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
//! Chroma-siting-aware 4:2:2 upsampling for the 8-bit PACKED `Yuyv422` /
//! `Uyvy422` / `Yvyu422` — the packed sibling of `chroma_siting_422` (planar
//! `Yuv422p`) and `chroma_siting_nv16` (semi-planar `Nv16`).
//!
//! A packed 4:2:2 row carries chroma 2:1 inline (`[Y0 U Y1 V]` for YUYV,
//! `[U Y0 V Y1]` for UYVY, `[Y0 V Y1 U]` for YVYU), so the centered
//! reconstruction de-interleaves the Y bytes to a full-width plane + the U / V
//! bytes to half-width planes, then reuses the SAME phase-0.5 horizontal
//! upsample + 4:4:4 decode the planar / semi-planar twins use. Covers, for ALL
//! layouts: the default / co-sited path
//! staying byte-identical to the pre-#302 fused nearest-neighbor decode (the
//! regression guard, negative-controlled by a ramp that makes the phase
//! observable); the centered RGB / RGBA / HSV decodes matching an independent
//! upsample-then-4:4:4 reference; cross-format equivalence to the planar
//! `Yuv422p` centered decode (the strongest catch for a Y / U / V de-interleave
//! bug); SIMD-vs-scalar parity; the `ColorSpec` siting flow; the
//! `ChromaDerivedNcl` matrix-tag-fallback consistency; the preflight-ordering
//! alloc-failure atomicity (luma untouched); and the no-output invariant (no row
//! math / no allocation when nothing is attached, including a
//! 32-bit-offset-overflow geometry).

#[allow(unused_imports)]
pub(super) use super::*;
pub(super) use crate::ChromaLocation;
use crate::ColorMatrix;

pub(super) const W: u32 = 16;
pub(super) const H: u32 = 8;

/// Non-flat luma + per-column chroma ramp on a half-width, **full-height**
/// chroma plane (one chroma row per luma row — 4:2:2). The Y ramp makes a Y
/// de-interleave bug observable; distinct adjacent chroma columns make the
/// horizontal phase observable; the `+ r` term keeps rows distinct so a vertical
/// mistake would surface.
pub(super) fn ramp_planes() -> (Vec<u8>, Vec<u8>, Vec<u8>) {
  let w = W as usize;
  let h = H as usize;
  let cw = w / 2;
  let mut y = std::vec![0u8; w * h];
  for r in 0..h {
    for x in 0..w {
      y[r * w + x] = (16 + x * 8 + r * 5).min(235) as u8;
    }
  }
  let mut u = std::vec![0u8; cw * h];
  let mut v = std::vec![0u8; cw * h];
  for r in 0..h {
    for c in 0..cw {
      u[r * cw + c] = (16 + c * 24 + r * 3).min(240) as u8;
      v[r * cw + c] = (240 - c * 24).max(16) as u8;
    }
  }
  (y, u, v)
}

/// Packs planar Y (full) + half-width U / V into a YUYV422 plane (`[Y0 U Y1 V]`).
pub(super) fn pack_yuyv(y: &[u8], u: &[u8], v: &[u8]) -> Vec<u8> {
  let w = W as usize;
  let h = H as usize;
  let cw = w / 2;
  let mut out = std::vec![0u8; 2 * w * h];
  for r in 0..h {
    for c in 0..cw {
      let base = r * 2 * w + c * 4;
      out[base] = y[r * w + 2 * c];
      out[base + 1] = u[r * cw + c];
      out[base + 2] = y[r * w + 2 * c + 1];
      out[base + 3] = v[r * cw + c];
    }
  }
  out
}

/// Packs planar Y (full) + half-width U / V into a UYVY422 plane (`[U Y0 V Y1]`).
pub(super) fn pack_uyvy(y: &[u8], u: &[u8], v: &[u8]) -> Vec<u8> {
  let w = W as usize;
  let h = H as usize;
  let cw = w / 2;
  let mut out = std::vec![0u8; 2 * w * h];
  for r in 0..h {
    for c in 0..cw {
      let base = r * 2 * w + c * 4;
      out[base] = u[r * cw + c];
      out[base + 1] = y[r * w + 2 * c];
      out[base + 2] = v[r * cw + c];
      out[base + 3] = y[r * w + 2 * c + 1];
    }
  }
  out
}

