 1 != 0 {
    return Err(MixedSinkerError::WidthAlignment(WidthAlignment::odd(w)));
  }
  if y_row.len() != w {
    return Err(MixedSinkerError::RowShapeMismatch(RowShapeMismatch::new(
      y_slice,
      idx,
      w,
      y_row.len(),
    )));
  }
  if u_half_row.len() != w / 2 {
    return Err(MixedSinkerError::RowShapeMismatch(RowShapeMismatch::new(
      u_slice,
      idx,
      w / 2,
      u_half_row.len(),
    )));
  }
  if v_half_row.len() != w / 2 {
    return Err(MixedSinkerError::RowShapeMismatch(RowShapeMismatch::new(
      v_slice,
      idx,
      w / 2,
      v_half_row.len(),
    )));
  }
  if a_row.len() != w {
    return Err(MixedSinkerError::RowShapeMismatch(RowShapeMismatch::new(
      a_slice,
      idx,
      w,
      a_row.len(),
    )));
  }
  if idx >= sinker.height {
    return Err(MixedSinkerError::RowIndexOutOfRange(
      RowIndexOutOfRange::new(idx, sinker.height),
    ));
  }

  let alpha_mode = sinker.alpha_mode;
  let MixedSinker {
    rgb,
    rgba,
    rgb_u16,
    rgba_u16,
    luma,
    luma_u16,
    hsv,
    rgba_scratch,
    rgb_scratch,
    rgba_scratch_u16,
    rgba_color_scratch_u16,
    luma_scratch_u16,
    chroma_full_u16,
    plan,
    rgba_stream,
    rgba_stream_u16,
    luma_stream_u16,
    rgba_filter_stream,
    rgba_filter_stream_u16,
    luma_filter_stream_u16,
    resample_outputs,
    frozen_alpha_mode,
    frozen_chroma_centered,
    frozen_chroma_bottom_v,
    frozen_chroma_top_v,
    chroma_prev_u16,
    chroma_prev_row,
    chroma_top_pending,
    chroma_top_y_u16,
    chroma_top_a_u16,
    ..
  } = sinker;
  let plan = plan.as_ref().expect("plan.is_some() checked by the caller");
  check_frozen_alpha_mode(*frozen_alpha_mode, alpha_mode, idx)?;
  // RFC #238 S6c / S6f — 4:2:0 HORIZONTAL + VERTICAL `Bottom` chroma siting for the
  // high-bit alpha-bearing source. The centered group (`chroma_420_center_sited_h`
  // — chroma at `+0.5` luma horizontally) reconstructs full-width chroma then feeds
  // the 4:4:4 (+ source-alpha) twins; the co-sited / unspecified group is phase 0
  // (today's fused 4:2:0 decode, byte-identical). The full-resolution α plane is
  // never subsampled, so it is siting-independent and passes through the 4:4:4
  // twins UNCHANGED. RFC #238 S6f adds the VERTICAL `Bottom` (`v = 1`) fold on top
  // ([`chroma_420_bottom_sited_v`], a strict sub-case of `center_sited`): the
  // reconstruction additionally box-blends the even output row's chroma with the
  // previous chroma row via the `chroma_prev_u16` lookback. `Center` keeps
  // `v_phase = 0` (co-sited vertical, byte-identical to S6c). This sink is
  // packed-only (no native code-domain tier), so `Bottom` folds via the RGB-domain
  // reconstruct-then-bin, never a native `area_chroma_phased_v` plan. RFC #238 Top
  // (`v = 0`, disjoint from `Bottom`) folds the FORWARD vertical triangle via a
  // one-row output delay: the odd row is HELD (its Y + full-resolution alpha) and
  // the following even row feeds TWO source rows through the SAME `Bottom`-EVEN
  // kernels — the RGB-domain reconstruct-then-bin, again with no native plan.
