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
//! Tier 4 Y212 sinker tests — Ship 11c.
//!
//! 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 Y212 builder -----------------------------------------

/// Builds a solid-color Y212 plane with one (Y, U, V) repeated. Each
/// row is `width x 2` u16 elements (`Y₀, U, Y₁, V` quadruples). All
/// samples are MSB-aligned 12-bit: each `u16` value's active 12 bits
/// occupy bits[15:4] and bits[3:0] are zero.
///
/// Width must be even (4:2:2 chroma pair).
pub(super) fn solid_y212_frame(width: u32, height: u32, y: u16, u: u16, v: u16) -> Vec<u16> {
  assert!(width.is_multiple_of(2), "Y212 requires even width");
  let row_elems = (width as usize) * 2;
  let mut buf = std::vec![0u16; row_elems * height as usize];
  let y_msb = (y & 0xFFF) << 4;
  let u_msb = (u & 0xFFF) << 4;
  let v_msb = (v & 0xFFF) << 4;
  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_msb;
      buf[base + 1] = u_msb;
      buf[base + 2] = y_msb;
      buf[base + 3] = v_msb;
    }
  }
  buf
}

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

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y212_luma_only_extracts_y_bytes_downshifted() {
  let buf = solid_y212_frame(6, 8, 200, 2048, 2048); // Y=200 (12-bit native).
  let src = Y212Frame::new(&buf, 6, 8, 12);
  let mut luma = std::vec![0u8; 6 * 8];
  let mut sink = MixedSinker::<Y212>::new(6, 8).with_luma(&mut luma).unwrap();
  y212_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
  // 12-bit Y=200 → 8-bit (200 >> 4) = 12.
  assert!(luma.iter().all(|&y| y == 12), "luma {luma:?}");
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y212_luma_u16_only_extracts_y_native_depth() {
  let buf = solid_y212_frame(6, 8, 200, 2048, 2048);
  let src = Y212Frame::new(&buf, 6, 8, 12);
  let mut luma = std::vec![0u16; 6 * 8];
  let mut sink = MixedSinker::<Y212>::new(6, 8)
    .with_luma_u16(&mut luma)
    .unwrap();
  y212_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
  assert!(luma.iter().all(|&y| y == 200), "luma_u16 {:?}", &luma[..16]);
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y212_rgb_only_converts_gray_to_gray() {
  let buf = solid_y212_frame(12, 4, 2048, 2048, 2048);
  let src = Y212Frame::new(&buf, 12, 4, 24);
  let mut rgb = std::vec![0u8; 12 * 4 * 3];
  let mut sink = MixedSinker::<Y212>::new(12, 4).with_rgb(&mut rgb).unwrap();
  y212_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
  for px in rgb.chunks(3) {
    assert!(px[0].abs_diff(128) <= 1);
    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 y212_rgba_only_converts_gray_to_gray_with_opaque_alpha() {
  let buf = solid_y212_frame(12, 4, 2048, 2048, 2048);
  let src = Y212Frame::new(&buf, 12, 4, 24);
  let mut rgba = std::vec![0u8; 12 * 4 * 4];
  let mut sink = MixedSinker::<Y212>::new(12, 4)
    .with_rgba(&mut rgba)
    .unwrap();
  y212_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 y212_rgb_u16_only_converts_gray_to_gray_native_depth() {
  let buf = solid_y212_frame(12, 4, 2048, 2048, 2048);
  let src = Y212Frame::new(&buf, 12, 4, 24);
  let mut rgb = std::vec![0u16; 12 * 4 * 3];
  let mut sink = MixedSinker::<Y212>::new(12, 4)
    .with_rgb_u16(&mut rgb)
    .unwrap();
  y212_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
  for px in rgb.chunks(3) {
    assert!(px[0].abs_diff(2048) <= 2, "expected ~2048, got {}", px[0]);
  }
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y212_rgba_u16_alpha_is_max() {
  let buf = solid_y212_frame(12, 4, 2048, 2048, 2048);
  let src = Y212Frame::new(&buf, 12, 4, 24);
  let mut rgba = std::vec![0u16; 12 * 4 * 4];
  let mut sink = MixedSinker::<Y212>::new(12, 4)
    .with_rgba_u16(&mut rgba)
    .unwrap();
  y212_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt601)).unwrap();
  for px in rgba.chunks(4) {
    assert_eq!(px[3], 4095);
  }
}

