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

//! Scalar reference kernels for the V30X packed YUV 4:4:4 10-bit
//! family (FFmpeg `AV_PIX_FMT_V30XLE`). One pixel per 32-bit word;
//! 10-bit V / Y / U channels with 2-bit padding at the **LSB** (unlike
//! V410, which pads at the MSB). 4:4:4 means no chroma deinterleave
//! step — each word yields a complete `(Y, U, V)` triple.
//!
//! Bit layout: `[31:22]=V, [21:12]=Y, [11:2]=U, [1:0]=pad`

use super::*;

/// Extract `(u, y, v)` from one V30X word.
#[cfg_attr(not(tarpaulin), inline(always))]
const fn extract_v30x(word: u32) -> (i32, i32, i32) {
  let u = ((word >> 2) & 0x3FF) as i32;
  let y = ((word >> 12) & 0x3FF) as i32;
  let v = ((word >> 22) & 0x3FF) as i32;
  (u, y, v)
}

// ---- u8 RGB / RGBA output ----------------------------------------------

#[cfg_attr(not(tarpaulin), inline(always))]
pub(crate) fn v30x_to_rgb_or_rgba_row<const ALPHA: bool>(
  packed: &[u32],
  out: &mut [u8],
  width: usize,
  matrix: ColorMatrix,
  full_range: bool,
) {
  debug_assert!(packed.len() >= width, "packed row too short");
  let bpp: usize = if ALPHA { 4 } else { 3 };
  debug_assert!(out.len() >= width * bpp, "out row too short");

  let coeffs = Coefficients::for_matrix(matrix);
  let (y_off, y_scale, c_scale) = range_params_n::<10, 8>(full_range);
  let bias = chroma_bias::<10>();

  for (x, &word) in packed[..width].iter().enumerate() {
    let (u, y, v) = extract_v30x(word);
    let u_d = q15_scale(u - bias, c_scale);
    let v_d = q15_scale(v - bias, c_scale);
    let r_chroma = q15_chroma(coeffs.r_u(), u_d, coeffs.r_v(), v_d);
    let g_chroma = q15_chroma(coeffs.g_u(), u_d, coeffs.g_v(), v_d);
    let b_chroma = q15_chroma(coeffs.b_u(), u_d, coeffs.b_v(), v_d);

    let y_s = q15_scale(y - y_off, y_scale);
    let off = x * bpp;
    out[off] = clamp_u8(y_s + r_chroma);
    out[off + 1] = clamp_u8(y_s + g_chroma);
    out[off + 2] = clamp_u8(y_s + b_chroma);
    if ALPHA {
      out[off + 3] = 0xFF;
    }
  }
}

// ---- u16 RGB / RGBA native-depth output --------------------------------

#[cfg_attr(not(tarpaulin), inline(always))]
pub(crate) fn v30x_to_rgb_u16_or_rgba_u16_row<const ALPHA: bool>(
  packed: &[u32],
  out: &mut [u16],
  width: usize,
  matrix: ColorMatrix,
  full_range: bool,
) {
  debug_assert!(packed.len() >= width, "packed row too short");
  let bpp: usize = if ALPHA { 4 } else { 3 };
  debug_assert!(out.len() >= width * bpp, "out row too short");

  let coeffs = Coefficients::for_matrix(matrix);
  let (y_off, y_scale, c_scale) = range_params_n::<10, 10>(full_range);
  let bias = chroma_bias::<10>();
  let alpha_max: u16 = 0x3FF;
  let out_max: i32 = 0x3FF;

  for (x, &word) in packed[..width].iter().enumerate() {
    let (u, y, v) = extract_v30x(word);
    let u_d = q15_scale(u - bias, c_scale);
    let v_d = q15_scale(v - bias, c_scale);
    let r_chroma = q15_chroma(coeffs.r_u(), u_d, coeffs.r_v(), v_d);
    let g_chroma = q15_chroma(coeffs.g_u(), u_d, coeffs.g_v(), v_d);
    let b_chroma = q15_chroma(coeffs.b_u(), u_d, coeffs.b_v(), v_d);

    let y_s = q15_scale(y - y_off, y_scale);
    let off = x * bpp;
    out[off] = (y_s + r_chroma).clamp(0, out_max) as u16;
    out[off + 1] = (y_s + g_chroma).clamp(0, out_max) as u16;
    out[off + 2] = (y_s + b_chroma).clamp(0, out_max) as u16;
    if ALPHA {
      out[off + 3] = alpha_max;
    }
  }
}