  let center_sited = chroma_420_center_sited_h(&chroma_location);
  let bottom_v = chroma_420_bottom_sited_v(&chroma_location);
  let top_v = chroma_420_top_sited_v(&chroma_location);
  // A colour output drives the centered chroma reconstruction (u8 OR native
  // u16); a luma-only row bins the native Y (siting-independent) and never
  // reconstructs chroma, so it stays on the co-sited fused arm.
  let want_color =
    rgb.is_some() || rgba.is_some() || hsv.is_some() || rgb_u16.is_some() || rgba_u16.is_some();
  // The EXACT set the packed tails treat as output-bearing (colour u8 / u16 or
  // luma); gates the siting freeze so a no-output call (which consumes no stream
  // state) freezes nothing. MUST include the u16 colour outputs — a u16-only
  // sink still advances a stream and must freeze the siting.
  let need_output = want_color || luma.is_some() || luma_u16.is_some();
  // Freeze the effective siting on the first output-bearing row: a later row
  // observing a different phase — in sequence or not — would bin a mixture of
  // co-sited and centered chroma, so it is rejected here before any
  // reconstruction. The matching SET rides each accepted output row below. RFC
  // #238 S6f also freezes the VERTICAL phase: `Center` and `Bottom` are both
  // horizontally centered, so a mid-frame Center ⇆ Bottom flip is invisible to
  // `center_sited` alone and is caught by the `frozen_chroma_bottom_v` companion.
  if need_output
    && let Some(frozen) = *frozen_chroma_centered
    && (frozen != center_sited
      || *frozen_chroma_bottom_v != Some(bottom_v)
      || *frozen_chroma_top_v != Some(top_v))
  {
    return Err(MixedSinkerError::ChromaSitingChanged(
      ChromaSitingChanged::new(idx),
    ));
  }

  // RFC #238 Top (`v = 0`) FORWARD one-row delay for the packed high-bit YUVA
  // reconstruction tiers (Area + single-kernel Filter). The whole odd row is HELD
  // (its wire Y `chroma_top_y_u16` + full-resolution alpha `chroma_top_a_u16` +
  // decode params) and the following even row feeds TWO source rows: the held odd
  // row (forward-blended through the `Bottom`-EVEN kernels at THIS even index) then
  // the current even row (co-sited). The trailing odd row clamps to a co-sited
  // decode; an ODD height's final even row is the two-row flush. Retry-atomic
  // (#180): the outer preflight + every reserve precede either feed, and the packed
  // streams size on row 0, so both feeds are infallible. Premultiplied + filter has
  // no analogue, so it is EXCLUDED here (`!(filter && premult)`) and falls through
  // to the normal filter arm, which surfaces the typed `UnsupportedFilter` before
  // any chroma reconstruction.
  if top_v && want_color && !(plan.kind().is_filter() && alpha_mode.is_premultiplied()) {
    match plan.kind() {
      crate::resample::SpanKind::Area => {
        let cursor = if rgb.is_some() || rgba.is_some() || hsv.is_some() {
          rgba_stream.as_ref().map_or(0, |s| s.next_y())
        } else {
          rgba_stream_u16.as_ref().map_or(0, |s| s.next_y())
        };
        let expected_source = cursor + usize::from(chroma_top_pending.is_some());
        if let core::ops::ControlFlow::Break(()) = super::resample_preflight_check_only(
          resample_outputs,
          luma,
          luma_u16,
          rgb,
          rgba,
          rgb_u16,
          rgba_u16,
          &None,
          &None,
          &None,
          &None,
          &None,
          hsv,
          &None,
          Some(expected_source),
          idx,
        )? {
          return Ok(());
        }
        reserve_420_chroma_full_u16(chroma_full_u16, w, h)?;
        reserve_420_chroma_prev_u16(chroma_prev_u16, w, h)?;
        reserve_420_chroma_top_u16(chroma_top_y_u16, w, h)?;
        reserve_420_chroma_top_u16(chroma_top_a_u16, w, h)?;
        let is_last = idx + 1 == h;
        if idx & 1 == 0 {
          if let Some((p_idx, p_matrix, p_full_range)) = chroma_top_pending.take() {
            let (u_full, v_full) = upsample_420_chroma_sited_u16::<BITS>(
              chroma_full_u16,
              chroma_prev_u16,
              chroma_prev_row,
              u_half_row,
              v_half_row,
              idx,
              true,
              center_sited,
              false,
              w,
              BE,
              use_simd,
            );
            let held_y: &[u16] = &chroma_top_y_u16[..w];
            let held_a: &[u16] = &chroma_top_a_u16[..w];
            packed_yuva444_resample::<BITS>(
              rgba_stream,