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

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y212_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_y212_frame(w, h, 700, 400, 600);
  let src = Y212Frame::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::<Y212>::new(w as usize, h as usize)
    .with_rgb(&mut rgb)
    .unwrap()
    .with_rgba(&mut rgba)
    .unwrap();
  y212_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 y212_with_rgb_u16_and_with_rgba_u16_byte_identical() {
  // Strategy A invariant on u16 path.
  let w = 12u32;
  let h = 4u32;
  let buf = solid_y212_frame(w, h, 700, 400, 600);
  let src = Y212Frame::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::<Y212>::new(w as usize, h as usize)
    .with_rgb_u16(&mut rgb)
    .unwrap()
    .with_rgba_u16(&mut rgba)
    .unwrap();
  y212_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], 4095);
  }
}

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

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn y212_with_simd_false_matches_with_simd_true() {
  // Pseudo-random Y212 across multiple widths covering the main loop
  // + scalar tail of every backend block size. Each u16 sample is
  // generated with 12 active bits, then shifted `<< 4` to become
  // MSB-aligned per the Y212 spec.
  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, 12);
    // MSB-align: each sample's active 12 bits move to bits[15:4].
    for s in &mut buf {
      *s <<= 4;
    }
    let src = Y212Frame::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::<Y212>::new(w, h)
      .with_rgb(&mut rgb_simd)
      .unwrap();
    let mut sink_scalar = MixedSinker::<Y212>::new(w, h)
      .with_rgb(&mut rgb_scalar)
      .unwrap()
      .with_simd(false);
    y212_to(
      &src,
      false,
      sink_simd.set_kernel_matrix(KernelMatrix::Bt709),
    )
    .unwrap();
    y212_to(
      &src,
      false,
      sink_scalar.set_kernel_matrix(KernelMatrix::Bt709),
    )
    .unwrap();
    assert_eq!(rgb_simd, rgb_scalar, "Y212 SIMD≠scalar at width {w}");
  }
}

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

#[test]
fn y212_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::<Y212>::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))
  );
}

// ---- Cross-format oracles ---------------------------------------------
//
// Planar parity: re-pack a Yuv422p12 source into Y212 layout and verify
// both produce byte-identical RGB. Y212 is just an MSB-aligned u16
// packing of 4:2:2 12-bit planar data, so the converted output MUST
// match Yuv422p12's output for the same samples.

/// Pack three 12-bit planes (Y / U / V at 4:2:2 subsampling) into the
/// Y212 layout — each row is `width x 2` u16 elements laid out as
/// `(Y₀, U, Y₁, V)` quadruples with each sample MSB-aligned (active 12
/// bits in bits[15:4]). Width must be even.
#[cfg(feature = "yuv-planar")]
fn pack_yuv422p12_to_y212(
  y: &[u16],
  u: &[u16],
  v: &[u16],
  width: usize,
  height: usize,
) -> Vec<u16> {
  let cw = width / 2;
  let mut out = std::vec![0u16; width * 2 * height];
  for row in 0..height {
    for c in 0..cw {
      let off = row * width * 2 + c * 4;
      out[off] = (y[row * width + 2 * c] << 4) & 0xFFF0;
      out[off + 1] = (u[row * cw + c] << 4) & 0xFFF0;
      out[off + 2] = (y[row * width + 2 * c + 1] << 4) & 0xFFF0;
      out[off + 3] = (v[row * cw + c] << 4) & 0xFFF0;
    }
  }
  out
}

