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
//! Sinker impl for the packed AYUV source format — 8-bit packed YUV 4:4:4
//! with real source alpha, alpha-first channel order.
//!
//! AYUV (FFmpeg `AV_PIX_FMT_AYUV`) packs **four u8 bytes per pixel**
//! (`[A, Y, U, V]`). The A byte (offset 0) is **real source alpha** — not
//! padding. It is the alpha-first channel re-ordering of
//! [`Vuya`](crate::source::Vuya); only the byte positions differ, so this
//! sink mirrors the `Vuya` sink exactly with the AYUV row kernels.
//!
//! Outputs map to the sink's standard channels:
//! - `with_rgb` — packed YUV → RGB 8-bit pipeline; alpha discarded.
//! - `with_rgba` — packed YUV → RGBA 8-bit pipeline; **source α byte
//! is passed through** verbatim from byte 0 of each pixel.
//! - `with_luma` — extracts the Y byte at offset 1 of each pixel directly.
//! - `with_luma_u16` — zero-extends the Y byte to u16.
//! - `with_hsv` — stages u8 RGB into the user's RGB buffer (if attached)
//! or a scratch buffer, then runs `rgb_to_hsv_row`.
//!
//! Alpha semantics are identical to `Vuya` (§ 7.2 / § 7.3): standalone
//! RGBA runs `ayuv_to_rgba_row` directly (source α through the kernel);
//! the RGB + RGBA combo derives RGBA from the computed RGB row then
//! overwrites the α slot from packed source byte 0 via
//! `alpha_extract::copy_alpha_packed_u8x4_at_0`.
use super::{
GeometryOverflow, InsufficientBuffer, MixedSinker, MixedSinkerError, RowIndexOutOfRange,
RowShapeMismatch, RowSlice, check_dimensions_match, check_frozen_alpha_mode,
packed_yuva444_filter_resample, packed_yuva444_resample, rgb_row_buf_or_scratch,
rgba_plane_row_slice,
};
use crate::{
PixelSink,
row::{
ayuv_to_hsv_row, ayuv_to_luma_row, ayuv_to_luma_u16_row, ayuv_to_rgb_row, ayuv_to_rgba_row,
rgb_to_hsv_row,
},
source::{Ayuv, AyuvRow, AyuvSink},
};
impl<'a, R> MixedSinker<'a, Ayuv, R> {
/// Attaches a **`u16`** luma output buffer. Y bytes from the packed AYUV
/// `[A, Y, U, V]` 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 AYUV is the
/// source, the per-pixel alpha byte is **sourced from the A byte of each
/// pixel quadruple** (offset 0) — not forced to `0xFF`.
///
/// Returns `Err(InsufficientRgbaBuffer)` if `buf.len() < width x height x 4`,
/// or `Err(GeometryOverflow)` on 32‑bit targets when the product overflows.
#[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> AyuvSink for MixedSinker<'_, Ayuv, R> {}
impl<R> PixelSink for MixedSinker<'_, Ayuv, R> {
type Input<'r> = AyuvRow<'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 4-channel u8 RGBA colour stream and the
// independent native-Y u16 luma stream, re-arm the alpha-mode snapshot.
// `Ayuv` exposes no u16 colour outputs, so its u16 RGBA streams are
// never created.
if let Some(stream) = self.rgba_stream.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();
}
self.resample_outputs = None;
self.frozen_alpha_mode = Some(self.alpha_mode);
Ok(())
}
fn process(&mut self, row: AyuvRow<'_>) -> Result<(), Self::Error> {
const BITS: u32 = 8;
let w = self.width;
let h = self.height;
let idx = row.row();
let use_simd = self.simd;
// AYUV 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::AyuvPacked,
idx,
packed_expected,
row.packed().len(),
)));
}
if idx >= self.height {
return Err(MixedSinkerError::RowIndexOutOfRange(
RowIndexOutOfRange::new(idx, self.height),
));
}
// Non-identity plan: `Ayuv` is packed 4:4:4 YUV with real source alpha
// (the A byte at offset 0). Route through the packed-YUVA tail at
// `SRC_BITS = 8`, exactly like `Vuya` but with the AYUV channel order.