              rgba_stream_u16,
              luma_stream_u16,
              resample_outputs,
              rgb,
              rgba,
              rgb_u16,
              rgba_u16,
              luma,
              luma_u16,
              hsv,
              rgba_scratch,
              rgb_scratch,
              rgba_scratch_u16,
              rgba_color_scratch_u16,
              luma_scratch_u16,
              w,
              plan,
              p_idx,
              use_simd,
              alpha_mode,
              |dst| {
                rgba_444_dispatch(
                  held_y,
                  u_full,
                  v_full,
                  held_a,
                  dst,
                  w,
                  p_matrix,
                  p_full_range,
                  use_simd,
                  BE,
                )
              },
              |dst| {
                rgba_u16_444_dispatch(
                  held_y,
                  u_full,
                  v_full,
                  held_a,
                  dst,
                  w,
                  p_matrix,
                  p_full_range,
                  use_simd,
                  BE,
                )
              },
              |dst| deinterleave_y_high_bit_masked::<BITS, BE>(held_y, dst, w),
            )?;
          }
          let (u_full, v_full) = upsample_420_chroma_sited_u16::<BITS>(
            chroma_full_u16,
            chroma_prev_u16,
            chroma_prev_row,
            u_half_row,
            v_half_row,
            idx,
            false,
            center_sited,
            false,
            w,
            BE,
            use_simd,
          );
          packed_yuva444_resample::<BITS>(
            rgba_stream,
            rgba_stream_u16,
            luma_stream_u16,
            resample_outputs,
            rgb,
            rgba,
            rgb_u16,
            rgba_u16,
            luma,
            luma_u16,
            hsv,
            rgba_scratch,
            rgb_scratch,
            rgba_scratch_u16,
            rgba_color_scratch_u16,
            luma_scratch_u16,
            w,
            plan,
            idx,
            use_simd,
            alpha_mode,
            |dst| {
              rgba_444_dispatch(
                y_row, u_full, v_full, a_row, dst, w, matrix, full_range, use_simd, BE,
              )
            },
            |dst| {
              rgba_u16_444_dispatch(
                y_row, u_full, v_full, a_row, dst, w, matrix, full_range, use_simd, BE,
              )
            },
            |dst| deinterleave_y_high_bit_masked::<BITS, BE>(y_row, dst, w),
          )?;
        } else if !is_last {
          chroma_top_y_u16[..w].copy_from_slice(&y_row[..w]);
          chroma_top_a_u16[..w].copy_from_slice(&a_row[..w]);
          *chroma_top_pending = Some((idx, matrix, full_range));
        } else {
          let (u_full, v_full) = upsample_420_chroma_sited_u16::<BITS>(
            chroma_full_u16,
            chroma_prev_u16,
            chroma_prev_row,
            u_half_row,
            v_half_row,
            idx,
            false,
            center_sited,
            false,
            w,
            BE,
            use_simd,
          );
          packed_yuva444_resample::<BITS>(
            rgba_stream,
            rgba_stream_u16,
            luma_stream_u16,
            resample_outputs,
            rgb,
            rgba,
            rgb_u16,
            rgba_u16,
            luma,
            luma_u16,
            hsv,
            rgba_scratch,
            rgb_scratch,
            rgba_scratch_u16,
            rgba_color_scratch_u16,
            luma_scratch_u16,
            w,
            plan,
            idx,
            use_simd,
            alpha_mode,
            |dst| {
              rgba_444_dispatch(
                y_row, u_full, v_full, a_row, dst, w, matrix, full_range, use_simd, BE,
              )
            },
            |dst| {
              rgba_u16_444_dispatch(
                y_row, u_full, v_full, a_row, dst, w, matrix, full_range, use_simd, BE,
              )
            },
            |dst| deinterleave_y_high_bit_masked::<BITS, BE>(y_row, dst, w),
          )?;
        }
        stage_420_chroma_prev_u16(
          chroma_prev_u16,
          chroma_prev_row,
          u_half_row,
          v_half_row,
          idx,
          w,
        );
      }
      crate::resample::SpanKind::Filter => {
        // Straight-alpha single-kernel filter (premult excluded by the guard).
