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
//! Sinker impl for the packed VUYX source format — Ship 12c (Tier 5
//! 8-bit packed YUV 4:4:4 with padding α byte).
//!
//! VUYX (FFmpeg `AV_PIX_FMT_VUYX`) packs **four u8 bytes per pixel**
//! (`[V, U, Y, X]`). The X byte is **padding** — not real source alpha.
//! RGBA outputs always force α to `0xFF`; the padding byte is ignored.
//! The packed slice type is `&[u8]`, with `4 x width` byte elements per
//! row. There is no chroma subsampling — every pixel carries its own
//! independent V / U / Y triplet (4:4:4).
//!
//! Outputs map to the sink's standard channels:
//! - `with_rgb` — packed YUV → RGB 8-bit pipeline; padding discarded.
//! - `with_rgba` — packed YUV → RGBA 8-bit pipeline; **α is forced to
//! `0xFF`** (the X byte is padding, never real alpha).
//! - `with_luma` — extracts the Y byte at offset 2 of each pixel
//! directly (no YUV→RGB pipeline).
//! - `with_luma_u16` — zero-extends the Y byte to u16
//! (`out[x] = Y_byte as u16`).
//! - `with_hsv` — stages u8 RGB into the user's RGB buffer (if
//! attached) or a scratch buffer, then runs `rgb_to_hsv_row`.
use super::{
GeometryOverflow, InsufficientBuffer, MixedSinker, MixedSinkerError, RowIndexOutOfRange,
RowShapeMismatch, RowSlice, check_dimensions_match, packed_yuv444_triple_resample,
packed_yuva444_filter_resample, rgb_row_buf_or_scratch, rgba_plane_row_slice,
};
// `NativeRouteChanged` is raised only by the native fast tier's route-flip
// guard, and `packed_vuyx_process_native` exists only when the reused 8-bit
// planar join is compiled in. Gated to the native tier's feature intersection.
#[cfg(all(feature = "yuv-444-packed", feature = "yuv-planar"))]
use super::{NativeRouteChanged, packed_vuyx_process_native};
// The RFC #238 insertion-point selector decides the native-vs-row-stage
// splice; consulted only inside the native tier's `cfg`, so its import
// shares that intersection.
#[cfg(all(feature = "yuv-444-packed", feature = "yuv-planar"))]
use crate::resample::{AveragingDomain, InsertionContext, InsertionPoint, select_insertion_point};
use crate::{
PixelSink,
row::{
expand_rgb_to_rgba_row, rgb_to_hsv_row, vuyx_to_hsv_row, vuyx_to_luma_row,
vuyx_to_luma_u16_row, vuyx_to_rgb_row, vuyx_to_rgba_row,
},
source::{Vuyx, VuyxRow, VuyxSink},
};
impl<'a, R> MixedSinker<'a, Vuyx, R> {
/// Attaches a **`u16`** luma output buffer. Y bytes from the packed VUYX
/// `[V, U, Y, X]` layout are 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)
}
/// Attaches a packed **8-bit** RGBA output buffer. When VUYX is the
/// source, the per-pixel alpha byte is always forced to `0xFF` —
/// the X (padding) byte in the source is never read as alpha.
///
/// Returns `Err(InsufficientRgbaBuffer)` if
/// `buf.len() < width x height x 4`, or `Err(GeometryOverflow)` on
/// 32‑bit targets when the product overflows.
///
/// ## Strategy note
///
/// α=`0xFF` is guaranteed in **all** paths (standalone or combined
/// with `with_rgb` / `with_hsv`). When combined with `with_rgb`,
/// RGBA is derived via Strategy A fan-out (`expand_rgb_to_rgba_row`)
/// instead of a second YUV→RGB kernel call — both produce α=`0xFF`,
/// so the outputs are semantically identical (spec § 8.4).
#[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)
}
}
impl<R> VuyxSink for MixedSinker<'_, Vuyx, R> {}
impl<R> PixelSink for MixedSinker<'_, Vuyx, R> {
type Input<'r> = VuyxRow<'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> {
check_dimensions_match(self.width, self.height, width, height)?;
// New frame: restart the row-stage streams (lazily created in
// `process`, area or filter kind) and drop the frozen output set.
