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
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

//! Tests for `crate::row::scalar::grayf32`.

use super::*;

/// Re-encode a host-native f32 slice as LE-encoded byte storage (each element
/// stored with LE u32-bit byte layout). Kernels called with `BE = false`
/// recover the intended host-native value via `u32::from_le` on both LE
/// (no-op) and BE (byte-swap) hosts.
fn as_le_f32(host: &[f32]) -> std::vec::Vec<f32> {
  host
    .iter()
    .map(|v| f32::from_bits(u32::from_ne_bytes(v.to_bits().to_le_bytes())))
    .collect()
}

// ---- grayf32_to_rgb_row --------------------------------------------------

#[test]
fn grayf32_to_rgb_zero() {
  let plane = [0.0f32];
  let mut out = [0xFFu8; 3];
  grayf32_to_rgb_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [0, 0, 0]);
}

#[test]
fn grayf32_to_rgb_max() {
  let plane = as_le_f32(&[1.0f32]);
  let mut out = [0u8; 3];
  grayf32_to_rgb_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [255, 255, 255]);
}

#[test]
fn grayf32_to_rgb_mid() {
  // Mid-gray Y=0.5 with round-half-up:
  //   0.5 * 255      = 127.5  (pure truncation would give 127)
  //   127.5 + 0.5    = 128.0  (round-half-up adds 0.5 first)
  //   trunc(128.0)   = 128
  // See module-level "Rounding (float → integer)" doc — `+ 0.5 then
  // truncate` is the contract this crate uses across scalar + SIMD.
  let plane = as_le_f32(&[0.5f32]);
  let mut out = [0u8; 3];
  grayf32_to_rgb_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [128, 128, 128]);
}

#[test]
fn grayf32_to_rgb_saturates_high() {
  let plane = as_le_f32(&[1.5f32]);
  let mut out = [0u8; 3];
  grayf32_to_rgb_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [255, 255, 255]);
}

#[test]
fn grayf32_to_rgb_saturates_low() {
  let plane = [-0.1f32];
  let mut out = [0xFFu8; 3];
  grayf32_to_rgb_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [0, 0, 0]);
}

// ---- grayf32_to_rgba_row -------------------------------------------------

#[test]
fn grayf32_to_rgba_zero_alpha_opaque() {
  let plane = [0.0f32];
  let mut out = [0u8; 4];
  grayf32_to_rgba_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [0, 0, 0, 0xFF]);
}

#[test]
fn grayf32_to_rgba_max_alpha_opaque() {
  let plane = as_le_f32(&[1.0f32]);
  let mut out = [0u8; 4];
  grayf32_to_rgba_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [255, 255, 255, 0xFF]);
}

// ---- grayf32_to_rgb_u16_row ----------------------------------------------

#[test]
fn grayf32_to_rgb_u16_zero() {
  let plane = [0.0f32];
  let mut out = [0xFFFFu16; 3];
  grayf32_to_rgb_u16_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [0, 0, 0]);
}

#[test]
fn grayf32_to_rgb_u16_max() {
  let plane = as_le_f32(&[1.0f32]);
  let mut out = [0u16; 3];
  grayf32_to_rgb_u16_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [65535, 65535, 65535]);
}

#[test]
fn grayf32_to_rgb_u16_saturates_high() {
  let plane = as_le_f32(&[2.0f32]);
  let mut out = [0u16; 3];
  grayf32_to_rgb_u16_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [65535, 65535, 65535]);
}

// ---- grayf32_to_rgba_u16_row ---------------------------------------------

#[test]
fn grayf32_to_rgba_u16_opaque() {
  let plane = as_le_f32(&[1.0f32]);
  let mut out = [0u16; 4];
  grayf32_to_rgba_u16_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [65535, 65535, 65535, 0xFFFF]);
}

// ---- grayf32_to_rgb_f32_row ----------------------------------------------

#[test]
fn grayf32_to_rgb_f32_lossless_replicate() {
  // Non-clamped value preserved exactly. Output is host-native f32.
  let plane = as_le_f32(&[1.5f32]);
  let mut out = [0.0f32; 3];
  grayf32_to_rgb_f32_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [1.5, 1.5, 1.5]);
}

#[test]
fn grayf32_to_rgb_f32_negative_preserved() {
  let plane = as_le_f32(&[-0.5f32]);
  let mut out = [0.0f32; 3];
  grayf32_to_rgb_f32_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [-0.5, -0.5, -0.5]);
}

// ---- grayf32_to_luma_row -------------------------------------------------

#[test]
fn grayf32_to_luma_zero() {
  let plane = [0.0f32];
  let mut out = [0xFFu8; 1];
  grayf32_to_luma_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [0]);
}