        plan.ensure_single_kernel_filter()?;
        let cursor = if rgb.is_some() || rgba.is_some() || hsv.is_some() {
          rgba_filter_stream.as_ref().map_or(0, |s| s.next_y())
        } else {
          rgba_filter_stream_u16.as_ref().map_or(0, |s| s.next_y())
        };
        let expected_source = cursor + usize::from(chroma_top_pending.is_some());
        if let core::ops::ControlFlow::Break(()) = super::resample_preflight_check_only(
          resample_outputs,
          luma,
          luma_u16,
          rgb,
          rgba,
          rgb_u16,
          rgba_u16,
          &None,
          &None,
          &None,
          &None,
          &None,
          hsv,
          &None,
          Some(expected_source),
          idx,
        )? {
          return Ok(());
        }
        reserve_420_chroma_full_u16(chroma_full_u16, w, h)?;
        reserve_420_chroma_prev_u16(chroma_prev_u16, w, h)?;
        reserve_420_chroma_top_u16(chroma_top_y_u16, w, h)?;
        reserve_420_chroma_top_u16(chroma_top_a_u16, w, h)?;
        let is_last = idx + 1 == h;
        if idx & 1 == 0 {
          if let Some((p_idx, p_matrix, p_full_range)) = chroma_top_pending.take() {
            let (u_full, v_full) = upsample_420_chroma_sited_u16::<BITS>(
              chroma_full_u16,
              chroma_prev_u16,
              chroma_prev_row,
              u_half_row,
              v_half_row,
              idx,
              true,
              center_sited,
              false,
              w,
              BE,
              use_simd,
            );
            let held_y: &[u16] = &chroma_top_y_u16[..w];
            let held_a: &[u16] = &chroma_top_a_u16[..w];
            packed_yuva444_filter_resample::<BITS, false, false>(
              rgba_filter_stream,
              rgba_filter_stream_u16,
              &mut None,
              luma_filter_stream_u16,
              resample_outputs,
              rgb,
              rgba,
              rgb_u16,
              rgba_u16,
              luma,
              luma_u16,
              hsv,
              rgba_scratch,
              rgb_scratch,
              rgba_scratch_u16,
              rgba_color_scratch_u16,
              luma_scratch_u16,
              w,
              plan,
              p_idx,
              use_simd,
              &[],
              None,
              |dst| {
                rgba_444_dispatch(
                  held_y,
                  u_full,
                  v_full,
                  held_a,
                  dst,
                  w,
                  p_matrix,
                  p_full_range,
                  use_simd,
                  BE,
                )
              },
              |dst| {
                rgba_u16_444_dispatch(
                  held_y,
                  u_full,
                  v_full,
                  held_a,
                  dst,
                  w,
                  p_matrix,
                  p_full_range,
                  use_simd,
                  BE,
                )
              },
              |dst| deinterleave_y_high_bit_masked::<BITS, BE>(held_y, dst, w),
              |_dst: &mut [u8]| {},
            )?;
          }
          let (u_full, v_full) = upsample_420_chroma_sited_u16::<BITS>(
            chroma_full_u16,
            chroma_prev_u16,
            chroma_prev_row,
            u_half_row,
            v_half_row,
            idx,
            false,
            center_sited,
            false,
            w,
            BE,
            use_simd,
          );
          packed_yuva444_filter_resample::<BITS, false, false>(
            rgba_filter_stream,
            rgba_filter_stream_u16,
            &mut None,
            luma_filter_stream_u16,
            resample_outputs,
            rgb,
            rgba,
            rgb_u16,
            rgba_u16,
