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
//! Tier 4 Y216 sinker tests — Ship 11d.
//!
//! Coverage matrix:
//! - Single-output paths (luma u8, luma u16, rgb, rgba, rgb_u16,
//!   rgba_u16) on solid-gray frames.
//! - Strategy A invariant (`with_rgb` + `with_rgba` byte-identical;
//!   same for the u16 variants).
//! - SIMD-vs-scalar parity across multiple widths covering the main
//!   loop + scalar tail of every backend block size.
//! - Three error-path tests: odd width, short packed slice, short
//!   luma_u16 buffer.

use super::*;

// ---- Solid-color Y216 builder -----------------------------------------

/// Builds a solid-color Y216 plane with one (Y, U, V) repeated. Each
/// row is `width x 2` u16 elements (`Y₀, U, Y₁, V` quadruples). All
/// samples are direct 16-bit values — Y216 uses the full u16 range
/// with no MSB-alignment shift (bits[15:0] all active).
///
/// Width must be even (4:2:2 chroma pair).
pub(super) fn solid_y216_frame(width: u32, height: u32, y: u16, u: u16, v: u16) -> Vec<u16> {
  assert!(width.is_multiple_of(2), "Y216 requires even width");
  let row_elems = (width as usize) * 2;
  let mut buf = std::vec![0u16; row_elems * height as usize];
  for row in 0..height as usize {
    let off = row * row_elems;
    for q in 0..(width as usize / 2) {
      let base = off + q * 4;
      buf[base] = y;
      buf[base + 1] = u;
      buf[base + 2] = y;
      buf[base + 3] = v;
    }
  }
  buf
}

// ---- Single-output gray-to-gray tests ---------------------------------

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y216_luma_only_extracts_y_bytes_downshifted() {
  // Y=51200 (16-bit direct) → 8-bit (51200 >> 8) = 200.
  let buf = solid_y216_frame(6, 8, 51200, 32768, 32768);
  let src = Y216Frame::new(&buf, 6, 8, 12);
  let mut luma = std::vec![0u8; 6 * 8];
  let mut sink = MixedSinker::<Y216>::new(6, 8).with_luma(&mut luma).unwrap();
  y216_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
  // 16-bit Y=51200 → 8-bit (51200 >> 8) = 200.
  assert!(luma.iter().all(|&y| y == 200), "luma {luma:?}");
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y216_luma_u16_only_extracts_y_native_depth() {
  // Y216 luma_u16 is a direct extract — no shift.
  let buf = solid_y216_frame(6, 8, 51200, 32768, 32768);
  let src = Y216Frame::new(&buf, 6, 8, 12);
  let mut luma = std::vec![0u16; 6 * 8];
  let mut sink = MixedSinker::<Y216>::new(6, 8)
    .with_luma_u16(&mut luma)
    .unwrap();
  y216_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
  assert!(
    luma.iter().all(|&y| y == 51200),
    "luma_u16 {:?}",
    &luma[..16]
  );
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y216_rgb_only_converts_gray_to_gray() {
  // Y=32896 (≈ 128/255 x 65535), U=V=32768 (neutral chroma, ≈ 0.5x65535).
  let buf = solid_y216_frame(12, 4, 32896, 32768, 32768);
  let src = Y216Frame::new(&buf, 12, 4, 24);
  let mut rgb = std::vec![0u8; 12 * 4 * 3];
  let mut sink = MixedSinker::<Y216>::new(12, 4).with_rgb(&mut rgb).unwrap();
  y216_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
  for px in rgb.chunks(3) {
    assert!(px[0].abs_diff(128) <= 2);
    assert_eq!(px[0], px[1]);
    assert_eq!(px[1], px[2]);
  }
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y216_rgba_only_converts_gray_to_gray_with_opaque_alpha() {
  let buf = solid_y216_frame(12, 4, 32896, 32768, 32768);
  let src = Y216Frame::new(&buf, 12, 4, 24);
  let mut rgba = std::vec![0u8; 12 * 4 * 4];
  let mut sink = MixedSinker::<Y216>::new(12, 4)
    .with_rgba(&mut rgba)
    .unwrap();
  y216_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
  for px in rgba.chunks(4) {
    assert_eq!(px[3], 0xFF);
  }
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y216_rgb_u16_only_converts_gray_to_gray_native_depth() {
  let buf = solid_y216_frame(12, 4, 32896, 32768, 32768);
  let src = Y216Frame::new(&buf, 12, 4, 24);
  let mut rgb = std::vec![0u16; 12 * 4 * 3];
  let mut sink = MixedSinker::<Y216>::new(12, 4)
    .with_rgb_u16(&mut rgb)
    .unwrap();
  y216_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
  for px in rgb.chunks(3) {
    // Mid-gray at 16-bit depth ≈ 32768; allow ±256 tolerance for
    // Q15 rounding on the 16-bit pipeline.
    assert!(
      px[0].abs_diff(32768) <= 512,
      "expected ~32768, got {}",
      px[0]
    );
  }
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y216_rgba_u16_alpha_is_max() {
  let buf = solid_y216_frame(12, 4, 32896, 32768, 32768);
  let src = Y216Frame::new(&buf, 12, 4, 24);
  let mut rgba = std::vec![0u16; 12 * 4 * 4];
  let mut sink = MixedSinker::<Y216>::new(12, 4)
    .with_rgba_u16(&mut rgba)
    .unwrap();
  y216_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
  for px in rgba.chunks(4) {
    assert_eq!(px[3], 0xFFFF);
  }
}

