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
//! AVX2 kernels for MSB-aligned high-bit planar GBR sources
//! (`AV_PIX_FMT_GBRP10MSB{LE,BE}` / `AV_PIX_FMT_GBRP12MSB{LE,BE}`).
//!
//! The MSB-aligned twins of [`planar_gbr_high_bit`](super::planar_gbr_high_bit).
//! The active sample is in the high `BITS` bits of each `u16`, so recovery is
//! `_mm256_srl_epi16(v, align)` (variable right-shift by `16 - BITS`, a 128-bit
//! count vector shared by the 256- and 128-bit shift forms) rather than the
//! low-bit family's `_mm256_and_si256` mask. These formats have no alpha
//! plane, so only the 3-plane kernels exist.
//!
//! Lane width: 16 pixels per iteration (16 × u16 per `__m256i`), draining an
//! 8-pixel block then a scalar tail.

use super::{endian::load_endian_u16x16, *};
use crate::row::arch::x86_sse41::endian::load_endian_u16x8;

// ---- u8 output, 3-channel (RGB) -----------------------------------------

/// AVX2 MSB-aligned G/B/R planar → packed `R, G, B` **bytes**.
/// Recovers each sample (`>> (16 - BITS)`), downshifts by `BITS - 8`, packs
/// to u8.
///
/// When `BE = true` each source u16 element is byte-swapped on load.
///
/// # Safety
///
/// 1. AVX2 must be available (caller obligation).
/// 2. `g.len()`, `b.len()`, `r.len()` ≥ `width`.
/// 3. `rgb_out.len()` ≥ `3 * width`.
#[inline]
#[target_feature(enable = "avx2")]
pub(crate) unsafe fn gbr_to_rgb_msb_row<const BITS: u32, const BE: bool>(
  g: &[u16],
  b: &[u16],
  r: &[u16],
  rgb_out: &mut [u8],
  width: usize,
) {
  debug_assert!(g.len() >= width, "g row too short");
  debug_assert!(b.len() >= width, "b row too short");
  debug_assert!(r.len() >= width, "r row too short");
  debug_assert!(rgb_out.len() >= width * 3, "rgb_out row too short");

  unsafe {
    let align_count = _mm_cvtsi32_si128((16 - BITS) as i32);
    let shr_count = _mm_cvtsi32_si128((BITS - 8) as i32);
    let zero256 = _mm256_setzero_si256();

    let mut x = 0usize;
    while x + 16 <= width {
      let r_v = _mm256_srl_epi16(
        load_endian_u16x16::<BE>(r.as_ptr().add(x).cast()),
        align_count,
      );
      let g_v = _mm256_srl_epi16(
        load_endian_u16x16::<BE>(g.as_ptr().add(x).cast()),
        align_count,
      );
      let b_v = _mm256_srl_epi16(
        load_endian_u16x16::<BE>(b.as_ptr().add(x).cast()),
        align_count,
      );

      let r_sh = _mm256_srl_epi16(r_v, shr_count);
      let g_sh = _mm256_srl_epi16(g_v, shr_count);
      let b_sh = _mm256_srl_epi16(b_v, shr_count);

      let r_packed = _mm256_permute4x64_epi64::<0xD8>(_mm256_packus_epi16(r_sh, zero256));
      let g_packed = _mm256_permute4x64_epi64::<0xD8>(_mm256_packus_epi16(g_sh, zero256));
      let b_packed = _mm256_permute4x64_epi64::<0xD8>(_mm256_packus_epi16(b_sh, zero256));

      let r_u8 = _mm256_castsi256_si128(r_packed);
      let g_u8 = _mm256_castsi256_si128(g_packed);
      let b_u8 = _mm256_castsi256_si128(b_packed);

      write_rgb_16(r_u8, g_u8, b_u8, rgb_out.as_mut_ptr().add(x * 3));