            luma,
            luma_u16,
            hsv,
            rgba_scratch,
            rgb_scratch,
            rgba_scratch_u16,
            rgba_color_scratch_u16,
            luma_scratch_u16,
            w,
            plan,
            idx,
            use_simd,
            &[],
            None,
            |dst| {
              rgba_444_dispatch(
                y_row, u_full, v_full, a_row, dst, w, matrix, full_range, use_simd, BE,
              )
            },
            |dst| {
              rgba_u16_444_dispatch(
                y_row, u_full, v_full, a_row, dst, w, matrix, full_range, use_simd, BE,
              )
            },
            |dst| deinterleave_y_high_bit_masked::<BITS, BE>(y_row, dst, w),
            |_dst: &mut [u8]| {},
          )?;
        } else if !is_last {
          chroma_top_y_u16[..w].copy_from_slice(&y_row[..w]);
          chroma_top_a_u16[..w].copy_from_slice(&a_row[..w]);
          *chroma_top_pending = Some((idx, matrix, full_range));
        } else {
          let (u_full, v_full) = upsample_420_chroma_sited_u16::<BITS>(
            chroma_full_u16,
            chroma_prev_u16,
            chroma_prev_row,
            u_half_row,
            v_half_row,
            idx,
            false,
            center_sited,
            false,
            w,
            BE,
            use_simd,
          );
          packed_yuva444_filter_resample::<BITS, false, false>(
            rgba_filter_stream,
            rgba_filter_stream_u16,
            &mut None,
            luma_filter_stream_u16,
            resample_outputs,
            rgb,
            rgba,
            rgb_u16,
            rgba_u16,
            luma,
            luma_u16,
            hsv,
            rgba_scratch,
            rgb_scratch,
            rgba_scratch_u16,
            rgba_color_scratch_u16,
            luma_scratch_u16,
            w,
            plan,
            idx,
            use_simd,
            &[],
            None,
            |dst| {
              rgba_444_dispatch(
                y_row, u_full, v_full, a_row, dst, w, matrix, full_range, use_simd, BE,
              )
            },
            |dst| {
              rgba_u16_444_dispatch(
                y_row, u_full, v_full, a_row, dst, w, matrix, full_range, use_simd, BE,
              )
            },
            |dst| deinterleave_y_high_bit_masked::<BITS, BE>(y_row, dst, w),
            |_dst: &mut [u8]| {},
          )?;
        }
        stage_420_chroma_prev_u16(
          chroma_prev_u16,
          chroma_prev_row,
          u_half_row,
          v_half_row,
          idx,
          w,
        );
      }
    }
    if need_output && frozen_chroma_centered.is_none() {
      *frozen_chroma_centered = Some(center_sited);
      *frozen_chroma_bottom_v = Some(bottom_v);
      *frozen_chroma_top_v = Some(top_v);
    }
    return Ok(());
  }

  // The span kind picks the engine (mirrors the 8-bit `Yuva420p` and packed
  // `Vuya`): `Area` bins (the alpha-aware tail — premultiplied colour binned
  // premultiplied then un-premultiplied) at native precision; `Filter` runs
  // the signed-coefficient filter on the same converted RGBA (straight alpha
  // only). Unlike the 8-bit planar sinks (whose native Y is `u8`), the
  // high-bit native Y is genuinely `u16`, so the filter route uses
  // `NATIVE_LUMA_U8 = false` — the same u16-luma branch `Vuya` / `Vuyx` use:
  // luma rides the `u16` filter stream over the de-interleaved native Y. The
  // filter tail clamps every sub-16-bit colour AND native-Y overshoot to the
  // source's native max `(1 << BITS) - 1` (a value no-op at `BITS = 16`),
  // matching the in-range area path. A premultiplied `Filter` plan has no
  // analogue (the engine cannot un-premultiply), so it is routed to the area
  // tail, which surfaces the typed `UnsupportedFilter`.