// ---- Strategy A invariant tests ---------------------------------------

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y216_with_rgb_and_with_rgba_byte_identical_u8() {
  // Strategy A invariant on u8 path — calling both `with_rgb` and
  // `with_rgba` must produce the same RGB bytes in both buffers, with
  // alpha = 0xFF in the RGBA buffer.
  let w = 12u32;
  let h = 4u32;
  let buf = solid_y216_frame(w, h, 45000, 20000, 50000);
  let src = Y216Frame::new(&buf, w, h, w * 2);
  let mut rgb = std::vec![0u8; (w * h) as usize * 3];
  let mut rgba = std::vec![0u8; (w * h) as usize * 4];
  let mut sink = MixedSinker::<Y216>::new(w as usize, h as usize)
    .with_rgb(&mut rgb)
    .unwrap()
    .with_rgba(&mut rgba)
    .unwrap();
  y216_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
  for i in 0..(w * h) as usize {
    assert_eq!(rgba[i * 4], rgb[i * 3]);
    assert_eq!(rgba[i * 4 + 1], rgb[i * 3 + 1]);
    assert_eq!(rgba[i * 4 + 2], rgb[i * 3 + 2]);
    assert_eq!(rgba[i * 4 + 3], 0xFF);
  }
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y216_with_rgb_u16_and_with_rgba_u16_byte_identical() {
  // Strategy A invariant on u16 path — alpha must be 0xFFFF (full
  // 16-bit opaque).
  let w = 12u32;
  let h = 4u32;
  let buf = solid_y216_frame(w, h, 45000, 20000, 50000);
  let src = Y216Frame::new(&buf, w, h, w * 2);
  let mut rgb = std::vec![0u16; (w * h) as usize * 3];
  let mut rgba = std::vec![0u16; (w * h) as usize * 4];
  let mut sink = MixedSinker::<Y216>::new(w as usize, h as usize)
    .with_rgb_u16(&mut rgb)
    .unwrap()
    .with_rgba_u16(&mut rgba)
    .unwrap();
  y216_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
  for i in 0..(w * h) as usize {
    assert_eq!(rgba[i * 4], rgb[i * 3]);
    assert_eq!(rgba[i * 4 + 1], rgb[i * 3 + 1]);
    assert_eq!(rgba[i * 4 + 2], rgb[i * 3 + 2]);
    assert_eq!(rgba[i * 4 + 3], 0xFFFF);
  }
}

// ---- SIMD-vs-scalar parity --------------------------------------------

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y216_with_simd_false_matches_with_simd_true() {
  // Pseudo-random Y216 across multiple widths covering the main loop
  // + scalar tail of every backend block size. Each u16 sample uses
  // the full 16-bit range — no MSB alignment needed for Y216.
  for w in [2usize, 4, 14, 16, 18, 30, 32, 34, 1920, 1922] {
    let h = 2usize;
    let mut buf = std::vec![0u16; w * 2 * h];
    pseudo_random_u16_low_n_bits(&mut buf, 0xC0FFEE, 16);
    let src = Y216Frame::new(&buf, w as u32, h as u32, (w * 2) as u32);

    let mut rgb_simd = std::vec![0u8; w * h * 3];
    let mut rgb_scalar = std::vec![0u8; w * h * 3];
    let mut sink_simd = MixedSinker::<Y216>::new(w, h)
      .with_rgb(&mut rgb_simd)
      .unwrap();
    let mut sink_scalar = MixedSinker::<Y216>::new(w, h)
      .with_rgb(&mut rgb_scalar)
      .unwrap()
      .with_simd(false);
    y216_to(
      &src,
      false,
      sink_simd.set_kernel_matrix(KernelMatrix::Bt709),
    )
    .unwrap();
    y216_to(
      &src,
      false,
      sink_scalar.set_kernel_matrix(KernelMatrix::Bt709),
    )
    .unwrap();
    assert_eq!(rgb_simd, rgb_scalar, "Y216 SIMD≠scalar at width {w}");
  }
}