// ---- Luma (u8) — `>> 2` ------------------------------------------------

#[cfg_attr(not(tarpaulin), inline(always))]
pub(crate) fn v30x_to_luma_row(packed: &[u32], out: &mut [u8], width: usize) {
  debug_assert!(packed.len() >= width);
  debug_assert!(out.len() >= width);
  for x in 0..width {
    let y = (packed[x] >> 12) & 0x3FF;
    out[x] = (y >> 2) as u8;
  }
}

// ---- Luma (u16, low-bit-packed at 10-bit) ------------------------------

#[cfg_attr(not(tarpaulin), inline(always))]
pub(crate) fn v30x_to_luma_u16_row(packed: &[u32], out: &mut [u16], width: usize) {
  debug_assert!(packed.len() >= width);
  debug_assert!(out.len() >= width);
  for x in 0..width {
    let y = (packed[x] >> 12) & 0x3FF;
    out[x] = y as u16;
  }
}

#[cfg(all(test, feature = "std"))]
mod tests {
  use super::*;
  use crate::ColorMatrix;

  /// Pack one V30X word from explicit U / Y / V samples.
  fn pack_v30x(u: u32, y: u32, v: u32) -> u32 {
    debug_assert!(u < 1024 && y < 1024 && v < 1024);
    (v << 22) | (y << 12) | (u << 2)
  }

  #[test]
  fn v30x_known_pattern_rgb() {
    // Limited-range BT.709, gray Y=64 (≈ 0 in [0, 255]) with neutral
    // chroma U=V=512. Both pixels should produce ~black [0, 0, 0]
    // before saturation.
    let p = vec![
      pack_v30x(512, 64, 512),
      pack_v30x(512, 64, 512),
      pack_v30x(512, 940, 512), // Y=940 ≈ 255 (limited-range white)
      pack_v30x(512, 940, 512),
    ];
    let mut out = vec![0u8; 4 * 3];
    v30x_to_rgb_or_rgba_row::<false>(&p, &mut out, 4, ColorMatrix::Bt709, false);
    // Two black pixels followed by two white pixels.
    assert_eq!(&out[0..3], &[0u8, 0, 0]);
    assert_eq!(&out[3..6], &[0u8, 0, 0]);
    assert_eq!(&out[6..9], &[255u8, 255, 255]);
    assert_eq!(&out[9..12], &[255u8, 255, 255]);
  }

  #[test]
  fn v30x_known_pattern_rgba_alpha_max() {
    let p = vec![pack_v30x(512, 940, 512)];
    let mut out = vec![0u8; 4];
    v30x_to_rgb_or_rgba_row::<true>(&p, &mut out, 1, ColorMatrix::Bt709, false);
    assert_eq!(out[3], 0xFF);
  }

  #[test]
  fn v30x_luma_extract_u8() {
    let p = vec![
      pack_v30x(0, 0x3FF, 0), // Y = 0x3FF (10-bit max)
      pack_v30x(0, 0x100, 0), // Y = 0x100
    ];
    let mut out = vec![0u8; 2];
    v30x_to_luma_row(&p, &mut out, 2);
    // 0x3FF >> 2 = 0xFF; 0x100 >> 2 = 0x40.
    assert_eq!(&out[..], &[0xFFu8, 0x40]);
  }

  #[test]
  fn v30x_luma_extract_u16_low_bit_packed() {
    let p = vec![pack_v30x(0, 0x3FF, 0), pack_v30x(0, 0x123, 0)];
    let mut out = vec![0u16; 2];
    v30x_to_luma_u16_row(&p, &mut out, 2);
    assert_eq!(&out[..], &[0x3FFu16, 0x123]);
  }

  #[test]
  fn v30x_known_pattern_rgba_u16_alpha_max() {
    let p = vec![pack_v30x(512, 940, 512)];
    let mut out = vec![0u16; 4];
    v30x_to_rgb_u16_or_rgba_u16_row::<true>(&p, &mut out, 1, ColorMatrix::Bt709, false);
    // 10-bit alpha max is 0x3FF (low-bit-packed).
    assert_eq!(out[3], 0x3FF);
  }
}