if self.plan.is_some() {
let alpha_mode = self.alpha_mode;
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,
rgba_scratch,
rgb_scratch,
rgba_scratch_u16,
rgba_color_scratch_u16,
luma_scratch_u16,
plan,
rgba_stream,
rgba_stream_u16,
luma_stream_u16,
rgba_filter_stream,
rgba_filter_stream_u16,
luma_filter_stream_u16,
resample_outputs,
frozen_alpha_mode,
..
} = self;
let plan = plan.as_ref().expect("plan.is_some() checked above");
check_frozen_alpha_mode(*frozen_alpha_mode, alpha_mode, idx)?;
return match plan.kind() {
crate::resample::SpanKind::Area => packed_yuva444_resample::<BITS>(
rgba_stream,
rgba_stream_u16,
luma_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,
alpha_mode,
|dst| ayuv_to_rgba_row(packed, dst, w, matrix, full_range, use_simd),
// `Ayuv` has no u16 colour outputs, so this closure is never called.
|_dst: &mut [u16]| {},
|dst| ayuv_to_luma_u16_row(packed, dst, w, use_simd),
),
crate::resample::SpanKind::Filter if alpha_mode.is_premultiplied() => {
// Premultiplied + filter has no analogue: route to the area tail
// with the filter plan so it returns the typed `UnsupportedFilter`.
packed_yuva444_resample::<BITS>(
rgba_stream,
rgba_stream_u16,
luma_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,
alpha_mode,
|dst| ayuv_to_rgba_row(packed, dst, w, matrix, full_range, use_simd),
|_dst: &mut [u16]| {},
|dst| ayuv_to_luma_u16_row(packed, dst, w, use_simd),
)
}
crate::resample::SpanKind::Filter => packed_yuva444_filter_resample::<BITS, false, false>(
rgba_filter_stream,
rgba_filter_stream_u16,
// Packed `Ayuv` 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 `Ayuv` 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,
|dst| ayuv_to_rgba_row(packed, dst, w, matrix, full_range, use_simd),
// `Ayuv` has no u16 colour outputs, so this closure is never called.
|_dst: &mut [u16]| {},
|dst| ayuv_to_luma_u16_row(packed, dst, w, use_simd),
// u16-luma path, so this u8 de-interleave is never called.
|_dst: &mut [u8]| {},
),
};
}
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
// `ayuv_to_rgba_row` kernel). 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 1 in each AYUV quadruple) directly.
if let Some(buf) = luma.as_deref_mut() {
ayuv_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() {
ayuv_to_luma_u16_row(
packed,
&mut buf[one_plane_start..one_plane_end],
w,
use_simd,
);
}
// ===== u8 RGB / RGBA / HSV 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();
ayuv_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(());
}
// 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.
if need_rgb_kernel {
let rgb_row = rgb_row_buf_or_scratch(
rgb.as_deref_mut(),
rgb_scratch,
one_plane_start,
one_plane_end,
w,
h,
)?;
ayuv_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,
);
}
// RGB + RGBA combo: AYUV carries source α at offset 0. The Strategy A+
// α-extract helpers only cover the offset-3 layout (`Vuya`), so the
// RGBA output is produced by the dedicated `ayuv_to_rgba_row` kernel
// (source α through the kernel) rather than the expand + α-overwrite
// path. The just-computed RGB row already populated the user's RGB /
// HSV outputs.
if want_rgba {
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)?;
ayuv_to_rgba_row(
packed,
rgba_row,
w,
row.matrix(),
row.full_range(),
use_simd,
);
}
}
// Standalone RGBA path — no RGB/HSV requested. Run ayuv_to_rgba_row
// directly from the packed source; source α passes through in the kernel.
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)?;
ayuv_to_rgba_row(
packed,
rgba_row,
w,
row.matrix(),
row.full_range(),
use_simd,
);
}
Ok(())
}
}