// ---- Error-path tests --------------------------------------------------

#[test]
fn y216_luma_u16_buffer_too_short_returns_err() {
  // Buffer holds 6x7 = 42 elements; a 6x8 frame needs 48.
  let mut luma = std::vec![0u16; 6 * 7];
  let result = MixedSinker::<Y216>::new(6, 8).with_luma_u16(&mut luma);
  let Err(err) = result else {
    panic!("expected InsufficientLumaU16Buffer");
  };
  assert_eq!(
    err,
    MixedSinkerError::InsufficientLumaU16Buffer(InsufficientBuffer::new(48, 42))
  );
}

// ---- Planar parity oracle ---------------------------------------------------

/// Pack three 16-bit planes (Y / U / V at 4:2:2 subsampling) into Y216
/// layout — each row is `width x 2` u16 elements laid out as `(Y₀, U,
/// Y₁, V)` quadruples. Y216 uses the full u16 range with no alignment
/// shift; samples are stored direct. Width must be even.
#[cfg(feature = "yuv-planar")]
fn pack_yuv422p16_to_y216(
  y_plane: &[u16],
  u_plane: &[u16],
  v_plane: &[u16],
  width: usize,
  height: usize,
) -> Vec<u16> {
  let mut packed = std::vec::Vec::with_capacity(width * 2 * height);
  for r in 0..height {
    for c in (0..width).step_by(2) {
      packed.push(y_plane[r * width + c]); // Y₀
      packed.push(u_plane[r * (width / 2) + c / 2]); // U
      packed.push(y_plane[r * width + c + 1]); // Y₁
      packed.push(v_plane[r * (width / 2) + c / 2]); // V
    }
  }
  packed
}

// Oracle compares against the planar `Yuv422p16` source, so this case
// only builds when that family is present.
#[cfg(feature = "yuv-planar")]
#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y216_planar_parity_with_yuv422p16() {
  // Oracle: Yuv422p16 (separate planes) and Y216 (interleaved u16 YUYV)
  // carry identical logical samples — both paths MUST produce
  // byte-identical RGB output. This exercises the i64 chroma pipeline
  // introduced for Y216 (the main point of the kernel family).
  let width = 16usize;
  let height = 4usize;
  let yp: Vec<u16> = (0..(width * height)).map(|i| (i * 17) as u16).collect();
  let up: Vec<u16> = (0..((width / 2) * height))
    .map(|i| ((i + 100) * 13) as u16)
    .collect();
  let vp: Vec<u16> = (0..((width / 2) * height))
    .map(|i| ((i + 200) * 11) as u16)
    .collect();

  let planar = Yuv422p16Frame::new(
    &yp,
    &up,
    &vp,
    width as u32,
    height as u32,
    width as u32,
    (width / 2) as u32,
    (width / 2) as u32,
  );

  let packed = pack_yuv422p16_to_y216(&yp, &up, &vp, width, height);
  let y216 = Y216Frame::new(&packed, width as u32, height as u32, (width * 2) as u32);

  // u8 RGB parity
  let mut p_rgb = std::vec![0u8; width * height * 3];
  let mut y_rgb = std::vec![0u8; width * height * 3];
  let mut p_sink = MixedSinker::<Yuv422p16>::new(width, height)
    .with_rgb(&mut p_rgb)
    .unwrap();
  let mut y_sink = MixedSinker::<Y216>::new(width, height)
    .with_rgb(&mut y_rgb)
    .unwrap();
  yuv422p16_to(
    &planar,
    false,
    p_sink.set_kernel_matrix(KernelMatrix::Bt709),
  )
  .unwrap();
  y216_to(&y216, false, y_sink.set_kernel_matrix(KernelMatrix::Bt709)).unwrap();
  assert_eq!(p_rgb, y_rgb, "Y216 vs Yuv422p16 u8 RGB diverges");

  // u16 RGB parity (the i64 chroma path — the whole point of this oracle)
  let mut p_rgb_u16 = std::vec![0u16; width * height * 3];
  let mut y_rgb_u16 = std::vec![0u16; width * height * 3];
  let mut p_sink2 = MixedSinker::<Yuv422p16>::new(width, height)
    .with_rgb_u16(&mut p_rgb_u16)
    .unwrap();
  let mut y_sink2 = MixedSinker::<Y216>::new(width, height)
    .with_rgb_u16(&mut y_rgb_u16)
    .unwrap();
  yuv422p16_to(
    &planar,
    false,
    p_sink2.set_kernel_matrix(KernelMatrix::Bt709),
  )
  .unwrap();
  y216_to(&y216, false, y_sink2.set_kernel_matrix(KernelMatrix::Bt709)).unwrap();
  assert_eq!(p_rgb_u16, y_rgb_u16, "Y216 vs Yuv422p16 u16 RGB diverges");
}

// Frame BE flag — Y216 LE/BE round-trip parity test.
// Mirrors the Y210 / Y212 / V210 LE/BE tests; see y210.rs for the full pattern.