/// Packs planar Y (full) + half-width U / V into a YVYU422 plane (`[Y0 V Y1 U]`).
pub(super) fn pack_yvyu(y: &[u8], u: &[u8], v: &[u8]) -> Vec<u8> {
  let w = W as usize;
  let h = H as usize;
  let cw = w / 2;
  let mut out = std::vec![0u8; 2 * w * h];
  for r in 0..h {
    for c in 0..cw {
      let base = r * 2 * w + c * 4;
      out[base] = y[r * w + 2 * c];
      out[base + 1] = v[r * cw + c];
      out[base + 2] = y[r * w + 2 * c + 1];
      out[base + 3] = u[r * cw + c];
    }
  }
  out
}

/// Independent reference for the centered-siting horizontal upsample — the
/// MPEG-1 / JPEG phase-0.5 `1/4`–`3/4` weights with edge clamp. Written
/// separately from the production kernel so it is a real oracle.
pub(super) fn ref_upsample_center_h(c_half: &[u8], width: usize) -> Vec<u8> {
  let half = width / 2;
  let mut out = std::vec![0u8; width];
  for j in 0..half {
    let l = c_half[j.saturating_sub(1)] as i32;
    let m = c_half[j] as i32;
    let r = c_half[if j + 1 < half { j + 1 } else { j }] as i32;
    out[2 * j] = ((l + 3 * m + 2) >> 2) as u8;
    out[2 * j + 1] = ((3 * m + r + 2) >> 2) as u8;
  }
  out
}

/// Full-resolution U / V planes a centered-siting 4:2:2 decode reconstructs:
/// each luma row `r` takes chroma row `r` (1:1 — NO vertical subsampling)
/// upsampled with the centered weights. Feeding these to a `Yuv444p` conversion
/// is the end-to-end oracle.
pub(super) fn ref_full_chroma(u422: &[u8], v422: &[u8]) -> (Vec<u8>, Vec<u8>) {
  let w = W as usize;
  let h = H as usize;
  let cw = w / 2;
  let mut u444 = std::vec![0u8; w * h];
  let mut v444 = std::vec![0u8; w * h];
  for r in 0..h {
    let urow = ref_upsample_center_h(&u422[r * cw..r * cw + cw], w);
    let vrow = ref_upsample_center_h(&v422[r * cw..r * cw + cw], w);
    u444[r * w..r * w + w].copy_from_slice(&urow);
    v444[r * w..r * w + w].copy_from_slice(&vrow);
  }
  (u444, v444)
}

/// Generates the full centered-siting suite for one packed 4:2:2 layout. Both
/// `Yuyv422` and `Uyvy422` get the identical coverage; only the source marker /
/// row / frame / walker and the planar→packed `$pack` differ.
macro_rules! packed_siting_suite {
  ($modname:ident, $Marker:ident, $Row:ident, $Frame:ident, $walker:ident, $pack:ident) => {
    mod $modname {
      use super::*;

      /// Centered identity-decode RGB for a siting + SIMD toggle.
      fn convert_rgb(loc: ChromaLocation, simd: bool) -> Vec<u8> {
        let (yp, up, vp) = ramp_planes();
        let packed = $pack(&yp, &up, &vp);
        let src = $Frame::new(&packed, W, H, W * 2);
        let mut rgb = std::vec![0u8; (W * H * 3) as usize];
        let mut sink = MixedSinker::<$Marker>::new(W as usize, H as usize)
          .with_rgb(&mut rgb)
          .unwrap()
          .with_chroma_location(loc.clone())
          .with_simd(simd);
        $walker(&src, false, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
        rgb
      }

      // ---- default / co-sited path is byte-identical (regression guard) ----

      #[test]
      #[cfg_attr(
        miri,
        ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
      )]
      fn default_and_cosited_sitings_are_byte_identical() {
        let baseline = convert_rgb(ChromaLocation::Unspecified, true);
        for loc in [
          ChromaLocation::Unspecified,
          ChromaLocation::other("unassigned-99"),
          ChromaLocation::Left,
          ChromaLocation::TopLeft,
          ChromaLocation::BottomLeft,
        ] {
          assert_eq!(
            convert_rgb(loc.clone(), true),
            baseline,
            "siting {loc:?} must keep the byte-identical default decode"
          );
        }
      }

