colconv 0.1.0

SIMD-dispatched color-conversion kernels covering the FFmpeg AVPixelFormat space, with a Sink-based API so consumers pick which derived outputs (RGB / Luma / HSV / custom) they want without paying for the ones they don't.
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

use super::super::*;
use crate::{ColorMatrix, row::scalar};

/// Builds a deterministic pseudo-random Y216-shaped u16 buffer with
/// `width * 2` u16 samples (one quadruple = 4 u16 = 2 pixels). Unlike
/// Y210/Y212, Y216 uses the **full** u16 range — no MSB shift; the
/// active bits fill all 16 bits.
fn pseudo_random_y216(width: usize, seed: usize) -> std::vec::Vec<u16> {
  (0..width * 2)
    .map(|i| ((i.wrapping_mul(seed).wrapping_add(seed * 3)) & 0xFFFF) as u16)
    .collect()
}

fn check_rgb(width: usize, matrix: ColorMatrix, full_range: bool) {
  let p = pseudo_random_y216(width, 0xAA55);
  let mut s = std::vec![0u8; width * 3];
  let mut k = std::vec![0u8; width * 3];
  scalar::y216_to_rgb_or_rgba_row::<false, false>(&p, &mut s, width, matrix, full_range);
  unsafe {
    y216_to_rgb_or_rgba_row::<false, false>(&p, &mut k, width, matrix, full_range);
  }
  assert_eq!(
    s, k,
    "simd128 y216→RGB diverges (width={width}, matrix={matrix:?}, full_range={full_range})"
  );
}

fn check_rgba(width: usize, matrix: ColorMatrix, full_range: bool) {
  let p = pseudo_random_y216(width, 0xAA55);
  let mut s = std::vec![0u8; width * 4];
  let mut k = std::vec![0u8; width * 4];
  scalar::y216_to_rgb_or_rgba_row::<true, false>(&p, &mut s, width, matrix, full_range);
  unsafe {
    y216_to_rgb_or_rgba_row::<true, false>(&p, &mut k, width, matrix, full_range);
  }
  assert_eq!(
    s, k,
    "simd128 y216→RGBA diverges (width={width}, matrix={matrix:?}, full_range={full_range})"
  );
}

fn check_rgb_u16(width: usize, matrix: ColorMatrix, full_range: bool) {
  let p = pseudo_random_y216(width, 0xAA55);
  let mut s = std::vec![0u16; width * 3];
  let mut k = std::vec![0u16; width * 3];
  scalar::y216_to_rgb_u16_or_rgba_u16_row::<false, false>(&p, &mut s, width, matrix, full_range);
  unsafe {
    y216_to_rgb_u16_or_rgba_u16_row::<false, false>(&p, &mut k, width, matrix, full_range);
  }
  assert_eq!(
    s, k,
    "simd128 y216→RGB u16 diverges (width={width}, matrix={matrix:?}, full_range={full_range})"
  );
}

fn check_rgba_u16(width: usize, matrix: ColorMatrix, full_range: bool) {
  let p = pseudo_random_y216(width, 0xAA55);
  let mut s = std::vec![0u16; width * 4];
  let mut k = std::vec![0u16; width * 4];
  scalar::y216_to_rgb_u16_or_rgba_u16_row::<true, false>(&p, &mut s, width, matrix, full_range);
  unsafe {
    y216_to_rgb_u16_or_rgba_u16_row::<true, false>(&p, &mut k, width, matrix, full_range);
  }
  assert_eq!(
    s, k,
    "simd128 y216→RGBA u16 diverges (width={width}, matrix={matrix:?}, full_range={full_range})"
  );
}

fn check_luma(width: usize) {
  let p = pseudo_random_y216(width, 0xC001);
  let mut s = std::vec![0u8; width];
  let mut k = std::vec![0u8; width];
  scalar::y216_to_luma_row::<false>(&p, &mut s, width);
  unsafe {
    y216_to_luma_row::<false>(&p, &mut k, width);
  }
  assert_eq!(s, k, "simd128 y216→luma u8 diverges (width={width})");
}

fn check_luma_u16(width: usize) {
  let p = pseudo_random_y216(width, 0xC001);
  let mut s = std::vec![0u16; width];
  let mut k = std::vec![0u16; width];
  scalar::y216_to_luma_u16_row::<false>(&p, &mut s, width);
  unsafe {
    y216_to_luma_u16_row::<false>(&p, &mut k, width);
  }
  assert_eq!(s, k, "simd128 y216→luma u16 diverges (width={width})");
}

// wasm has no runtime CPU detection — `simd128` is a compile-time
// feature, so no `is_*_feature_detected!` early-return guard. The
// `#[cfg_attr(miri, ignore)]` attribute is included for parity with
// other backends; miri does not currently target wasm32-wasip1.