  let r = match plan.kind() {
    crate::resample::SpanKind::Area => {
      if (center_sited || bottom_v) && want_color {
        // Centered area: reconstruct full-width chroma and bin the 4:4:4-decoded
        // RGBA — but ONLY after the resample preflight (frozen-output +
        // sequence), so an out-of-sequence / rejected row is caught before the
        // chroma reservation (#180). `packed_yuva444_resample` re-runs the
        // idempotent preflight. The α plane is full-resolution and
        // siting-independent: it flows into the 4:4:4 converters UNCHANGED.
        let expected = if rgb.is_some() || rgba.is_some() || hsv.is_some() {
          rgba_stream.as_ref().map_or(0, |s| s.next_y())
        } else {
          rgba_stream_u16.as_ref().map_or(0, |s| s.next_y())
        };
        match super::resample_preflight_check_only(
          resample_outputs,
          luma,
          luma_u16,
          rgb,
          rgba,
          rgb_u16,
          rgba_u16,
          &None,
          &None,
          &None,
          &None,
          &None,
          hsv,
          &None,
          Some(expected),
          idx,
        )? {
          core::ops::ControlFlow::Break(()) => Ok(()),
          core::ops::ControlFlow::Continue(()) => {
            reserve_420_chroma_full_u16(chroma_full_u16, w, h)?;
            // RFC #238 S6f: the `Bottom` even output row box-blends its chroma with
            // the previous chroma row, so reserve the lookback up front — AFTER the
            // preflight, BEFORE the commit (#180).
            if bottom_v {
              reserve_420_chroma_prev_u16(chroma_prev_u16, w, h)?;
            }
            // `stage = false`: DEFER the lookback advance until AFTER the fallible
            // packed commit accepts the row (below), so a rejected row leaves the
            // predecessor for a clean retry (#180 state-atomicity).
            let (u_full, v_full) = upsample_420_chroma_sited_u16::<BITS>(
              chroma_full_u16,
              chroma_prev_u16,
              chroma_prev_row,
              u_half_row,
              v_half_row,
              idx,
              bottom_v,
              center_sited,
              false,
              w,
              BE,
              use_simd,
            );
            let r = packed_yuva444_resample::<BITS>(
              rgba_stream,
              rgba_stream_u16,
              luma_stream_u16,
              resample_outputs,
              rgb,
              rgba,
              rgb_u16,
              rgba_u16,
              luma,
              luma_u16,
              hsv,
              rgba_scratch,
              rgb_scratch,
              rgba_scratch_u16,
              rgba_color_scratch_u16,
              luma_scratch_u16,
              w,
              plan,
              idx,
              use_simd,
              alpha_mode,
              |dst| {
                rgba_444_dispatch(
                  y_row, u_full, v_full, a_row, dst, w, matrix, full_range, use_simd, BE,
                )
              },
              |dst| {
                rgba_u16_444_dispatch(
                  y_row, u_full, v_full, a_row, dst, w, matrix, full_range, use_simd, BE,
                )
              },
              |dst| deinterleave_y_high_bit_masked::<BITS, BE>(y_row, dst, w),
            );
            // Bottom lookback: advance ONLY after the packed area resample accepts
            // the row, so a rejected row leaves the predecessor for a clean retry —
            // the sited reconstruction read it above but did not stage (deferred).