      #[test]
      #[cfg_attr(
        miri,
        ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
      )]
      fn default_path_does_not_allocate_chroma_scratch() {
        let (yp, up, vp) = ramp_planes();
        let packed = $pack(&yp, &up, &vp);
        let src = $Frame::new(&packed, W, H, W * 2);
        let mut rgb = std::vec![0u8; (W * H * 3) as usize];
        let mut sink = MixedSinker::<$Marker>::new(W as usize, H as usize)
          .with_rgb(&mut rgb)
          .unwrap()
          .with_chroma_location(ChromaLocation::Left);
        $walker(&src, false, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
        let chroma_len = sink.chroma_full.len();
        let half_len = sink.packed_yuv_u_half.len();
        let y_len = sink.packed_yuv_y_full.len();
        drop(sink);
        assert_eq!(
          chroma_len, 0,
          "co-sited path must not grow the chroma scratch"
        );
        assert_eq!(
          half_len, 0,
          "co-sited path must not grow the de-interleave scratch"
        );
        assert_eq!(
          y_len, 0,
          "co-sited path must not grow the Y de-interleave scratch"
        );
      }

      // ---- centered path correctness ----

      #[test]
      #[cfg_attr(
        miri,
        ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
      )]
      fn center_rgb_matches_upsample_then_444_reference() {
        let (yp, up, vp) = ramp_planes();
        let (u444, v444) = ref_full_chroma(&up, &vp);
        let ref_src = Yuv444pFrame::new(&yp, &u444, &v444, W, H, W, W, W);
        let mut rgb_ref = std::vec![0u8; (W * H * 3) as usize];
        let mut ref_sink = MixedSinker::<Yuv444p>::new(W as usize, H as usize)
          .with_rgb(&mut rgb_ref)
          .unwrap();
        yuv444p_to(&ref_src, false, ref_sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
        assert_eq!(
          convert_rgb(ChromaLocation::Center, true),
          rgb_ref,
          "centered packed RGB must equal upsample-then-4:4:4"
        );
      }

      #[test]
      #[cfg_attr(
        miri,
        ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
      )]
      fn center_grows_chroma_scratch_to_full_width() {
        let (yp, up, vp) = ramp_planes();
        let packed = $pack(&yp, &up, &vp);
        let src = $Frame::new(&packed, W, H, W * 2);
        let mut rgb = std::vec![0u8; (W * H * 3) as usize];
        let mut sink = MixedSinker::<$Marker>::new(W as usize, H as usize)
          .with_rgb(&mut rgb)
          .unwrap()
          .with_chroma_location(ChromaLocation::Center);
        $walker(&src, false, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
        let chroma_len = sink.chroma_full.len();
        let half_len = sink.packed_yuv_u_half.len();
        let y_len = sink.packed_yuv_y_full.len();
        drop(sink);
        assert_eq!(
          chroma_len,
          2 * W as usize,
          "centered path stages U+V at full width"
        );
        assert_eq!(
          half_len,
          (W / 2) as usize,
          "centered path stages the de-interleaved half-width chroma"
        );
        assert_eq!(
          y_len,
          W as usize,
          "centered path stages the de-interleaved full-width Y"
        );
      }

      #[test]
      #[cfg_attr(
        miri,
        ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
      )]
      fn top_and_bottom_route_like_center_horizontally() {
        // Top / Bottom share Center's horizontal (centered) phase; 4:2:2 has no
        // vertical axis to drive, so all three agree.
        let center = convert_rgb(ChromaLocation::Center, true);
        assert_eq!(convert_rgb(ChromaLocation::Top, true), center);
        assert_eq!(convert_rgb(ChromaLocation::Bottom, true), center);
      }

      #[test]
      #[cfg_attr(
        miri,
        ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
      )]
      fn centered_phase_differs_from_default() {
        // Negative control for the byte-identity guard: on a chroma ramp the
        // centered phase must move chroma relative to the co-sited default.
        assert_ne!(
          convert_rgb(ChromaLocation::Center, true),
          convert_rgb(ChromaLocation::Left, true),
          "centered siting must shift chroma vs the co-sited default"
        );
      }

