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

use super::super::*;

// ---- Tier 3 packed YUV 4:2:2 AVX-512 scalar-equivalence -----------

fn packed_yuv422_buffer(width: usize, seed: usize) -> std::vec::Vec<u8> {
  (0..width * 2)
    .map(|i| ((i.wrapping_mul(seed).wrapping_add(seed * 3)) & 0xFF) as u8)
    .collect()
}

fn check_yuyv422_rgb(width: usize, matrix: ColorMatrix, full_range: bool) {
  if !std::arch::is_x86_feature_detected!("avx512bw") {
    return;
  }
  let p = packed_yuv422_buffer(width, 37);
  let mut s = std::vec![0u8; width * 3];
  let mut k = std::vec![0u8; width * 3];
  scalar::yuyv422_to_rgb_row(&p, &mut s, width, matrix, full_range);
  unsafe {
    yuyv422_to_rgb_row(&p, &mut k, width, matrix, full_range);
  }
  assert_eq!(s, k, "AVX-512 yuyv422→RGB diverges (width={width})");
}

fn check_yuyv422_rgba(width: usize, matrix: ColorMatrix, full_range: bool) {
  if !std::arch::is_x86_feature_detected!("avx512bw") {
    return;
  }
  let p = packed_yuv422_buffer(width, 37);
  let mut s = std::vec![0u8; width * 4];
  let mut k = std::vec![0u8; width * 4];
  scalar::yuyv422_to_rgba_row(&p, &mut s, width, matrix, full_range);
  unsafe {
    yuyv422_to_rgba_row(&p, &mut k, width, matrix, full_range);
  }
  assert_eq!(s, k, "AVX-512 yuyv422→RGBA diverges (width={width})");
}

fn check_uyvy422_rgb(width: usize, matrix: ColorMatrix, full_range: bool) {
  if !std::arch::is_x86_feature_detected!("avx512bw") {
    return;
  }
  let p = packed_yuv422_buffer(width, 37);
  let mut s = std::vec![0u8; width * 3];
  let mut k = std::vec![0u8; width * 3];
  scalar::uyvy422_to_rgb_row(&p, &mut s, width, matrix, full_range);
  unsafe {
    uyvy422_to_rgb_row(&p, &mut k, width, matrix, full_range);
  }
  assert_eq!(s, k, "AVX-512 uyvy422→RGB diverges (width={width})");
}

fn check_uyvy422_rgba(width: usize, matrix: ColorMatrix, full_range: bool) {
  if !std::arch::is_x86_feature_detected!("avx512bw") {
    return;
  }
  let p = packed_yuv422_buffer(width, 37);
  let mut s = std::vec![0u8; width * 4];
  let mut k = std::vec![0u8; width * 4];
  scalar::uyvy422_to_rgba_row(&p, &mut s, width, matrix, full_range);
  unsafe {
    uyvy422_to_rgba_row(&p, &mut k, width, matrix, full_range);
  }
  assert_eq!(s, k, "AVX-512 uyvy422→RGBA diverges (width={width})");
}

fn check_yvyu422_rgb(width: usize, matrix: ColorMatrix, full_range: bool) {
  if !std::arch::is_x86_feature_detected!("avx512bw") {
    return;
  }
  let p = packed_yuv422_buffer(width, 37);
  let mut s = std::vec![0u8; width * 3];
  let mut k = std::vec![0u8; width * 3];
  scalar::yvyu422_to_rgb_row(&p, &mut s, width, matrix, full_range);
  unsafe {
    yvyu422_to_rgb_row(&p, &mut k, width, matrix, full_range);
  }
  assert_eq!(s, k, "AVX-512 yvyu422→RGB diverges (width={width})");
}

fn check_yvyu422_rgba(width: usize, matrix: ColorMatrix, full_range: bool) {
  if !std::arch::is_x86_feature_detected!("avx512bw") {
    return;
  }
  let p = packed_yuv422_buffer(width, 37);
  let mut s = std::vec![0u8; width * 4];
  let mut k = std::vec![0u8; width * 4];
  scalar::yvyu422_to_rgba_row(&p, &mut s, width, matrix, full_range);
  unsafe {
    yvyu422_to_rgba_row(&p, &mut k, width, matrix, full_range);
  }
  assert_eq!(s, k, "AVX-512 yvyu422→RGBA diverges (width={width})");
}