            if r.is_ok() && bottom_v {
              stage_420_chroma_prev_u16(
                chroma_prev_u16,
                chroma_prev_row,
                u_half_row,
                v_half_row,
                idx,
                w,
              );
            }
            r
          }
        }
      } else {
        packed_yuva444_resample::<BITS>(
          rgba_stream,
          rgba_stream_u16,
          luma_stream_u16,
          resample_outputs,
          rgb,
          rgba,
          rgb_u16,
          rgba_u16,
          luma,
          luma_u16,
          hsv,
          rgba_scratch,
          rgb_scratch,
          rgba_scratch_u16,
          rgba_color_scratch_u16,
          luma_scratch_u16,
          w,
          plan,
          idx,
          use_simd,
          alpha_mode,
          |dst| {
            rgba_dispatch(
              y_row, u_half_row, v_half_row, a_row, dst, w, matrix, full_range, use_simd, BE,
            )
          },
          |dst| {
            rgba_u16_dispatch(
              y_row, u_half_row, v_half_row, a_row, dst, w, matrix, full_range, use_simd, BE,
            )
          },
          |dst| deinterleave_y_high_bit_masked::<BITS, BE>(y_row, dst, w),
        )
      }
    }
    crate::resample::SpanKind::Filter if alpha_mode.is_premultiplied() => {
      // Premultiplied + filter has no analogue: route to the area tail with the
      // filter plan so it returns the typed `UnsupportedFilter`. Siting is
      // irrelevant — the row is rejected before any chroma reconstruction, so it
      // stays on the co-sited kernels (no wasted reservation, #180).
      packed_yuva444_resample::<BITS>(
        rgba_stream,
        rgba_stream_u16,
        luma_stream_u16,
        resample_outputs,
        rgb,
        rgba,
        rgb_u16,
        rgba_u16,
        luma,
        luma_u16,
        hsv,
        rgba_scratch,
        rgb_scratch,
        rgba_scratch_u16,
        rgba_color_scratch_u16,
        luma_scratch_u16,
        w,
        plan,
        idx,
        use_simd,
        alpha_mode,
        |dst| {
          rgba_dispatch(
            y_row, u_half_row, v_half_row, a_row, dst, w, matrix, full_range, use_simd, BE,
          )
        },
        |dst| {
          rgba_u16_dispatch(
            y_row, u_half_row, v_half_row, a_row, dst, w, matrix, full_range, use_simd, BE,
          )
        },
        |dst| deinterleave_y_high_bit_masked::<BITS, BE>(y_row, dst, w),
      )
    }
    crate::resample::SpanKind::Filter => {
      // Reject a multi-kernel (BICUBLIN) filter plan BEFORE the centered
      // preflight/reserve below: `packed_yuva444_filter_resample`'s first act is
      // this same check, so without hoisting it a BICUBLIN plan would freeze the
      // output set and reserve/reconstruct chroma before being rejected as
      // `UnsupportedFilter` (#180 reject-before-allocation). Idempotent — the
      // delegate re-runs it.
      plan.ensure_single_kernel_filter()?;
      if (center_sited || bottom_v) && want_color {
        // Centered filter: reconstruct full-width chroma and filter the
        // 4:4:4-decoded RGBA — but ONLY after the filter preflight (frozen-output
        // + sequence), so an out-of-sequence / rejected row is caught before the
        // chroma reservation (#180). The α plane is siting-independent and flows
        // into the 4:4:4 converters UNCHANGED.
        let expected = if rgb.is_some() || rgba.is_some() || hsv.is_some() {
          rgba_filter_stream.as_ref().map_or(0, |s| s.next_y())
        } else {
          rgba_filter_stream_u16.as_ref().map_or(0, |s| s.next_y())
        };
        match super::resample_preflight_check_only(
          resample_outputs,
          luma,
          luma_u16,
          rgb,
          rgba,
          rgb_u16,
          rgba_u16,
          &None,
          &None,
          &None,
          &None,
          &None,
          hsv,
          &None,
          Some(expected),
          idx,
        )? {
          core::ops::ControlFlow::Break(()) => Ok(()),
          core::ops::ControlFlow::Continue(()) => {
            reserve_420_chroma_full_u16(chroma_full_u16, w, h)?;
            // RFC #238 S6f: reserve the bottom-sited lookback the even output row
            // box-blends against — AFTER the preflight, BEFORE the commit (#180).
            if bottom_v {
              reserve_420_chroma_prev_u16(chroma_prev_u16, w, h)?;
            }
            // `stage = false`: DEFER the lookback advance until AFTER the fallible
            // filter commit accepts the row (below), so a rejected row leaves the
            // predecessor for a clean retry (#180 state-atomicity).