      #[test]
      #[cfg_attr(
        miri,
        ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
      )]
      fn centered_path_simd_matches_scalar() {
        assert_eq!(
          convert_rgb(ChromaLocation::Center, true),
          convert_rgb(ChromaLocation::Center, false),
          "centered path must be bit-identical across the SIMD and scalar tiers"
        );
      }

      #[test]
      #[cfg_attr(
        miri,
        ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
      )]
      fn centered_matches_yuv422p_centered() {
        // De-interleaving the packed row back to planar Y / U / V must
        // reconstruct the SAME planes the planar twin holds, so a centered packed
        // decode is byte-identical to a centered Yuv422p decode — the strongest
        // catch for a Y / U / V swap in the de-interleave. Bt601: the equivalence
        // holds on the shared matrix-tag path (`ChromaDerivedNcl` is separate).
        let (yp, up, vp) = ramp_planes();
        let src422 = Yuv422pFrame::new(&yp, &up, &vp, W, H, W, W / 2, W / 2);
        let mut rgb422 = std::vec![0u8; (W * H * 3) as usize];
        let mut sink422 = MixedSinker::<Yuv422p>::new(W as usize, H as usize)
          .with_rgb(&mut rgb422)
          .unwrap()
          .with_chroma_location(ChromaLocation::Center);
        yuv422p_to(&src422, false, sink422.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
        assert_eq!(
          convert_rgb(ChromaLocation::Center, true),
          rgb422,
          "centered packed must equal centered Yuv422p of the same planes"
        );
      }

      // ---- ChromaDerivedNcl: packed uses the matrix-tag fallback ----

      #[test]
      #[cfg_attr(
        miri,
        ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
      )]
      fn centered_chroma_derived_ncl_uses_matrix_tag_fallback() {
        // Packed is NOT ChromaDerivedNcl-primaries-wired. BOTH paths (the default
        // fused kernel AND the centered 4:4:4 kernel) resolve ChromaDerivedNcl via
        // the shared BT.709 matrix-tag fallback (`Coefficients::for_matrix`),
        // IGNORING the ColorSpec primaries — so default and centered stay
        // internally consistent (the centered phase shift is the ONLY difference).
        use crate::{ColorInfo, ColorSpec, DynamicRange, PixelFormat, Primaries, Transfer};

        let (yp, up, vp) = ramp_planes();
        let packed = $pack(&yp, &up, &vp);
        let spec = |loc: ChromaLocation| {
          ColorSpec::from_info(
            PixelFormat::Yuv422p,
            ColorInfo::new(
              Primaries::Bt2020,
              Transfer::Bt709,
              ColorMatrix::ChromaDerivedNcl,
              DynamicRange::Limited,
              loc,
            ),
          )
        };
        let decode_cdn = |loc: ChromaLocation| -> Vec<u8> {
          let src = $Frame::new(&packed, W, H, W * 2);
          let mut rgb = std::vec![0u8; (W * H * 3) as usize];
          let mut sink = MixedSinker::<$Marker>::new(W as usize, H as usize)
            .with_rgb(&mut rgb)
            .unwrap()
            .with_color_spec(&spec(loc)).unwrap();
          $walker(&src, false, sink.set_kernel_matrix(KernelMatrix::ChromaDerivedNcl)).unwrap();
          rgb
        };
        let decode_bt709 = |loc: ChromaLocation| -> Vec<u8> {
          let src = $Frame::new(&packed, W, H, W * 2);
          let mut rgb = std::vec![0u8; (W * H * 3) as usize];
          let mut sink = MixedSinker::<$Marker>::new(W as usize, H as usize)
            .with_rgb(&mut rgb)
            .unwrap()
            .with_chroma_location(loc.clone());
          $walker(&src, false, sink.set_kernel_matrix(KernelMatrix::Bt709)).unwrap();
          rgb
        };

        assert_eq!(
          decode_cdn(ChromaLocation::Center),
          decode_bt709(ChromaLocation::Center),
          "centered ChromaDerivedNcl must resolve via the BT.709 matrix-tag fallback"
        );
        assert_eq!(
          decode_cdn(ChromaLocation::Left),
          decode_bt709(ChromaLocation::Left),
          "default ChromaDerivedNcl must resolve via the same matrix-tag fallback"
        );
        assert_ne!(
          decode_cdn(ChromaLocation::Center),
          decode_cdn(ChromaLocation::Left),
          "centered ChromaDerivedNcl must still shift chroma vs the co-sited default"
        );
      }

