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
//! Fused-downscale coverage for the high-bit packed RGB family
//! (`Rgb48` / `Bgr48`): the wire row converts to source-width host u16
//! RGB, binning runs at native 16-bit depth, the native-depth
//! `rgb_u16` / `rgba_u16` outputs are exact area means, and the u8 /
//! `luma_u16` outputs derive from a single `>> 8` narrowing — the same
//! source-of-truth ordering the direct path uses.

use crate::{
  KernelMatrix, PixelSink,
  resample::{AreaResampler, ResampleError},
  sinker::{MixedSinker, MixedSinkerError},
  source::{Bgr48, Bgr48Row, Rgb48, Rgb48Row, bgr48_to, rgb48_to},
};
use mediaframe::frame::{Bgr48Frame, Rgb48Frame};

const SRC: usize = 8;
const OUT: usize = 4;

/// Re-encode a host-native u16 slice as LE-wire byte storage, so a
/// fixture reads back identically on LE (no-op) and BE (byte-swap)
/// hosts.
fn as_le_u16(host: &[u16]) -> Vec<u16> {
  host
    .iter()
    .map(|v| u16::from_ne_bytes(v.to_le_bytes()))
    .collect()
}

/// Per-channel full-range u16 ramps; every value interior so the
/// derived kernels (luma / hsv) see real math and the wide
/// accumulator carries bits a u8 path would drop.
fn packed_frame_u16() -> Vec<u16> {
  let mut buf = vec![0u16; SRC * SRC * 3];
  for (i, px) in buf.chunks_exact_mut(3).enumerate() {
    px[0] = 4000 + (i as u16) * 600;
    px[1] = 60000 - (i as u16) * 700;
    px[2] = 1000 + ((i % 8) as u16) * 5000;
  }
  buf
}

/// Exact 2x2 block mean with round-half-up over host u16 values — the
/// contract for integer-ratio area downscale at native depth.
fn expected_block_mean_u16(src: &[u16], ox: usize, oy: usize, c: usize) -> u16 {
  let mut acc = 0u64;
  for dy in 0..2 {
    for dx in 0..2 {
      acc += src[((oy * 2 + dy) * SRC + ox * 2 + dx) * 3 + c] as u64;
    }
  }
  ((acc + 2) / 4) as u16
}

#[test]
fn rgb48_downscale_rgb_u16_is_exact_area_mean() {
  let host = packed_frame_u16();
  let wire = as_le_u16(&host);
  let src = Rgb48Frame::new(&wire, SRC as u32, SRC as u32, (SRC * 3) as u32);

  let mut rgb_u16 = vec![0u16; OUT * OUT * 3];
  {
    let mut sink =
      MixedSinker::<Rgb48, AreaResampler>::with_resampler(SRC, SRC, AreaResampler::to(OUT, OUT))
        .unwrap()
        .with_rgb_u16(&mut rgb_u16)
        .unwrap();
    rgb48_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt709)).unwrap();
  }
  for oy in 0..OUT {
    for ox in 0..OUT {
      for c in 0..3 {
        assert_eq!(
          rgb_u16[(oy * OUT + ox) * 3 + c],
          expected_block_mean_u16(&host, ox, oy, c),
          "({ox},{oy}) c{c}"
        );
      }
    }
  }
}

#[test]
fn rgb48_derived_outputs_come_from_binned_rgb() {
  // Every attached output — native-depth u16 and narrowed u8 — must be
  // exactly what the direct full-res Rgb48 sink produces over a frame
  // that already holds the binned u16 RGB.
  let host = packed_frame_u16();
  let wire = as_le_u16(&host);
  let src = Rgb48Frame::new(&wire, SRC as u32, SRC as u32, (SRC * 3) as u32);

  let mut rgb = vec![0u8; OUT * OUT * 3];
  let mut rgb_u16 = vec![0u16; OUT * OUT * 3];
  let mut rgba = vec![0u8; OUT * OUT * 4];
  let mut rgba_u16 = vec![0u16; OUT * OUT * 4];
  let mut luma = vec![0u8; OUT * OUT];
  let mut luma_u16 = vec![0u16; OUT * OUT];
  let mut h = vec![0u8; OUT * OUT];
  let mut s_ = vec![0u8; OUT * OUT];
  let mut v_ = vec![0u8; OUT * OUT];
  {
    let mut sink =
      MixedSinker::<Rgb48, AreaResampler>::with_resampler(SRC, SRC, AreaResampler::to(OUT, OUT))
        .unwrap()
        .with_rgb(&mut rgb)
        .unwrap()
        .with_rgb_u16(&mut rgb_u16)
        .unwrap()
        .with_rgba(&mut rgba)
        .unwrap()
        .with_rgba_u16(&mut rgba_u16)
        .unwrap()
        .with_luma(&mut luma)
        .unwrap()
        .with_luma_u16(&mut luma_u16)
        .unwrap()
        .with_hsv(&mut h, &mut s_, &mut v_)
        .unwrap();
    rgb48_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt709)).unwrap();
  }