// `Vuyx` exposes no u16 colour outputs, so its u16 colour streams are
// never created; resetting them unconditionally is a harmless no-op. The
// area arm uses the 3-channel colour streams; the filter arm uses the
// 4-channel `rgba_filter_stream` (the padding byte filters as a constant
// opaque α) and the native-Y `luma_filter_stream_u16`.
if let Some(stream) = self.rgb_stream.as_mut() {
stream.reset();
}
if let Some(stream) = self.rgb_stream_u16.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_u16.as_mut() {
stream.reset();
}
// New frame: restart the native join and clear the per-frame frozen
// native/row-stage route so the next frame may pick either tier; a
// mid-frame flip stays rejected. Gated to the native tier's feature
// intersection (the planar join the native tier reuses is compiled only
// under `yuv-planar`).
#[cfg(all(feature = "yuv-444-packed", feature = "yuv-planar"))]
if let Some(native) = self.native_planar.as_mut() {
native.reset();
}
#[cfg(all(feature = "yuv-444-packed", feature = "yuv-planar"))]
{
self.frozen_native_route = None;
}
self.resample_outputs = None;
Ok(())
}
fn process(&mut self, row: VuyxRow<'_>) -> Result<(), Self::Error> {
const BITS: u32 = 8;
let w = self.width;
let h = self.height;
let idx = row.row();
let use_simd = self.simd;
// VUYX row = `width x 4` bytes (one quadruple per pixel).
let packed_expected =
w.checked_mul(4)
.ok_or(MixedSinkerError::GeometryOverflow(GeometryOverflow::new(
w, h, 4,
)))?;
if row.packed().len() != packed_expected {
return Err(MixedSinkerError::RowShapeMismatch(RowShapeMismatch::new(
RowSlice::VuyxPacked,
idx,
packed_expected,
row.packed().len(),
)));
}
if idx >= self.height {
return Err(MixedSinkerError::RowIndexOutOfRange(
RowIndexOutOfRange::new(idx, self.height),
));
}
// Non-identity plan: `Vuyx` has NO alpha (the X byte is padding,
// forced opaque). The span kind picks the engine:
//
// - `Area` bins through the no-alpha three-stream tail exactly like the
// no-alpha packed 4:4:4 YUV siblings (`V30X` / `V410` / `Xv36`) — u8
// colour bins the converted u8 RGB row, native Y bins the
// de-interleaved Y plane (the colour outputs force α opaque, the
// padding byte is never read).
// - `Filter` runs the signed-coefficient filter on the converted RGBA
// via the shared packed-YUVA filter tail. `vuyx_to_rgba_row` writes a
// constant `0xFF` α plane, and a constant channel filters to itself
// (partition of unity), so the 4-channel filter reproduces the
// no-alpha 3-channel result with α pinned opaque — no separate
// padding-byte filter path is needed. Straight alpha (`Vuyx` has no
// alpha mode).
//
// `Vuyx` exposes no u16 colour outputs, so the u16 colour resampling is
// never active (`rgb_u16` / `rgba_u16` stay `None`) and the
// `convert_*_u16` closure is therefore never invoked on either arm.
if let Some(plan) = self.plan.as_ref() {
let matrix = row.matrix();
let full_range = row.full_range();
let packed = row.packed();
let Self {
rgb,
rgba,
rgb_u16,
rgba_u16,
luma,
luma_u16,
hsv,
rgb_scratch,
rgb_scratch_u16,
rgba_scratch,
rgba_scratch_u16,
rgba_color_scratch_u16,
luma_scratch_u16,
rgb_stream,
rgb_stream_u16,
luma_stream_u16,
rgba_filter_stream,
rgba_filter_stream_u16,
luma_filter_stream_u16,
resample_outputs,
#[cfg(all(feature = "yuv-444-packed", feature = "yuv-planar"))]
native,
#[cfg(all(feature = "yuv-444-packed", feature = "yuv-planar"))]
native_planar,
#[cfg(all(feature = "yuv-444-packed", feature = "yuv-planar"))]
packed_444_y_full,
#[cfg(all(feature = "yuv-444-packed", feature = "yuv-planar"))]
packed_444_u_full,
#[cfg(all(feature = "yuv-444-packed", feature = "yuv-planar"))]
packed_444_v_full,
#[cfg(all(feature = "yuv-444-packed", feature = "yuv-planar"))]
frozen_native_route,
..