      // ---- RGBA / HSV identity outputs also honor siting ----

      #[test]
      #[cfg_attr(
        miri,
        ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
      )]
      fn center_rgba_and_hsv_match_444_reference() {
        let (yp, up, vp) = ramp_planes();
        let (u444, v444) = ref_full_chroma(&up, &vp);
        let packed = $pack(&yp, &up, &vp);

        // RGBA-only path.
        {
          let src = $Frame::new(&packed, W, H, W * 2);
          let mut rgba = std::vec![0u8; (W * H * 4) as usize];
          let mut sink = MixedSinker::<$Marker>::new(W as usize, H as usize)
            .with_rgba(&mut rgba)
            .unwrap()
            .with_chroma_location(ChromaLocation::Center);
          $walker(&src, false, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();

          let ref_src = Yuv444pFrame::new(&yp, &u444, &v444, W, H, W, W, W);
          let mut rgba_ref = std::vec![0u8; (W * H * 4) as usize];
          let mut ref_sink = MixedSinker::<Yuv444p>::new(W as usize, H as usize)
            .with_rgba(&mut rgba_ref)
            .unwrap();
          yuv444p_to(&ref_src, false, ref_sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
          assert_eq!(rgba, rgba_ref, "centered RGBA must equal upsample-then-4:4:4");
        }

        // HSV-direct path (no RGB / RGBA attached).
        {
          let src = $Frame::new(&packed, W, H, W * 2);
          let (mut h, mut s, mut v) = (
            std::vec![0u8; (W * H) as usize],
            std::vec![0u8; (W * H) as usize],
            std::vec![0u8; (W * H) as usize],
          );
          let mut sink = MixedSinker::<$Marker>::new(W as usize, H as usize)
            .with_hsv(&mut h, &mut s, &mut v)
            .unwrap()
            .with_chroma_location(ChromaLocation::Center);
          $walker(&src, false, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();

          let ref_src = Yuv444pFrame::new(&yp, &u444, &v444, W, H, W, W, W);
          let (mut hr, mut sr, mut vr) = (
            std::vec![0u8; (W * H) as usize],
            std::vec![0u8; (W * H) as usize],
            std::vec![0u8; (W * H) as usize],
          );
          let mut ref_sink = MixedSinker::<Yuv444p>::new(W as usize, H as usize)
            .with_hsv(&mut hr, &mut sr, &mut vr)
            .unwrap();
          yuv444p_to(&ref_src, false, ref_sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
          assert_eq!(
            (h, s, v),
            (hr, sr, vr),
            "centered HSV must equal upsample-then-4:4:4"
          );
        }
      }

      // ---- end-to-end ColorSpec flow (no manual with_chroma_location) ----

      #[test]
      #[cfg_attr(
        miri,
        ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
      )]
      fn color_spec_center_drives_decode_without_manual_chroma_call() {
        use crate::{
          ColorInfo, ColorSpec, DynamicRange, PixelFormat, Primaries, Transfer, YuvOptions,
        };

        let (yp, up, vp) = ramp_planes();
        let packed = $pack(&yp, &up, &vp);
        let src = $Frame::new(&packed, W, H, W * 2);

        let info = ColorInfo::new(
          Primaries::Bt709,
          Transfer::Bt709,
          ColorMatrix::Bt601,
          DynamicRange::Limited,
          ChromaLocation::Center,
        );
        let spec = ColorSpec::from_info(PixelFormat::Yuv422p, info);
        let opts = YuvOptions::from_color_spec(&spec);
        let mut rgb = std::vec![0u8; (W * H * 3) as usize];
        let mut sink = MixedSinker::<$Marker>::new(W as usize, H as usize)
          .with_rgb(&mut rgb)
          .unwrap()
          .with_color_spec(&spec).unwrap();
        $walker(&src, opts.full_range(), &mut sink).unwrap();
        drop(sink);

        assert_ne!(
          rgb,
          convert_rgb(ChromaLocation::Unspecified, true),
          "ColorSpec ChromaLocation::Center must change the decode via the options path"
        );
        assert_eq!(
          rgb,
          convert_rgb(ChromaLocation::Center, true),
          "ColorSpec-driven centered decode must equal the explicit centered path"
        );
      }