// Oracle compares against the planar `Yuv422p12` 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 y212_reconstructed_from_yuv422p12_matches_yuv422p12_to_rgb() {
  let w = 16usize;
  let h = 4usize;
  let mut y_plane = std::vec![0u16; w * h];
  let mut u_plane = std::vec![0u16; (w / 2) * h];
  let mut v_plane = std::vec![0u16; (w / 2) * h];
  pseudo_random_u16_low_n_bits(&mut y_plane, 0xC0FFEE, 12);
  pseudo_random_u16_low_n_bits(&mut u_plane, 0xBADF00D, 12);
  pseudo_random_u16_low_n_bits(&mut v_plane, 0xFEEDFACE, 12);

  let planar = Yuv422p12Frame::new(
    &y_plane,
    &u_plane,
    &v_plane,
    w as u32,
    h as u32,
    w as u32,
    (w / 2) as u32,
    (w / 2) as u32,
  );
  let packed = pack_yuv422p12_to_y212(&y_plane, &u_plane, &v_plane, w, h);
  let y212 = Y212Frame::new(&packed, w as u32, h as u32, (w * 2) as u32);

  let mut rgb_planar = std::vec![0u8; w * h * 3];
  let mut rgb_packed = std::vec![0u8; w * h * 3];
  let mut s_planar = MixedSinker::<Yuv422p12>::new(w, h)
    .with_rgb(&mut rgb_planar)
    .unwrap();
  let mut s_packed = MixedSinker::<Y212>::new(w, h)
    .with_rgb(&mut rgb_packed)
    .unwrap();
  yuv422p12_to(
    &planar,
    false,
    s_planar.set_kernel_matrix(KernelMatrix::Bt709),
  )
  .unwrap();
  y212_to(
    &y212,
    false,
    s_packed.set_kernel_matrix(KernelMatrix::Bt709),
  )
  .unwrap();

  assert_eq!(rgb_planar, rgb_packed);
}
// Phase 4 — Frame BE flag, Tier 4 Y212 LE/BE round-trip parity test.
// Mirrors the Y210 / V210 / Y216 LE/BE tests; see y210.rs for the full pattern.
fn y212_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 y212_le_be_roundtrip_byte_identical() {
  // Mid-range 12-bit Y/U/V samples (MSB-aligned).
  let intended = solid_y212_frame(8, 4, 2400, 1600, 2800);
  let pix_le = intended.clone();
  let pix_be = y212_as_be_u16(&intended);

  let frame_le = Y212LeFrame::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::<Y212>::new(8, 4)
    .with_simd(false)
    .with_rgba(&mut out_le_rgba)
    .unwrap()
    .with_luma_u16(&mut out_le_luma_u16)
    .unwrap();
  y212_to(
    &frame_le,
    true,
    sink_le.set_kernel_matrix(KernelMatrix::Bt709),
  )
  .unwrap();

  let frame_be = Y212BeFrame::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::<Y212<true>>::new(8, 4)
    .with_simd(false)
    .with_rgba(&mut out_be_rgba)
    .unwrap()
    .with_luma_u16(&mut out_be_luma_u16)
    .unwrap();
  y212_to_endian(
    &frame_be,
    true,
    sink_be.set_kernel_matrix(KernelMatrix::Bt709),
  )
  .unwrap();

  assert_eq!(
    out_le_rgba, out_be_rgba,
    "Y212 RGBA u8 LE/BE outputs diverge — `<const BE>` propagation broken"
  );
  assert_eq!(
    out_le_luma_u16, out_be_luma_u16,
    "Y212 luma u16 LE/BE outputs diverge — `<const BE>` propagation broken"
  );
}

// ---- Atomicity (#308): RGB-scratch preflight before output writes -------
//
// Y212'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
// `y212_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 y212_rgb_scratch_alloc_failure_leaves_outputs_untouched() {
  use crate::resample::ResampleError;

  let buf = solid_y212_frame(16, 8, 2048, 2048, 2048);
  let src = Y212Frame::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::<Y212>::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 = y212_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"
  );
}