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn wasm_simd128_y216_rgb_matches_scalar_all_matrices() {
  for m in [
    ColorMatrix::Bt601,
    ColorMatrix::Bt709,
    ColorMatrix::Bt2020Ncl,
    ColorMatrix::Smpte240m,
    ColorMatrix::Fcc,
    ColorMatrix::YCgCo,
  ] {
    for full in [true, false] {
      check_rgb(16, m, full);
      check_rgba(16, m, full);
      check_rgb_u16(16, m, full);
      check_rgba_u16(16, m, full);
    }
  }
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn wasm_simd128_y216_matches_scalar_widths() {
  for w in [2usize, 4, 14, 16, 18, 30, 32, 34, 62, 64, 66, 1920, 1922] {
    check_rgb(w, ColorMatrix::Bt709, false);
    check_rgba(w, ColorMatrix::Bt709, true);
    check_rgb_u16(w, ColorMatrix::Bt2020Ncl, true);
    check_rgba_u16(w, ColorMatrix::Bt601, false);
  }
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn wasm_simd128_y216_luma_matches_scalar_widths() {
  for w in [2usize, 4, 14, 16, 18, 30, 32, 34, 62, 64, 66, 1920, 1922] {
    check_luma(w);
    check_luma_u16(w);
  }
}

/// Multi-channel lane-order regression — encodes pixel index in
/// BOTH Y AND U so we catch per-channel asymmetric mask bugs that
/// a Y-only test would miss. Pattern from Ship 12d AYUV64 backport.
///
/// Y216 is 16-bit native (no MSB shift) — exercises the i64 chroma
/// path at u16 RGB output.
///
/// Y216 layout: YUYV-shape u16x2 `[Y0, U, Y1, V]` per 2 pixels.
/// - `Y[n] = (n + 1) as u16` (16-bit native, no shift)
/// - `U[k] = (2k + 1) as u16` (one U per pair)
/// - `V = 0x8000` (neutral u16 midpoint)
///
/// wasm u16-RGB path threshold: 8 px/iter. W=16 covers 2 full iterations.
/// No runtime feature guard — simd128 is a compile-time feature on wasm.
fn build_y216_packed_y_n_plus_1_u_2k_plus_1_v_neutral(width: usize) -> std::vec::Vec<u16> {
  let mut packed = std::vec![0u16; width * 2];
  for k in 0..(width / 2) {
    let y0 = (2 * k as u16) + 1;
    let y1 = (2 * k as u16) + 2;
    let u = (2 * k as u16) + 1;
    packed[k * 4] = y0;
    packed[k * 4 + 1] = u;
    packed[k * 4 + 2] = y1;
    packed[k * 4 + 3] = 0x8000; // V neutral
  }
  packed
}

#[test]
#[cfg_attr(
  miri,
  ignore = "SIMD-dispatched row kernels use intrinsics unsupported by Miri"
)]
fn wasm_simd128_y216_lane_order_per_pixel_y_and_u() {
  const W: usize = 16;
  let packed = build_y216_packed_y_n_plus_1_u_2k_plus_1_v_neutral(W);

  // Part 1: Luma natural-order at u16
  let mut luma_u16 = std::vec![0u16; W];
  unsafe {
    y216_to_luma_u16_row::<false>(&packed, &mut luma_u16, W);
  }
  let expected_luma: std::vec::Vec<u16> = (1..=W as u16).collect();
  assert_eq!(
    luma_u16, expected_luma,
    "wasm_simd128 y216 luma_u16 reorder bug"
  );

  // Part 2: SIMD vs scalar parity at u16 RGB (i64 chroma path)
  let mut simd_rgb = std::vec![0u16; W * 3];
  let mut scalar_rgb = std::vec![0u16; W * 3];
  unsafe {
    y216_to_rgb_u16_or_rgba_u16_row::<false, false>(
      &packed,
      &mut simd_rgb,
      W,
      ColorMatrix::Bt709,
      false,
    );
  }
  scalar::y216_to_rgb_u16_or_rgba_u16_row::<false, false>(
    &packed,
    &mut scalar_rgb,
    W,
    ColorMatrix::Bt709,
    false,
  );
  assert_eq!(
    simd_rgb, scalar_rgb,
    "wasm_simd128 y216 SIMD vs scalar diverges (u16 RGB, i64 chroma)"
  );
}