      // ---- preflight-ordering atomicity (#302, cf. #180 / #308) ----

      #[test]
      #[cfg_attr(
        miri,
        ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
      )]
      fn centered_alloc_failure_leaves_outputs_untouched() {
        use crate::resample::ResampleError;

        let (yp, up, vp) = ramp_planes();
        let packed = $pack(&yp, &up, &vp);

        // Negative control: unarmed, the SAME luma + centered-RGB config DOES
        // write luma — so the armed "untouched" assertion below is non-vacuous.
        {
          let src = $Frame::new(&packed, W, H, W * 2);
          let mut luma_ok = std::vec![0xABu8; (W * H) as usize];
          let mut rgb_ok = std::vec![0xCDu8; (W * H * 3) as usize];
          let mut sink_ok = MixedSinker::<$Marker>::new(W as usize, H as usize)
            .with_luma(&mut luma_ok)
            .unwrap()
            .with_rgb(&mut rgb_ok)
            .unwrap()
            .with_chroma_location(ChromaLocation::Center);
          $walker(&src, false, sink_ok.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
          drop(sink_ok);
          assert!(
            luma_ok.iter().any(|&b| b != 0xAB),
            "control: the centered path writes luma when the scratch alloc is not armed"
          );
        }

        // Armed: a centered RGB decode whose chroma-scratch allocation fails must
        // leave EVERY output — luma included — untouched, because the scratch is
        // reserved (fallibly) BEFORE any output row is written.
        let src = $Frame::new(&packed, W, H, W * 2);
        let mut luma = std::vec![0xABu8; (W * H) as usize];
        let mut rgb = std::vec![0xCDu8; (W * H * 3) as usize];
        let mut sink = MixedSinker::<$Marker>::new(W as usize, H as usize)
          .with_luma(&mut luma)
          .unwrap()
          .with_rgb(&mut rgb)
          .unwrap()
          .with_chroma_location(ChromaLocation::Center);

        crate::sinker::mixed::arm_chroma_full_alloc_failure();
        let err = $walker(&src, false, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap_err();
        drop(sink);

        assert!(
          matches!(
            err,
            MixedSinkerError::Resample(ResampleError::AllocationFailed(_))
          ),
          "centered chroma-scratch refusal must surface as a recoverable AllocationFailed, got {err:?}"
        );
        assert!(
          luma.iter().all(|&b| b == 0xAB),
          "luma must be untouched on the centered alloc-failure path"
        );
        assert!(
          rgb.iter().all(|&b| b == 0xCD),
          "rgb must be untouched on the centered alloc-failure path"
        );
      }

      // ---- no-output invariant: guard runs before the row-offset math ----

      #[test]
      #[cfg_attr(
        miri,
        ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
      )]
      fn no_output_row_does_not_allocate_chroma_scratch() {
        // A centered-siting sink with NO outputs attached returns before the
        // centered preflight, so neither chroma scratch is ever reserved.
        let (yp, up, vp) = ramp_planes();
        let packed = $pack(&yp, &up, &vp);
        let src = $Frame::new(&packed, W, H, W * 2);
        let mut sink = MixedSinker::<$Marker>::new(W as usize, H as usize)
          .with_chroma_location(ChromaLocation::Center);
        $walker(&src, false, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
        let chroma_len = sink.chroma_full.len();
        let half_len = sink.packed_yuv_u_half.len();
        drop(sink);
        assert_eq!(
          chroma_len, 0,
          "a no-output centered row must not reserve the chroma scratch"
        );
        assert_eq!(
          half_len, 0,
          "a no-output centered row must not reserve the de-interleave scratch"
        );
      }

      #[test]
      #[cfg_attr(
        miri,
        ignore = "constructs an absurd geometry; the no-op contract is the point, not Miri"
      )]
      fn no_output_row_large_geometry_does_not_overflow() {
        // The no-output guard must run BEFORE the `idx * w` single-plane offset
        // arithmetic. A no-output `process` call never ran an attach-time
        // `w x h x 1` validation, so on a 32-bit target (`usize == u32`) an absurd
        // geometry where `idx * w` exceeds `u32::MAX` would overflow that offset
        // and panic under overflow checks. With no outputs attached, `process`
        // must return `Ok(())` having done NO row math and NO allocation.
        //
        // w = 4, idx = 2^30 -> idx * w = 2^32 = u32::MAX + 1 (overflows u32).
        let w: usize = 4;
        let idx: usize = 1 << 30;
        let h: usize = idx + 1; // idx < height so the row-index check passes
        assert!(
          (idx as u64) * (w as u64) > u32::MAX as u64,
          "test geometry must exceed u32::MAX to exercise the 32-bit offset overflow"
        );