} = self;
// A `Filter` plan routes straight to the signed-coefficient filter tail
// (the native fast tier is area-only); branched FIRST, before the
// area native-route machinery below.
if let crate::resample::SpanKind::Filter = plan.kind() {
return packed_yuva444_filter_resample::<BITS, false, false>(
rgba_filter_stream,
rgba_filter_stream_u16,
// Packed `Vuyx` 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,
// Packed `Vuyx` routes luma through `deinterleave_y` + the u16
// stream (no contiguous native-Y plane), so the u8-luma input and
// its de-interleave scratch are unused.
&[],
None,
// α forced opaque (`0xFF`) by `vuyx_to_rgba_row` — the padding
// byte is never read as alpha.
|dst| vuyx_to_rgba_row(packed, dst, w, matrix, full_range, use_simd),
// `Vuyx` has no u16 colour outputs, so this closure is never called.
|_dst: &mut [u16]| {},
|dst| vuyx_to_luma_u16_row(packed, dst, w, use_simd),
// u16-luma path, so this u8 de-interleave is never called.
|_dst: &mut [u8]| {},
);
}
// Area plan. When the native tier is enabled (and the planar join it
// reuses is compiled in), de-pack the interleaved `V U Y X` row into
// Y / U / V scratch (4:4:4, full-width chroma) and bin those planes at
// output resolution, converting once per output row (Vuyx: V at 0 / U at
// 1 / Y at 2 / X padding at 3). Otherwise (or under `with_native(false)`)
// take the row-stage tier. The output set is frozen on the first
// resampled row, so the native/row-stage route stays stable across a
// frame and the mid-frame flip guard catches a `set_native` toggle.
#[cfg(all(feature = "yuv-444-packed", feature = "yuv-planar"))]
{
let need_output =
luma.is_some() || luma_u16.is_some() || rgb.is_some() || rgba.is_some() || hsv.is_some();
// The RFC #238 splice stage. A filter plan already returned above, so
// `area_plan` is true and the selector reproduces the former `*native`
// boolean bit-for-bit (`cfg!` is true wherever this block compiles).
let take_native = matches!(
select_insertion_point(
AveragingDomain::Encoded,
InsertionContext {
native_eligible: cfg!(all(feature = "yuv-444-packed", feature = "yuv-planar")),
with_native: *native,
area_plan: true,
},
),
InsertionPoint::NativeCodes
);
// Reject a mid-frame native/row-stage route flip BEFORE either tier's
// dispatch (the #186 CHECK-before / SET-after template).
if need_output
&& let Some(frozen) = *frozen_native_route
&& frozen != take_native
{
return Err(MixedSinkerError::NativeRouteChanged(
NativeRouteChanged::new(idx),
));
}
if take_native {
// Dispatch first; freeze the route to native ONLY after the call
// returns Ok on an output-bearing row.
packed_vuyx_process_native(
plan,
native_planar,
packed_444_y_full,
packed_444_u_full,
packed_444_v_full,
resample_outputs,
rgb,
rgba,
luma,
luma_u16,
hsv,
rgb_scratch,
packed,
matrix,
full_range,
idx,
w,
h,
use_simd,
)?;
if frozen_native_route.is_none() && need_output {
*frozen_native_route = Some(true);
}
return Ok(());
}
}
// Row-stage area tail (the only route under a `yuv-444-packed`-solo build
// where the planar join is absent). Same CHECK-before / SET-after split.
packed_yuv444_triple_resample::<BITS>(
rgb_stream,
rgb_stream_u16,
luma_stream_u16,
resample_outputs,
rgb,
rgba,
rgb_u16,
rgba_u16,
luma,
luma_u16,
hsv,
rgb_scratch,
rgb_scratch_u16,
luma_scratch_u16,
w,
plan,
idx,
use_simd,
matrix,
full_range,
|scratch| vuyx_to_rgb_row(packed, scratch, w, matrix, full_range, use_simd),
// `Vuyx` has no u16 colour outputs, so `need_u16_color` is always
// false in the tail and this closure is never called.
|_scratch: &mut [u16]| {},
|scratch| vuyx_to_luma_u16_row(packed, scratch, w, use_simd),
)?;
#[cfg(all(feature = "yuv-444-packed", feature = "yuv-planar"))]
{
let need_output =
luma.is_some() || luma_u16.is_some() || rgb.is_some() || rgba.is_some() || hsv.is_some();
if frozen_native_route.is_none() && need_output {
*frozen_native_route = Some(false);
}
}
return Ok(());
}
let Self {
rgb,
rgba,
luma,
luma_u16,
hsv,
rgb_scratch,
..