      x += 16;
    }
    if x + 8 <= width {
      let zero = _mm_setzero_si128();
      let r_v = _mm_srl_epi16(
        load_endian_u16x8::<BE>(r.as_ptr().add(x).cast()),
        align_count,
      );
      let g_v = _mm_srl_epi16(
        load_endian_u16x8::<BE>(g.as_ptr().add(x).cast()),
        align_count,
      );
      let b_v = _mm_srl_epi16(
        load_endian_u16x8::<BE>(b.as_ptr().add(x).cast()),
        align_count,
      );
      let r_sh = _mm_srl_epi16(r_v, shr_count);
      let g_sh = _mm_srl_epi16(g_v, shr_count);
      let b_sh = _mm_srl_epi16(b_v, shr_count);
      let r_u8 = _mm_packus_epi16(r_sh, zero);
      let g_u8 = _mm_packus_epi16(g_sh, zero);
      let b_u8 = _mm_packus_epi16(b_sh, zero);
      let mut tmp = [0u8; 48];
      write_rgb_16(r_u8, g_u8, b_u8, tmp.as_mut_ptr());
      core::ptr::copy_nonoverlapping(tmp.as_ptr(), rgb_out.as_mut_ptr().add(x * 3), 24);
      x += 8;
    }
    if x < width {
      scalar::gbr_to_rgb_msb_row::<BITS, BE>(
        &g[x..width],
        &b[x..width],
        &r[x..width],
        &mut rgb_out[x * 3..width * 3],
        width - x,
      );
    }
  }
}

// ---- u8 output, 4-channel RGBA with constant opaque alpha ----------------

/// AVX2 MSB-aligned G/B/R planar → packed `R, G, B, A` **bytes** with constant
/// opaque alpha (`0xFF`).
///
/// When `BE = true` each source u16 element is byte-swapped on load.
///
/// # Safety
///
/// 1. AVX2 must be available (caller obligation).
/// 2. `g.len()`, `b.len()`, `r.len()` ≥ `width`.
/// 3. `rgba_out.len()` ≥ `4 * width`.
#[inline]
#[target_feature(enable = "avx2")]
pub(crate) unsafe fn gbr_to_rgba_opaque_msb_row<const BITS: u32, const BE: bool>(
  g: &[u16],
  b: &[u16],
  r: &[u16],
  rgba_out: &mut [u8],
  width: usize,
) {
  debug_assert!(g.len() >= width, "g row too short");
  debug_assert!(b.len() >= width, "b row too short");
  debug_assert!(r.len() >= width, "r row too short");
  debug_assert!(rgba_out.len() >= width * 4, "rgba_out row too short");

  unsafe {
    let align_count = _mm_cvtsi32_si128((16 - BITS) as i32);
    let shr_count = _mm_cvtsi32_si128((BITS - 8) as i32);
    let zero256 = _mm256_setzero_si256();
    let zero = _mm_setzero_si128();
    let opaque_u8 = _mm_set1_epi8(-1i8);

    let mut x = 0usize;
    while x + 16 <= width {
      let r_v = _mm256_srl_epi16(
        load_endian_u16x16::<BE>(r.as_ptr().add(x).cast()),
        align_count,
      );
      let g_v = _mm256_srl_epi16(
        load_endian_u16x16::<BE>(g.as_ptr().add(x).cast()),
        align_count,
      );
      let b_v = _mm256_srl_epi16(
        load_endian_u16x16::<BE>(b.as_ptr().add(x).cast()),
        align_count,
      );

      let r_sh = _mm256_srl_epi16(r_v, shr_count);
      let g_sh = _mm256_srl_epi16(g_v, shr_count);
      let b_sh = _mm256_srl_epi16(b_v, shr_count);

      let r_packed = _mm256_permute4x64_epi64::<0xD8>(_mm256_packus_epi16(r_sh, zero256));
      let g_packed = _mm256_permute4x64_epi64::<0xD8>(_mm256_packus_epi16(g_sh, zero256));
      let b_packed = _mm256_permute4x64_epi64::<0xD8>(_mm256_packus_epi16(b_sh, zero256));

      let r_u8 = _mm256_castsi256_si128(r_packed);
      let g_u8 = _mm256_castsi256_si128(g_packed);
      let b_u8 = _mm256_castsi256_si128(b_packed);

      write_rgba_16(
        r_u8,
        g_u8,
        b_u8,
        opaque_u8,
        rgba_out.as_mut_ptr().add(x * 4),
      );