        let packed = std::vec![128u8; 2 * w]; // one packed row: 2 * w bytes
        let mut sink =
          MixedSinker::<$Marker>::new(w, h).with_chroma_location(ChromaLocation::Center);
        let row = crate::source::$Row::for_tests(&packed, idx, KernelMatrix::Bt601, false);
        crate::PixelSink::process(&mut sink, row).unwrap();
        let chroma_len = sink.chroma_full.len();
        drop(sink);
        assert_eq!(
          chroma_len, 0,
          "a no-output large-geometry row must allocate nothing"
        );
      }

      // ---- mid-frame siting-flip rejection (identity path freeze) ---------

      #[test]
      #[cfg_attr(
        miri,
        ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
      )]
      fn direct_path_mid_frame_siting_flip_is_rejected() {
        // The identity (no-resample) packed 4:2:2 decode freezes the horizontal
        // siting on its first output-bearing row, mirroring the planar / semi-planar
        // twins. Flipping co-sited ⇆ centered mid-frame must reject the next
        // in-sequence row with `ChromaSitingChanged` (leaving the de-interleave /
        // chroma scratch untouched); flipping back and retrying then matches a clean
        // single-phase decode byte-for-byte.
        let (yp, up, vp) = ramp_planes();
        let packed = $pack(&yp, &up, &vp);
        let w = W as usize;
        let h = H as usize;
        for (loc1, loc2) in [
          (ChromaLocation::Left, ChromaLocation::Center),
          (ChromaLocation::Center, ChromaLocation::Left),
        ] {
          let want = convert_rgb(loc1.clone(), true);
          let mut rgb = std::vec![0u8; w * h * 3];
          let mut sink = MixedSinker::<$Marker>::new(w, h)
            .with_rgb(&mut rgb)
            .unwrap()
            .with_chroma_location(loc1.clone())
            .with_simd(true);
          crate::PixelSink::begin_frame(&mut sink, W, H).unwrap();
          let row0 = crate::source::$Row::for_tests(&packed[0..2 * w], 0, KernelMatrix::Bt601, false);
          crate::PixelSink::process(&mut sink, row0).unwrap();
          let scratch_len = sink.chroma_full.len();

          sink.set_chroma_location(loc2.clone());
          let row1 = crate::source::$Row::for_tests(&packed[2 * w..4 * w], 1, KernelMatrix::Bt601, false);
          let err = crate::PixelSink::process(&mut sink, row1).unwrap_err();
          assert!(
            matches!(err, MixedSinkerError::ChromaSitingChanged(_)),
            "direct path {loc1:?}->{loc2:?}: want ChromaSitingChanged, got {err:?}"
          );
          assert_eq!(
            sink.chroma_full.len(),
            scratch_len,
            "{loc1:?}->{loc2:?}: a rejected flip must not grow the chroma scratch"
          );

          sink.set_chroma_location(loc1.clone());
          for r in 1..h {
            let row = crate::source::$Row::for_tests(
              &packed[r * 2 * w..(r + 1) * 2 * w],
              r,
              KernelMatrix::Bt601,
              false,
            );
            crate::PixelSink::process(&mut sink, row).unwrap();
          }
          drop(sink);
          assert_eq!(
            rgb, want,
            "{loc1:?}: retry after a rejected flip must match a clean in-order decode"
          );
        }
      }
    }
  };
}

packed_siting_suite!(
  yuyv,
  Yuyv422,
  Yuyv422Row,
  Yuyv422Frame,
  yuyv422_to,
  pack_yuyv
);
packed_siting_suite!(
  uyvy,
  Uyvy422,
  Uyvy422Row,
  Uyvy422Frame,
  uyvy422_to,
  pack_uyvy
);
packed_siting_suite!(
  yvyu,
  Yvyu422,
  Yvyu422Row,
  Yvyu422Frame,
  yvyu422_to,
  pack_yvyu
);