#[test]
fn avx512_packed_yuv422_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_yuyv422_rgb(64, m, full);
      check_yuyv422_rgba(64, m, full);
      check_uyvy422_rgb(64, m, full);
      check_uyvy422_rgba(64, m, full);
      check_yvyu422_rgb(64, m, full);
      check_yvyu422_rgba(64, m, full);
    }
  }
}

#[test]
fn avx512_packed_yuv422_matches_scalar_widths() {
  // 64-pixel block boundary; tail-width sweep covers main loop +
  // scalar fallback at every block-size boundary including
  // 16/32/64-pixel splits where lower-tier code paths would land.
  for w in [
    2usize, 4, 14, 16, 18, 30, 32, 34, 62, 64, 66, 126, 128, 130, 1920, 1922,
  ] {
    check_yuyv422_rgb(w, ColorMatrix::Bt709, false);
    check_yuyv422_rgba(w, ColorMatrix::Bt709, true);
    check_uyvy422_rgb(w, ColorMatrix::Bt2020Ncl, true);
    check_uyvy422_rgba(w, ColorMatrix::Bt601, false);
    check_yvyu422_rgb(w, ColorMatrix::Smpte240m, false);
    check_yvyu422_rgba(w, ColorMatrix::YCgCo, true);
  }
}

fn check_luma(width: usize, format: char) {
  if !std::arch::is_x86_feature_detected!("avx512bw") {
    return;
  }
  let p = packed_yuv422_buffer(width, 53);
  let mut s = std::vec![0u8; width];
  let mut k = std::vec![0u8; width];
  match format {
    'y' => unsafe {
      scalar::yuyv422_to_luma_row(&p, &mut s, width);
      yuyv422_to_luma_row(&p, &mut k, width);
    },
    'u' => unsafe {
      scalar::uyvy422_to_luma_row(&p, &mut s, width);
      uyvy422_to_luma_row(&p, &mut k, width);
    },
    'v' => unsafe {
      scalar::yvyu422_to_luma_row(&p, &mut s, width);
      yvyu422_to_luma_row(&p, &mut k, width);
    },
    _ => unreachable!(),
  }
  assert_eq!(
    s, k,
    "AVX-512 luma diverges (format={format}, width={width})"
  );
}

#[test]
fn avx512_packed_yuv422_luma_matches_scalar_widths() {
  for w in [
    2usize, 4, 14, 16, 30, 32, 34, 62, 64, 66, 126, 128, 130, 1920, 1922,
  ] {
    check_luma(w, 'y');
    check_luma(w, 'u');
    check_luma(w, 'v');
  }
}

// ===== u16 luma extraction (3 formats; YUYV / UYVY / YVYU) =====

fn check_luma_u16(width: usize, format: char) {
  if !std::arch::is_x86_feature_detected!("avx512bw") {
    return;
  }
  let p = packed_yuv422_buffer(width, 71);
  let mut s = std::vec![0u16; width];
  let mut k = std::vec![0u16; width];
  match format {
    'y' => unsafe {
      scalar::yuyv422_to_luma_u16_row(&p, &mut s, width);
      yuyv422_to_luma_u16_row(&p, &mut k, width);
    },
    'u' => unsafe {
      scalar::uyvy422_to_luma_u16_row(&p, &mut s, width);
      uyvy422_to_luma_u16_row(&p, &mut k, width);
    },
    'v' => unsafe {
      scalar::yvyu422_to_luma_u16_row(&p, &mut s, width);
      yvyu422_to_luma_u16_row(&p, &mut k, width);
    },
    _ => unreachable!(),
  }
  assert_eq!(
    s, k,
    "AVX-512 luma_u16 diverges (format={format}, width={width})"
  );
}

#[test]
fn avx512_packed_yuv422_luma_u16_matches_scalar_widths() {
  // Sweep covers the 64-px AVX-512 main loop boundary plus tails on
  // both sides (63 / 65 / 95 / 127 / 129).
  for w in [
    1usize, 2, 4, 8, 14, 16, 30, 32, 34, 62, 63, 64, 65, 66, 95, 126, 127, 128, 129, 130, 1920,
    1922,
  ] {
    check_luma_u16(w, 'y');
    check_luma_u16(w, 'u');
    check_luma_u16(w, 'v');
  }
}