  // Reference: the full-res sink over the (exact) binned u16 RGB.
  let binned_wire = as_le_u16(&rgb_u16);
  let mut ref_rgb = vec![0u8; OUT * OUT * 3];
  let mut ref_rgba = vec![0u8; OUT * OUT * 4];
  let mut ref_rgba_u16 = vec![0u16; OUT * OUT * 4];
  let mut ref_luma = vec![0u8; OUT * OUT];
  let mut ref_luma_u16 = vec![0u16; OUT * OUT];
  let mut ref_h = vec![0u8; OUT * OUT];
  let mut ref_s = vec![0u8; OUT * OUT];
  let mut ref_v = vec![0u8; OUT * OUT];
  {
    let binned = Rgb48Frame::new(&binned_wire, OUT as u32, OUT as u32, (OUT * 3) as u32);
    let mut sink = MixedSinker::<Rgb48>::new(OUT, OUT)
      .with_rgb(&mut ref_rgb)
      .unwrap()
      .with_rgba(&mut ref_rgba)
      .unwrap()
      .with_rgba_u16(&mut ref_rgba_u16)
      .unwrap()
      .with_luma(&mut ref_luma)
      .unwrap()
      .with_luma_u16(&mut ref_luma_u16)
      .unwrap()
      .with_hsv(&mut ref_h, &mut ref_s, &mut ref_v)
      .unwrap();
    rgb48_to(&binned, true, sink.set_kernel_matrix(KernelMatrix::Bt709)).unwrap();
  }
  assert_eq!(rgb, ref_rgb, "rgb");
  assert_eq!(rgba, ref_rgba, "rgba");
  assert_eq!(rgba_u16, ref_rgba_u16, "rgba_u16");
  assert_eq!(luma, ref_luma, "luma");
  assert_eq!(luma_u16, ref_luma_u16, "luma_u16");
  assert_eq!(h, ref_h, "h");
  assert_eq!(s_, ref_s, "s");
  assert_eq!(v_, ref_v, "v");
}

#[test]
fn rgb48_identity_plan_matches_new_sink() {
  let host = packed_frame_u16();
  let wire = as_le_u16(&host);
  let src = Rgb48Frame::new(&wire, SRC as u32, SRC as u32, (SRC * 3) as u32);

  let mut direct = vec![0u16; SRC * SRC * 3];
  {
    let mut sink = MixedSinker::<Rgb48>::new(SRC, SRC)
      .with_rgb_u16(&mut direct)
      .unwrap();
    rgb48_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt709)).unwrap();
  }
  let mut via_area = vec![0u16; SRC * SRC * 3];
  {
    let mut sink =
      MixedSinker::<Rgb48, AreaResampler>::with_resampler(SRC, SRC, AreaResampler::to(SRC, SRC))
        .unwrap()
        .with_rgb_u16(&mut via_area)
        .unwrap();
    rgb48_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt709)).unwrap();
  }
  assert_eq!(direct, via_area);
}

#[test]
fn rgb48_contracts_hold_on_the_fused_path() {
  let host = packed_frame_u16();
  let wire = as_le_u16(&host);
  let row0 = &wire[..SRC * 3];

  // Out-of-order direct process.
  let mut rgb_u16 = vec![0u16; OUT * OUT * 3];
  {
    let mut sink =
      MixedSinker::<Rgb48, AreaResampler>::with_resampler(SRC, SRC, AreaResampler::to(OUT, OUT))
        .unwrap()
        .with_rgb_u16(&mut rgb_u16)
        .unwrap();
    sink.begin_frame(SRC as u32, SRC as u32).unwrap();
    let err = sink
      .process(Rgb48Row::for_tests(row0, 3, KernelMatrix::Bt709, true))
      .unwrap_err();
    assert!(matches!(
      err,
      MixedSinkerError::Resample(ResampleError::OutOfSequenceRow(_))
    ));
  }