} = self;
let one_plane_start = idx * w;
let one_plane_end = one_plane_start + w;
let packed = row.packed();
// Output mode resolution — hoisted ABOVE the luma writes so the atomicity
// preflight runs ahead of any output mutation.
let want_rgb = rgb.is_some();
let want_rgba = rgba.is_some();
let want_hsv = hsv.is_some();
// Atomicity preflight (#308, cf. the crate's #180 resample fix and the
// merged packed 4:2:2 / planar / semi-planar siblings): reserve the only
// fallible row scratch this identity row can grow BEFORE any output row
// (luma / luma_u16 included) is written, so an allocator refusal returns a
// typed `AllocationFailed` leaving the output frame untouched rather than
// partially mutated. The sole growable scratch is the RGB row buffer, taken
// exactly when the colour decode needs an RGB row but no caller RGB buffer
// is borrowable — `want_hsv && want_rgba && !want_rgb` (the
// `rgb_row_buf_or_scratch` scratch arm; an attached RGB buffer is borrowed
// and never allocates, and standalone `with_rgba` uses the dedicated
// `vuyx_to_rgba_row` kernel — α forced opaque). The later `need_rgb_kernel`
// decode reuses the already-sized buffer.
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,
)?;
}
// Luma u8 — extract Y byte (offset 2 in each VUYX quadruple) directly.
// `vuyx_to_luma_row` is a re-export of `vuya_to_luma_row` — the
// byte stream is identical (Y at offset 2 regardless of α semantics).
if let Some(buf) = luma.as_deref_mut() {
vuyx_to_luma_row(
packed,
&mut buf[one_plane_start..one_plane_end],
w,
use_simd,
);
}
// Luma u16 — extract Y bytes and zero-extend to u16.
if let Some(buf) = luma_u16.as_deref_mut() {
vuyx_to_luma_u16_row(
packed,
&mut buf[one_plane_start..one_plane_end],
w,
use_simd,
);
}
// ===== u8 RGB / RGBA / HSV path (Strategy A for VUYX) =====
// HSV-without-RGB-or-RGBA goes through the direct `vuyx_to_hsv_row`
// kernel — no source-width RGB scratch (the SIMD path stages a fixed
// 64-pixel 8-bit RGB chunk internally). When RGB or RGBA is also
// attached the RGB kernel runs anyway, so HSV derives off that buffer
// for free and `need_rgb_kernel` keeps it alive. Resample row-stage
// HSV-only is a #263 follow-up — HSV stays correct via the
// convert-once path.
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 (h, s, v) = hsv.hsv();
vuyx_to_hsv_row(
packed,
&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(());
}
// Standalone RGBA fast path — no RGB / HSV requested. Run the
// dedicated RGBA kernel directly into the output buffer; avoids
// both the scratch allocation and the RGB→RGBA expand pass.
// α is forced to `0xFF` by `vuyx_to_rgba_row` (ALPHA_SRC = false).
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)?;
vuyx_to_rgba_row(
packed,
rgba_row,
w,
row.matrix(),
row.full_range(),
use_simd,
);
return Ok(());
}
if !need_rgb_kernel {
return Ok(());
}
// RGB kernel — write into the user's RGB buffer (if attached) or the
// internal scratch buffer. Required when with_rgb or with_hsv is set.
let rgb_row = rgb_row_buf_or_scratch(
rgb.as_deref_mut(),
rgb_scratch,
one_plane_start,
one_plane_end,
w,
h,
)?;
// `vuyx_to_rgb_row` is a re-export of `vuya_to_rgb_row` — the padding
// byte is irrelevant when there is no α channel in the output.
vuyx_to_rgb_row(packed, 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,
);
}
// Strategy A u8 fan-out — derive RGBA from the just-computed RGB
// row instead of running a second YUV→RGB kernel. For VUYX,
// α=`0xFF` is semantically correct in both paths (padding byte
// is never real alpha), so Strategy A applies (spec § 8.4).
if let Some(buf) = rgba.as_deref_mut() {
let rgba_row = rgba_plane_row_slice(buf, one_plane_start, one_plane_end, w, h)?;
expand_rgb_to_rgba_row(rgb_row, rgba_row, w);
}
Ok(())
}
}