            let (u_full, v_full) = upsample_420_chroma_sited_u16::<BITS>(
              chroma_full_u16,
              chroma_prev_u16,
              chroma_prev_row,
              u_half_row,
              v_half_row,
              idx,
              bottom_v,
              center_sited,
              false,
              w,
              BE,
              use_simd,
            );
            let r = packed_yuva444_filter_resample::<BITS, false, false>(
              rgba_filter_stream,
              rgba_filter_stream_u16,
              &mut None,
              luma_filter_stream_u16,
              resample_outputs,
              rgb,
              rgba,
              rgb_u16,
              rgba_u16,
              luma,
              luma_u16,
              hsv,
              rgba_scratch,
              rgb_scratch,
              rgba_scratch_u16,
              rgba_color_scratch_u16,
              luma_scratch_u16,
              w,
              plan,
              idx,
              use_simd,
              &[],
              None,
              |dst| {
                rgba_444_dispatch(
                  y_row, u_full, v_full, a_row, dst, w, matrix, full_range, use_simd, BE,
                )
              },
              |dst| {
                rgba_u16_444_dispatch(
                  y_row, u_full, v_full, a_row, dst, w, matrix, full_range, use_simd, BE,
                )
              },
              |dst| deinterleave_y_high_bit_masked::<BITS, BE>(y_row, dst, w),
              |_dst: &mut [u8]| {},
            );
            // Bottom lookback: advance ONLY after the filter resample accepts the
            // row, so a rejected row leaves the predecessor for a clean retry — the
            // sited reconstruction read it above but did not stage (deferred, #180).
            if r.is_ok() && bottom_v {
              stage_420_chroma_prev_u16(
                chroma_prev_u16,
                chroma_prev_row,
                u_half_row,
                v_half_row,
                idx,
                w,
              );
            }
            r
          }
        }
      } else {
        packed_yuva444_filter_resample::<BITS, false, false>(
          rgba_filter_stream,
          rgba_filter_stream_u16,
          // High-bit planar YUVA never uses the u8 native-Y luma stream
          // (`NATIVE_LUMA_U8 = false`); pass an inert slot.
          &mut None,
          luma_filter_stream_u16,
          resample_outputs,
          rgb,
          rgba,
          rgb_u16,
          rgba_u16,
          luma,
          luma_u16,
          hsv,
          rgba_scratch,
          rgb_scratch,
          rgba_scratch_u16,
          rgba_color_scratch_u16,
          luma_scratch_u16,
          w,
          plan,
          idx,
          use_simd,
          // Luma rides `deinterleave_y` + the u16 stream (native Y is u16), so the
          // u8-luma input is unused.
          &[],
          None,
          |dst| {
            rgba_dispatch(
              y_row, u_half_row, v_half_row, a_row, dst, w, matrix, full_range, use_simd, BE,
            )
          },
          |dst| {
            rgba_u16_dispatch(
              y_row, u_half_row, v_half_row, a_row, dst, w, matrix, full_range, use_simd, BE,
            )
          },
          |dst| deinterleave_y_high_bit_masked::<BITS, BE>(y_row, dst, w),
          |_dst: &mut [u8]| {},
        )
      }
    }
  };
  // RFC #238 S6c: freeze the horizontal 4:2:0 chroma siting on the first
  // ACCEPTED output-bearing resampled row (a rejected row returns Err above and
  // leaves it unset for a corrected retry); S6f / Top freeze the vertical `Bottom` /
  // `Top` phase alongside it (the Top forward-delay path returned earlier with its
  // own freeze). Both packed tails carry one plan kind for the frame, so there is
  // no native/row-stage route to co-freeze here.
  if r.is_ok() && need_output && frozen_chroma_centered.is_none() {
    *frozen_chroma_centered = Some(center_sited);
    *frozen_chroma_bottom_v = Some(bottom_v);
    *frozen_chroma_top_v = Some(top_v);
  }
  r
}