      x += 16;
    }
    if x + 8 <= width {
      let r_v = _mm_srl_epi16(
        load_endian_u16x8::<BE>(r.as_ptr().add(x).cast()),
        align_count,
      );
      let g_v = _mm_srl_epi16(
        load_endian_u16x8::<BE>(g.as_ptr().add(x).cast()),
        align_count,
      );
      let b_v = _mm_srl_epi16(
        load_endian_u16x8::<BE>(b.as_ptr().add(x).cast()),
        align_count,
      );
      let r_sh = _mm_srl_epi16(r_v, shr_count);
      let g_sh = _mm_srl_epi16(g_v, shr_count);
      let b_sh = _mm_srl_epi16(b_v, shr_count);
      let r_u8 = _mm_packus_epi16(r_sh, zero);
      let g_u8 = _mm_packus_epi16(g_sh, zero);
      let b_u8 = _mm_packus_epi16(b_sh, zero);
      let mut tmp = [0u8; 64];
      write_rgba_16(r_u8, g_u8, b_u8, opaque_u8, tmp.as_mut_ptr());
      core::ptr::copy_nonoverlapping(tmp.as_ptr(), rgba_out.as_mut_ptr().add(x * 4), 32);
      x += 8;
    }
    if x < width {
      scalar::gbr_to_rgba_opaque_msb_row::<BITS, BE>(
        &g[x..width],
        &b[x..width],
        &r[x..width],
        &mut rgba_out[x * 4..width * 4],
        width - x,
      );
    }
  }
}

// ---- u16 output, 3-channel (RGB) ----------------------------------------

/// AVX2 MSB-aligned G/B/R planar → packed `R, G, B` **u16** samples.
/// Recovers each sample (`>> (16 - BITS)`), reorders G/B/R → R/G/B. Processes
/// 16 pixels per outer loop via two 8-pixel `write_rgb_u16_8` calls.
///
/// When `BE = true` each source u16 element is byte-swapped on load.
///
/// # Safety
///
/// 1. AVX2 must be available (caller obligation).
/// 2. `g.len()`, `b.len()`, `r.len()` ≥ `width`.
/// 3. `rgb_u16_out.len()` ≥ `3 * width`.
#[inline]
#[target_feature(enable = "avx2")]
pub(crate) unsafe fn gbr_to_rgb_u16_msb_row<const BITS: u32, const BE: bool>(
  g: &[u16],
  b: &[u16],
  r: &[u16],
  rgb_u16_out: &mut [u16],
  width: usize,
) {
  debug_assert!(g.len() >= width, "g row too short");
  debug_assert!(b.len() >= width, "b row too short");
  debug_assert!(r.len() >= width, "r row too short");
  debug_assert!(rgb_u16_out.len() >= width * 3, "rgb_u16_out row too short");

  unsafe {
    let align_count = _mm_cvtsi32_si128((16 - BITS) as i32);
    let mut x = 0usize;
    while x + 16 <= width {
      let r_lo = _mm_srl_epi16(
        load_endian_u16x8::<BE>(r.as_ptr().add(x).cast()),
        align_count,
      );
      let g_lo = _mm_srl_epi16(
        load_endian_u16x8::<BE>(g.as_ptr().add(x).cast()),
        align_count,
      );
      let b_lo = _mm_srl_epi16(
        load_endian_u16x8::<BE>(b.as_ptr().add(x).cast()),
        align_count,
      );
      write_rgb_u16_8(r_lo, g_lo, b_lo, rgb_u16_out.as_mut_ptr().add(x * 3));

      let r_hi = _mm_srl_epi16(
        load_endian_u16x8::<BE>(r.as_ptr().add(x + 8).cast()),
        align_count,
      );
      let g_hi = _mm_srl_epi16(
        load_endian_u16x8::<BE>(g.as_ptr().add(x + 8).cast()),
        align_count,
      );
      let b_hi = _mm_srl_epi16(
        load_endian_u16x8::<BE>(b.as_ptr().add(x + 8).cast()),
        align_count,
      );
      write_rgb_u16_8(r_hi, g_hi, b_hi, rgb_u16_out.as_mut_ptr().add((x + 8) * 3));

      x += 16;
    }
    if x + 8 <= width {
      let r_v = _mm_srl_epi16(
        load_endian_u16x8::<BE>(r.as_ptr().add(x).cast()),
        align_count,
      );
      let g_v = _mm_srl_epi16(
        load_endian_u16x8::<BE>(g.as_ptr().add(x).cast()),
        align_count,
      );
      let b_v = _mm_srl_epi16(
        load_endian_u16x8::<BE>(b.as_ptr().add(x).cast()),
        align_count,
      );
      write_rgb_u16_8(r_v, g_v, b_v, rgb_u16_out.as_mut_ptr().add(x * 3));
      x += 8;
    }
    if x < width {
      scalar::gbr_to_rgb_u16_msb_row::<BITS, BE>(
        &g[x..width],
        &b[x..width],
        &r[x..width],
        &mut rgb_u16_out[x * 3..width * 3],
        width - x,
      );
    }
  }
}

