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

//! NEON kernels for planar GBR sources (Tier 10).
//!
//! NEON's `vst3q_u8` / `vst4q_u8` make the planar→packed interleave
//! nearly free — load three (or four) channel vectors and store them
//! interleaved in one shot. Both kernels process 16 pixels per
//! iteration with a scalar tail.

use core::arch::aarch64::*;

use crate::row::scalar;

/// Interleaves three planar G/B/R rows into packed `R, G, B`. Loads
/// each plane as a `uint8x16_t` (16 bytes), assembles a
/// `uint8x16x3_t` in **R, G, B** order, and stores via `vst3q_u8` —
/// which interleaves the three vectors into one packed RGB stream
/// without per-pixel byte arithmetic.
///
/// # Safety
///
/// 1. NEON must be available (caller obligation).
/// 2. `g.len()`, `b.len()`, `r.len()` ≥ `width`.
/// 3. `rgb_out.len()` ≥ `3 * width`.
#[inline]
#[target_feature(enable = "neon")]
pub(crate) unsafe fn gbr_to_rgb_row(
  g: &[u8],
  b: &[u8],
  r: &[u8],
  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");

  // SAFETY: NEON is available per caller obligation. All pointer adds
  // are bounded by the `while x + 16 <= width` condition.
  unsafe {
    let mut x = 0usize;
    while x + 16 <= width {
      let g_v = vld1q_u8(g.as_ptr().add(x));
      let b_v = vld1q_u8(b.as_ptr().add(x));
      let r_v = vld1q_u8(r.as_ptr().add(x));
      let triple = uint8x16x3_t(r_v, g_v, b_v);
      vst3q_u8(rgb_out.as_mut_ptr().add(x * 3), triple);
      x += 16;
    }
    if x < width {
      scalar::gbr_to_rgb_row(
        &g[x..width],
        &b[x..width],
        &r[x..width],
        &mut rgb_out[x * 3..width * 3],
        width - x,
      );
    }
  }
}

/// Interleaves four planar G/B/R/A rows into packed `R, G, B, A`.
/// Same shape as [`gbr_to_rgb_row`] with `vst4q_u8` and an extra
/// alpha lane.
///
/// # Safety
///
/// 1. NEON must be available.
/// 2. `g.len()`, `b.len()`, `r.len()`, `a.len()` ≥ `width`.
/// 3. `rgba_out.len()` ≥ `4 * width`.
#[inline]
#[target_feature(enable = "neon")]
pub(crate) unsafe fn gbra_to_rgba_row(
  g: &[u8],
  b: &[u8],
  r: &[u8],
  a: &[u8],
  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!(a.len() >= width, "a row too short");
  debug_assert!(rgba_out.len() >= width * 4, "rgba_out row too short");

  // SAFETY: see `gbr_to_rgb_row`.
  unsafe {
    let mut x = 0usize;
    while x + 16 <= width {
      let g_v = vld1q_u8(g.as_ptr().add(x));
      let b_v = vld1q_u8(b.as_ptr().add(x));
      let r_v = vld1q_u8(r.as_ptr().add(x));
      let a_v = vld1q_u8(a.as_ptr().add(x));
      let quad = uint8x16x4_t(r_v, g_v, b_v, a_v);
      vst4q_u8(rgba_out.as_mut_ptr().add(x * 4), quad);
      x += 16;
    }
    if x < width {
      scalar::gbra_to_rgba_row(
        &g[x..width],
        &b[x..width],
        &r[x..width],
        &a[x..width],
        &mut rgba_out[x * 4..width * 4],
        width - x,
      );
    }
  }
}

/// Interleaves three planar G/B/R rows into packed `R, G, B, A` with
/// constant `α = 0xFF`. Used by `Gbrp` (no alpha plane) for the
/// `with_rgba` standalone path.
///
/// # Safety
///
/// 1. NEON must be available.
/// 2. `g.len()`, `b.len()`, `r.len()` ≥ `width`.
/// 3. `rgba_out.len()` ≥ `4 * width`.
#[inline]
#[target_feature(enable = "neon")]
pub(crate) unsafe fn gbr_to_rgba_opaque_row(
  g: &[u8],
  b: &[u8],
  r: &[u8],
  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");

  // SAFETY: see `gbr_to_rgb_row`.
  unsafe {
    let opaque = vdupq_n_u8(0xFF);
    let mut x = 0usize;
    while x + 16 <= width {
      let g_v = vld1q_u8(g.as_ptr().add(x));
      let b_v = vld1q_u8(b.as_ptr().add(x));
      let r_v = vld1q_u8(r.as_ptr().add(x));
      let quad = uint8x16x4_t(r_v, g_v, b_v, opaque);
      vst4q_u8(rgba_out.as_mut_ptr().add(x * 4), quad);
      x += 16;
    }
    if x < width {
      scalar::gbr_to_rgba_opaque_row(
        &g[x..width],
        &b[x..width],
        &r[x..width],
        &mut rgba_out[x * 4..width * 4],
        width - x,
      );
    }
  }
}