#[test]
fn grayf32_to_luma_max() {
  let plane = as_le_f32(&[1.0f32]);
  let mut out = [0u8; 1];
  grayf32_to_luma_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [255]);
}

// ---- grayf32_to_luma_u16_row ---------------------------------------------

#[test]
fn grayf32_to_luma_u16_max() {
  let plane = as_le_f32(&[1.0f32]);
  let mut out = [0u16; 1];
  grayf32_to_luma_u16_row::<false>(&plane, &mut out, 1);
  assert_eq!(out, [65535]);
}

// ---- grayf32_to_luma_f32_row ---------------------------------------------

#[test]
fn grayf32_to_luma_f32_identity() {
  let plane = as_le_f32(&[0.0f32, 0.5, 1.0, 1.5, -0.1]);
  let mut out = [99.0f32; 5];
  grayf32_to_luma_f32_row::<false>(&plane, &mut out, 5);
  // Lossless pass-through — exact bit equality. Output is host-native f32.
  assert_eq!(out, [0.0, 0.5, 1.0, 1.5, -0.1]);
}

// ---- grayf32_to_hsv_row --------------------------------------------------

#[test]
fn grayf32_to_hsv_zero() {
  let plane = [0.0f32];
  let mut h = [0xFFu8; 1];
  let mut s = [0xFFu8; 1];
  let mut v = [0u8; 1];
  grayf32_to_hsv_row::<false>(&plane, &mut h, &mut s, &mut v, 1);
  assert_eq!(h[0], 0, "H must be 0 for achromatic source");
  assert_eq!(s[0], 0, "S must be 0 for achromatic source");
  assert_eq!(v[0], 0);
}

#[test]
fn grayf32_to_hsv_max() {
  let plane = as_le_f32(&[1.0f32]);
  let mut h = [0u8; 1];
  let mut s = [0u8; 1];
  let mut v = [0u8; 1];
  grayf32_to_hsv_row::<false>(&plane, &mut h, &mut s, &mut v, 1);
  assert_eq!(h[0], 0);
  assert_eq!(s[0], 0);
  assert_eq!(v[0], 255);
}

#[test]
fn grayf32_to_hsv_mid() {
  // 0.5 → (0.5 * 255 + 0.5) as u8 = 128
  let plane = as_le_f32(&[0.5f32]);
  let mut h = [0u8; 1];
  let mut s = [0u8; 1];
  let mut v = [0u8; 1];
  grayf32_to_hsv_row::<false>(&plane, &mut h, &mut s, &mut v, 1);
  assert_eq!(h[0], 0);
  assert_eq!(s[0], 0);
  assert_eq!(v[0], 128);
}

#[test]
fn grayf32_to_hsv_clamps_hdr() {
  // HDR value > 1.0 saturates to V=255.
  let plane = as_le_f32(&[2.0f32]);
  let mut h = [0u8; 1];
  let mut s = [0u8; 1];
  let mut v = [0u8; 1];
  grayf32_to_hsv_row::<false>(&plane, &mut h, &mut s, &mut v, 1);
  assert_eq!(v[0], 255);
}

#[test]
fn grayf32_to_rgb_multi_pixel() {
  let plane = as_le_f32(&[0.0f32, 1.0, 0.5]);
  let mut out = [0u8; 9];
  grayf32_to_rgb_row::<false>(&plane, &mut out, 3);
  assert_eq!(&out[0..3], &[0, 0, 0]);
  assert_eq!(&out[3..6], &[255, 255, 255]);
  assert_eq!(&out[6..9], &[128, 128, 128]); // 0.5 → 128
}

// ---- BE parity tests: grayf32 ---------------------------------------------
// Pattern: build a single host-native `intended` fixture, materialise it as
// LE-encoded bytes via `as_le_f32` and BE-encoded bytes via `as_be_f32`, run
// both `<false>` and `<true>` kernels, and pin each output against an
// absolute scalar reference so the parity assertion cannot pass on two
// equally corrupt decodes.

/// Mirror of `as_le_f32` for kernels invoked with `BE = true`. Combined
/// with `as_le_f32`, lets a single host-native `intended` fixture drive
/// both `<false>` and `<true>` kernel paths so they decode the same logical
/// bit pattern on every host.
fn as_be_f32(host: &[f32]) -> std::vec::Vec<f32> {
  host
    .iter()
    .map(|v| f32::from_bits(u32::from_ne_bytes(v.to_bits().to_be_bytes())))
    .collect()
}

// -- Scalar references for the BE-parity tests --
//
// Walk host-native `intended` planes and reproduce each kernel's documented
// behaviour without going through any byte-order conversion. Pinning the
// LE / BE outputs against these absolute references prevents the parity
// assertion from passing in lock-step on two equally corrupt decode paths.