  // Mid-frame output change.
  let mut luma = vec![0u8; OUT * OUT];
  {
    let mut sink =
      MixedSinker::<Rgb48, AreaResampler>::with_resampler(SRC, SRC, AreaResampler::to(OUT, OUT))
        .unwrap()
        .with_rgb_u16(&mut rgb_u16)
        .unwrap();
    sink.begin_frame(SRC as u32, SRC as u32).unwrap();
    sink
      .process(Rgb48Row::for_tests(row0, 0, KernelMatrix::Bt709, true))
      .unwrap();
    sink.set_luma(&mut luma).unwrap();
    let err = sink
      .process(Rgb48Row::for_tests(
        &wire[SRC * 3..SRC * 6],
        1,
        KernelMatrix::Bt709,
        true,
      ))
      .unwrap_err();
    assert!(matches!(err, MixedSinkerError::ResampleOutputsChanged(_)));
  }
  assert!(luma.iter().all(|&l| l == 0));
}

#[test]
fn bgr48_downscale_rgb_u16_is_exact_area_mean() {
  // Bgr48 stores B, G, R; the converted host RGB swaps channel 0 and 2,
  // so the binned rgb_u16 output equals the block mean of the source
  // with the same swap applied.
  let host = packed_frame_u16();
  let wire = as_le_u16(&host);
  let src = Bgr48Frame::new(&wire, SRC as u32, SRC as u32, (SRC * 3) as u32);

  let mut rgb_u16 = vec![0u16; OUT * OUT * 3];
  {
    let mut sink =
      MixedSinker::<Bgr48, AreaResampler>::with_resampler(SRC, SRC, AreaResampler::to(OUT, OUT))
        .unwrap()
        .with_rgb_u16(&mut rgb_u16)
        .unwrap();
    bgr48_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt709)).unwrap();
  }
  let swap = [2usize, 1, 0];
  for oy in 0..OUT {
    for ox in 0..OUT {
      for c in 0..3 {
        assert_eq!(
          rgb_u16[(oy * OUT + ox) * 3 + c],
          expected_block_mean_u16(&host, ox, oy, swap[c]),
          "({ox},{oy}) c{c}"
        );
      }
    }
  }
}

#[test]
fn rgb48_u16_only_downscale_does_not_size_the_narrow_scratch() {
  let host = packed_frame_u16();
  let wire = as_le_u16(&host);
  let src = Rgb48Frame::new(&wire, SRC as u32, SRC as u32, (SRC * 3) as u32);

  // Native-u16-only sink (only rgb_u16 attached): the binned row is
  // copied at native depth, so the u8 narrow scratch is never sized.
  let mut rgb_u16 = vec![0u16; OUT * OUT * 3];
  let mut sink =
    MixedSinker::<Rgb48, AreaResampler>::with_resampler(SRC, SRC, AreaResampler::to(OUT, OUT))
      .unwrap()
      .with_rgb_u16(&mut rgb_u16)
      .unwrap();
  rgb48_to(&src, true, sink.set_kernel_matrix(KernelMatrix::Bt709)).unwrap();
  assert_eq!(
    sink.rgb_scratch_capacity(),
    0,
    "native-u16-only sink grew the u8 narrow scratch"
  );

  // Positive control: attaching a u8 output sizes it to the out-width
  // RGB row.
  let mut rgb_u16b = vec![0u16; OUT * OUT * 3];
  let mut rgb = vec![0u8; OUT * OUT * 3];
  let mut sink2 =
    MixedSinker::<Rgb48, AreaResampler>::with_resampler(SRC, SRC, AreaResampler::to(OUT, OUT))
      .unwrap()
      .with_rgb_u16(&mut rgb_u16b)
      .unwrap()
      .with_rgb(&mut rgb)
      .unwrap();
  rgb48_to(&src, true, sink2.set_kernel_matrix(KernelMatrix::Bt709)).unwrap();
  assert!(
    sink2.rgb_scratch_capacity() >= OUT * 3,
    "u8 output did not size the narrow scratch"
  );
}

#[test]
fn rgb48_out_of_sequence_first_row_rejected_before_allocation() {
  let host = packed_frame_u16();
  let wire = as_le_u16(&host);
  let row3 = &wire[3 * SRC * 3..4 * SRC * 3];