// ---- u16 output, 4-channel RGBA with constant opaque alpha ---------------

/// AVX2 MSB-aligned G/B/R planar → packed `R, G, B, A` **u16** samples with
/// constant opaque alpha `(1 << BITS) - 1`.
///
/// When `BE = true` each source u16 element is byte-swapped on load.
///
/// # Safety
///
/// 1. AVX2 must be available (caller obligation).
/// 2. `g.len()`, `b.len()`, `r.len()` ≥ `width`.
/// 3. `rgba_u16_out.len()` ≥ `4 * width`.
#[inline]
#[target_feature(enable = "avx2")]
pub(crate) unsafe fn gbr_to_rgba_opaque_u16_msb_row<const BITS: u32, const BE: bool>(
  g: &[u16],
  b: &[u16],
  r: &[u16],
  rgba_u16_out: &mut [u16],
  width: usize,
) {
  debug_assert!(g.len() >= width, "g row too short");
  debug_assert!(b.len() >= width, "b row too short");
  debug_assert!(r.len() >= width, "r row too short");
  debug_assert!(
    rgba_u16_out.len() >= width * 4,
    "rgba_u16_out row too short"
  );

  unsafe {
    let align_count = _mm_cvtsi32_si128((16 - BITS) as i32);
    let opaque = _mm_set1_epi16(((1u32 << BITS) - 1) as u16 as i16);

    let mut x = 0usize;
    while x + 16 <= width {
      let r_lo = _mm_srl_epi16(
        load_endian_u16x8::<BE>(r.as_ptr().add(x).cast()),
        align_count,
      );
      let g_lo = _mm_srl_epi16(
        load_endian_u16x8::<BE>(g.as_ptr().add(x).cast()),
        align_count,
      );
      let b_lo = _mm_srl_epi16(
        load_endian_u16x8::<BE>(b.as_ptr().add(x).cast()),
        align_count,
      );
      write_rgba_u16_8(
        r_lo,
        g_lo,
        b_lo,
        opaque,
        rgba_u16_out.as_mut_ptr().add(x * 4),
      );

      let r_hi = _mm_srl_epi16(
        load_endian_u16x8::<BE>(r.as_ptr().add(x + 8).cast()),
        align_count,
      );
      let g_hi = _mm_srl_epi16(
        load_endian_u16x8::<BE>(g.as_ptr().add(x + 8).cast()),
        align_count,
      );
      let b_hi = _mm_srl_epi16(
        load_endian_u16x8::<BE>(b.as_ptr().add(x + 8).cast()),
        align_count,
      );
      write_rgba_u16_8(
        r_hi,
        g_hi,
        b_hi,
        opaque,
        rgba_u16_out.as_mut_ptr().add((x + 8) * 4),
      );

      x += 16;
    }
    if x + 8 <= width {
      let r_v = _mm_srl_epi16(
        load_endian_u16x8::<BE>(r.as_ptr().add(x).cast()),
        align_count,
      );
      let g_v = _mm_srl_epi16(
        load_endian_u16x8::<BE>(g.as_ptr().add(x).cast()),
        align_count,
      );
      let b_v = _mm_srl_epi16(
        load_endian_u16x8::<BE>(b.as_ptr().add(x).cast()),
        align_count,
      );
      write_rgba_u16_8(r_v, g_v, b_v, opaque, rgba_u16_out.as_mut_ptr().add(x * 4));
      x += 8;
    }
    if x < width {
      scalar::gbr_to_rgba_opaque_u16_msb_row::<BITS, BE>(
        &g[x..width],
        &b[x..width],
        &r[x..width],
        &mut rgba_u16_out[x * 4..width * 4],
        width - x,
      );
    }
  }
}