fn ref_grayf32_to_rgb(intended: &[f32], width: usize) -> std::vec::Vec<u8> {
  let mut out = std::vec![0u8; width * 3];
  for (x, &y) in intended[..width].iter().enumerate() {
    let v = f32_to_u8(y);
    out[x * 3] = v;
    out[x * 3 + 1] = v;
    out[x * 3 + 2] = v;
  }
  out
}

fn ref_grayf32_to_luma(intended: &[f32], width: usize) -> std::vec::Vec<u8> {
  let mut out = std::vec![0u8; width];
  for (x, &y) in intended[..width].iter().enumerate() {
    out[x] = f32_to_u8(y);
  }
  out
}

/// Lossless f32 → f32 reference: pure host-native pass-through.
fn ref_grayf32_to_luma_f32(intended: &[f32], width: usize) -> std::vec::Vec<f32> {
  intended[..width].to_vec()
}

#[test]
fn grayf32_be_parity_rgb() {
  let intended = [0.5f32];
  let le = as_le_f32(&intended);
  let be = as_be_f32(&intended);
  let mut out_le = [0u8; 3];
  let mut out_be = [0u8; 3];
  grayf32_to_rgb_row::<false>(&le, &mut out_le, 1);
  grayf32_to_rgb_row::<true>(&be, &mut out_be, 1);
  let expected = ref_grayf32_to_rgb(&intended, 1);
  assert_eq!(
    out_le.as_slice(),
    expected,
    "LE path must match scalar reference"
  );
  assert_eq!(
    out_be.as_slice(),
    expected,
    "BE path must match scalar reference"
  );
  assert_eq!(out_le, out_be, "BE and LE grayf32 rgb outputs must agree");
}

#[test]
fn grayf32_be_parity_luma() {
  let intended = [0.25f32];
  let le = as_le_f32(&intended);
  let be = as_be_f32(&intended);
  let mut out_le = [0u8; 1];
  let mut out_be = [0u8; 1];
  grayf32_to_luma_row::<false>(&le, &mut out_le, 1);
  grayf32_to_luma_row::<true>(&be, &mut out_be, 1);
  let expected = ref_grayf32_to_luma(&intended, 1);
  assert_eq!(
    out_le.as_slice(),
    expected,
    "LE path must match scalar reference"
  );
  assert_eq!(
    out_be.as_slice(),
    expected,
    "BE path must match scalar reference"
  );
  assert_eq!(out_le, out_be, "BE and LE grayf32 luma outputs must agree");
}

/// The integer-output `grayf32_be_parity_*` tests don't reach the lossless
/// `grayf32_to_luma_f32_row::<true>` (BE-encoded f32 → host-native f32)
/// fast/slow paths. Build a single host-native `intended` fixture,
/// materialise it as LE / BE byte storage, run both kernels, pin both
/// outputs against an absolute scalar reference (compared bitwise to be
/// NaN-safe), and additionally assert bit-for-bit LE↔BE parity.
///
/// Path coverage by host:
///   LE host: LE kernel = memcpy fast path; BE kernel = slow swap path.
///   BE host: LE kernel = slow swap path; BE kernel = memcpy fast path.
/// Either way both outputs must agree bit-for-bit, exercising the
/// `BE == HOST_NATIVE_BE` gate from both directions.
#[test]
fn grayf32_to_luma_f32_row_be_le_parity_lossless() {
  // Mix of normal, HDR, negative, subnormal, and exact-zero values to
  // ensure non-symmetric byte layouts in every position.
  let intended: std::vec::Vec<f32> = std::vec![0.25f32, 1.5, -0.5, 1e-5, 0.0, 65504.0];
  let width = intended.len();
  let le = as_le_f32(&intended);
  let be = as_be_f32(&intended);
  let mut out_le = std::vec![0.0f32; width];
  let mut out_be = std::vec![0.0f32; width];
  grayf32_to_luma_f32_row::<false>(&le, &mut out_le, width);
  grayf32_to_luma_f32_row::<true>(&be, &mut out_be, width);
  let expected = ref_grayf32_to_luma_f32(&intended, width);
  // Bitwise equality is NaN-safe and confirms no rounding/clamping was
  // applied along either path.
  let bits_le: std::vec::Vec<u32> = out_le.iter().map(|v| v.to_bits()).collect();
  let bits_be: std::vec::Vec<u32> = out_be.iter().map(|v| v.to_bits()).collect();
  let bits_expected: std::vec::Vec<u32> = expected.iter().map(|v| v.to_bits()).collect();
  assert_eq!(
    bits_le, bits_expected,
    "LE path must match scalar reference"
  );
  assert_eq!(
    bits_be, bits_expected,
    "BE path must match scalar reference"
  );
  assert_eq!(
    bits_le, bits_be,
    "BE and LE grayf32 luma_f32 outputs must match bit-for-bit"
  );
}