fn y216_as_be_u16(host: &[u16]) -> Vec<u16> {
  host
    .iter()
    .map(|v| u16::from_ne_bytes(v.to_be_bytes()))
    .collect()
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y216_le_be_roundtrip_byte_identical() {
  // Mid-range full 16-bit samples (no MSB alignment for Y216).
  let intended = solid_y216_frame(8, 4, 38400, 25600, 44800);
  let pix_le = intended.clone();
  let pix_be = y216_as_be_u16(&intended);

  let frame_le = Y216LeFrame::try_new(&pix_le, 8, 4, 16).unwrap();
  let mut out_le_rgba = std::vec![0u8; 8 * 4 * 4];
  let mut out_le_luma_u16 = std::vec![0u16; 8 * 4];
  let mut sink_le = MixedSinker::<Y216>::new(8, 4)
    .with_simd(false)
    .with_rgba(&mut out_le_rgba)
    .unwrap()
    .with_luma_u16(&mut out_le_luma_u16)
    .unwrap();
  y216_to(
    &frame_le,
    true,
    sink_le.set_kernel_matrix(KernelMatrix::Bt709),
  )
  .unwrap();

  let frame_be = Y216BeFrame::try_new(&pix_be, 8, 4, 16).unwrap();
  let mut out_be_rgba = std::vec![0u8; 8 * 4 * 4];
  let mut out_be_luma_u16 = std::vec![0u16; 8 * 4];
  let mut sink_be = MixedSinker::<Y216<true>>::new(8, 4)
    .with_simd(false)
    .with_rgba(&mut out_be_rgba)
    .unwrap()
    .with_luma_u16(&mut out_be_luma_u16)
    .unwrap();
  y216_to_endian(
    &frame_be,
    true,
    sink_be.set_kernel_matrix(KernelMatrix::Bt709),
  )
  .unwrap();

  assert_eq!(
    out_le_rgba, out_be_rgba,
    "Y216 RGBA u8 LE/BE outputs diverge — `<const BE>` propagation broken"
  );
  assert_eq!(
    out_le_luma_u16, out_be_luma_u16,
    "Y216 luma u16 LE/BE outputs diverge — `<const BE>` propagation broken"
  );
}

// ---- Atomicity (#308): RGB-scratch preflight before output writes -------
//
// Y216's identity `process` hoists the RGB-scratch reservation above every
// output write (luma, luma_u16, then the u16 RGB / RGBA fan-out), so an
// allocator refusal of the scratch grow surfaces as a recoverable
// `AllocationFailed` BEFORE any output plane is touched. The allocating
// (rgb = None) arm of `rgb_row_buf_or_scratch` is reached only at
// `want_hsv && want_rgba && !want_rgb` (HSV-only routes through the direct
// `y216_to_hsv_row_endian` kernel and never allocates): attach luma + luma_u16
// + rgba + hsv (no rgb) and arm the shared failpoint. `yuva`-gated (shares the
// crate's RGB-scratch failpoint).
#[cfg(feature = "yuva")]
#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y216_rgb_scratch_alloc_failure_leaves_outputs_untouched() {
  use crate::resample::ResampleError;

  let buf = solid_y216_frame(16, 8, 32768, 32768, 32768);
  let src = Y216Frame::new(&buf, 16, 8, 32);
  let mut luma = std::vec![0xABu8; 16 * 8];
  let mut luma_u16 = std::vec![0xABCDu16; 16 * 8];
  let mut rgba = std::vec![0xCDu8; 16 * 8 * 4];
  let (mut hh, mut ss, mut vv) = (
    std::vec![0xCDu8; 16 * 8],
    std::vec![0xCDu8; 16 * 8],
    std::vec![0xCDu8; 16 * 8],
  );
  let mut sink = MixedSinker::<Y216>::new(16, 8)
    .with_luma(&mut luma)
    .unwrap()
    .with_luma_u16(&mut luma_u16)
    .unwrap()
    .with_rgba(&mut rgba)
    .unwrap()
    .with_hsv(&mut hh, &mut ss, &mut vv)
    .unwrap();

  super::super::arm_rgb_scratch_alloc_failure();
  let err = y216_to(&src, false, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap_err();
  drop(sink);

  assert!(
    matches!(
      err,
      MixedSinkerError::Resample(ResampleError::AllocationFailed(_))
    ),
    "RGB-scratch refusal must surface as a recoverable AllocationFailed, got {err:?}"
  );
  assert!(
    luma.iter().all(|&b| b == 0xAB),
    "luma must be untouched on the rgb-scratch alloc-failure path"
  );
  assert!(
    luma_u16.iter().all(|&b| b == 0xABCD),
    "luma_u16 must be untouched on the rgb-scratch alloc-failure path"
  );
}