  let mut rgb_u16 = vec![0u16; OUT * OUT * 3];
  let mut sink =
    MixedSinker::<Rgb48, AreaResampler>::with_resampler(SRC, SRC, AreaResampler::to(OUT, OUT))
      .unwrap()
      .with_rgb_u16(&mut rgb_u16)
      .unwrap();
  sink.begin_frame(SRC as u32, SRC as u32).unwrap();
  let err = sink
    .process(Rgb48Row::for_tests(row3, 3, KernelMatrix::Bt709, true))
    .unwrap_err();
  assert!(
    matches!(
      err,
      MixedSinkerError::Resample(ResampleError::OutOfSequenceRow(_))
    ),
    "expected OutOfSequenceRow, got {err:?}"
  );
  // The out-of-sequence first row must be rejected before the u16
  // stream or the narrow scratch is allocated.
  assert!(
    !sink.rgb_stream_u16_allocated(),
    "stream allocated for a rejected row"
  );
  assert_eq!(
    sink.rgb_scratch_capacity(),
    0,
    "scratch grown for a rejected row"
  );
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn bgr48_first_row_scratch_oom_leaves_stream_and_freeze_uncommitted_for_retry() {
  let host = packed_frame_u16();
  let wire = as_le_u16(&host);
  let mut rgb_u16 = vec![0u16; OUT * OUT * 3];
  let mut luma = vec![0u8; OUT * OUT];
  {
    let mut sink =
      MixedSinker::<Bgr48, AreaResampler>::with_resampler(SRC, SRC, AreaResampler::to(OUT, OUT))
        .unwrap()
        .with_rgb_u16(&mut rgb_u16)
        .unwrap();
    sink.begin_frame(SRC as u32, SRC as u32).unwrap();
    // Arm the single-shot source-RGB-u16-scratch grow failpoint: Bgr48 stages its
    // wire row into the host-u16 scratch, so the row-0 stream box builds OK and the
    // scratch grow then refuses — surfacing AllocationFailed with BOTH the stream
    // field `None` and the freeze uncommitted (the commit-together atomic shape; the
    // pre-fix ordering would have left the stream committed).
    crate::sinker::mixed::arm_source_rgb_u16_scratch_failure();
    let err = sink
      .process(Bgr48Row::for_tests(
        &wire[..SRC * 3],
        0,
        KernelMatrix::Bt709,
        true,
      ))
      .unwrap_err();
    assert!(
      matches!(
        err,
        MixedSinkerError::Resample(ResampleError::AllocationFailed(_))
      ),
      "first-row scratch OOM must surface AllocationFailed, got {err:?}"
    );
    // DIRECT stream-rollback proof (the non-vacuous part): the u16 area stream field
    // must be `None` — the fix builds it into a LOCAL and inserts only after the
    // scratch grows, so a failed scratch leaves nothing committed. The insert-before-
    // scratch shape would leave it `Some` here, so this assertion FAILS against that
    // shape and PASSES against the fix.
    assert!(
      !sink.rgb_stream_u16_allocated(),
      "a scratch OOM must leave the rgb_stream_u16 field None (no partial commit)"
    );
    // The freeze was likewise never committed — attaching luma (a CHANGED output
    // set) and replaying from row 0 is ACCEPTED, not ResampleOutputsChanged.
    sink.set_luma(&mut luma).unwrap();
    for r in 0..SRC {
      sink
        .process(Bgr48Row::for_tests(
          &wire[r * SRC * 3..(r + 1) * SRC * 3],
          r,
          KernelMatrix::Bt709,
          true,
        ))
        .expect("frame replay after a first-row scratch OOM must succeed");
    }
  }
  assert!(
    rgb_u16.iter().any(|&b| b != 0),
    "the recovered frame produced real rgb_u16 output (scratch OOM was recoverable)"
  );
  assert!(
    luma.iter().any(|&b| b != 0),
    "the newly-attached luma output was driven on the accepted retry"
  );
}

#[test]
fn rgb48_rejected_first_row_does_not_poison_output_retry() {
  // A rejected out-of-sequence FIRST row must store no frozen-output
  // snapshot (the split high-bit packed-RGB preflight rejects the OOS
  // first row before its freeze), so retrying row 0 after attaching a new
  // output succeeds instead of tripping ResampleOutputsChanged.
  let host = packed_frame_u16();
  let wire = as_le_u16(&host);
  let mut rgb_u16 = vec![0u16; OUT * OUT * 3];
  let mut sink =
    MixedSinker::<Rgb48, AreaResampler>::with_resampler(SRC, SRC, AreaResampler::to(OUT, OUT))
      .unwrap()
      .with_rgb_u16(&mut rgb_u16)
      .unwrap();
  sink.begin_frame(SRC as u32, SRC as u32).unwrap();
  let err = sink
    .process(Rgb48Row::for_tests(
      &wire[3 * SRC * 3..4 * SRC * 3],
      3,
      KernelMatrix::Bt709,
      true,
    ))
    .unwrap_err();
  assert!(
    matches!(
      err,
      MixedSinkerError::Resample(ResampleError::OutOfSequenceRow(_))
    ),
    "expected OutOfSequenceRow, got {err:?}"
  );
  let mut luma = vec![0u8; OUT * OUT];
  sink.set_luma(&mut luma).unwrap();
  sink
    .process(Rgb48Row::for_tests(
      &wire[..SRC * 3],
      0,
      KernelMatrix::Bt709,
      true,
    ))
    .expect("row 0 must succeed after a rejected out-of-sequence first row");
}