| 1 |
use crate::codecs::avif::yuv::{ |
| 2 |
check_rgb_preconditions, check_yuv_plane_preconditions, qrshr, yuv420_to_rgbx_invoker, |
| 3 |
yuv422_to_rgbx_invoker, CbCrInverseTransform, HalvedRowHandler, PlaneDefinition, |
| 4 |
YuvChromaRange, YuvIntensityRange, YuvPlanarImage, YuvStandardMatrix, |
| 5 |
}; |
| 6 |
use crate::ImageError; |
| 7 |
use num_traits::AsPrimitive; |
| 8 |
|
| 9 |
/// Computes YCgCo inverse in limited range |
| 10 |
/// # Arguments |
| 11 |
/// - `dst` - dest buffer |
| 12 |
/// - `y_value` - Y value with subtracted bias |
| 13 |
/// - `cb` - Cb value with subtracted bias |
| 14 |
/// - `cr` - Cr value with subtracted bias |
| 15 |
#[inline(always)] |
| 16 |
fn ycgco_execute_limited< |
| 17 |
V: Copy + AsPrimitive<i32> + 'static + Sized, |
| 18 |
const PRECISION: i32, |
| 19 |
const CHANNELS: usize, |
| 20 |
const BIT_DEPTH: usize, |
| 21 |
>( |
| 22 |
dst: &mut [V; CHANNELS], |
| 23 |
y_value: i32, |
| 24 |
cg: i32, |
| 25 |
co: i32, |
| 26 |
scale: i32, |
| 27 |
) where |
| 28 |
i32: AsPrimitive<V>, |
| 29 |
{ |
| 30 |
let t0 = y_value - cg; |
| 31 |
|
| 32 |
let r = qrshr::<PRECISION, BIT_DEPTH>((t0 + co) * scale); |
| 33 |
let b = qrshr::<PRECISION, BIT_DEPTH>((t0 - co) * scale); |
| 34 |
let g = qrshr::<PRECISION, BIT_DEPTH>((y_value + cg) * scale); |
| 35 |
|
| 36 |
if CHANNELS == 4 { |
| 37 |
dst[0] = r.as_(); |
| 38 |
dst[1] = g.as_(); |
| 39 |
dst[2] = b.as_(); |
| 40 |
dst[3] = ((1i32 << BIT_DEPTH) - 1).as_(); |
| 41 |
} else if CHANNELS == 3 { |
| 42 |
dst[0] = r.as_(); |
| 43 |
dst[1] = g.as_(); |
| 44 |
dst[2] = b.as_(); |
| 45 |
} else { |
| 46 |
unreachable!(); |
| 47 |
} |
| 48 |
} |
| 49 |
|
| 50 |
/// Computes YCgCo inverse in full range |
| 51 |
/// # Arguments |
| 52 |
/// - `dst` - dest buffer |
| 53 |
/// - `y_value` - Y value with subtracted bias |
| 54 |
/// - `cb` - Cb value with subtracted bias |
| 55 |
/// - `cr` - Cr value with subtracted bias |
| 56 |
#[inline(always)] |
| 57 |
fn ycgco_execute_full< |
| 58 |
V: Copy + AsPrimitive<i32> + 'static + Sized, |
| 59 |
const PRECISION: i32, |
| 60 |
const CHANNELS: usize, |
| 61 |
const BIT_DEPTH: usize, |
| 62 |
>( |
| 63 |
dst: &mut [V; CHANNELS], |
| 64 |
y_value: i32, |
| 65 |
cg: i32, |
| 66 |
co: i32, |
| 67 |
) where |
| 68 |
i32: AsPrimitive<V>, |
| 69 |
{ |
| 70 |
let t0 = y_value - cg; |
| 71 |
|
| 72 |
let max_value = (1i32 << BIT_DEPTH) - 1; |
| 73 |
|
| 74 |
let r = (t0 + co).clamp(0, max_value); |
| 75 |
let b = (t0 - co).clamp(0, max_value); |
| 76 |
let g = (y_value + cg).clamp(0, max_value); |
| 77 |
|
| 78 |
if CHANNELS == 4 { |
| 79 |
dst[0] = r.as_(); |
| 80 |
dst[1] = g.as_(); |
| 81 |
dst[2] = b.as_(); |
| 82 |
dst[3] = max_value.as_(); |
| 83 |
} else if CHANNELS == 3 { |
| 84 |
dst[0] = r.as_(); |
| 85 |
dst[1] = g.as_(); |
| 86 |
dst[2] = b.as_(); |
| 87 |
} else { |
| 88 |
unreachable!(); |
| 89 |
} |
| 90 |
} |
| 91 |
|
| 92 |
#[inline(always)] |
| 93 |
fn process_halved_chroma_row_cgco< |
| 94 |
V: Copy + AsPrimitive<i32> + 'static + Sized, |
| 95 |
const PRECISION: i32, |
| 96 |
const CHANNELS: usize, |
| 97 |
const BIT_DEPTH: usize, |
| 98 |
>( |
| 99 |
image: YuvPlanarImage<V>, |
| 100 |
rgba: &mut [V], |
| 101 |
_: &CbCrInverseTransform<i32>, |
| 102 |
range: &YuvChromaRange, |
| 103 |
) where |
| 104 |
i32: AsPrimitive<V>, |
| 105 |
{ |
| 106 |
let max_value = (1i32 << BIT_DEPTH) - 1; |
| 107 |
|
| 108 |
// If the stride is larger than the plane size, |
| 109 |
// it might contain junk data beyond the actual valid region. |
| 110 |
// To avoid processing artifacts when working with odd-sized images, |
| 111 |
// the buffer is reshaped to its actual size, |
| 112 |
// preventing accidental use of invalid values from the trailing region. |
| 113 |
|
| 114 |
let y_plane = &image.y_plane[..image.width]; |
| 115 |
let chroma_size = image.width.div_ceil(2); |
| 116 |
let u_plane = &image.u_plane[..chroma_size]; |
| 117 |
let v_plane = &image.v_plane[..chroma_size]; |
| 118 |
let rgba = &mut rgba[..image.width * CHANNELS]; |
| 119 |
|
| 120 |
let bias_y = range.bias_y as i32; |
| 121 |
let bias_uv = range.bias_uv as i32; |
| 122 |
let y_iter = y_plane.chunks_exact(2); |
| 123 |
let rgb_chunks = rgba.chunks_exact_mut(CHANNELS * 2); |
| 124 |
|
| 125 |
let scale_coef = ((max_value as f32 / range.range_y as f32) * (1 << PRECISION) as f32) as i32; |
| 126 |
|
| 127 |
for (((y_src, &u_src), &v_src), rgb_dst) in y_iter.zip(u_plane).zip(v_plane).zip(rgb_chunks) { |
| 128 |
let y_value0: i32 = y_src[0].as_() - bias_y; |
| 129 |
let cg_value: i32 = u_src.as_() - bias_uv; |
| 130 |
let co_value: i32 = v_src.as_() - bias_uv; |
| 131 |
|
| 132 |
let dst0 = &mut rgb_dst[..CHANNELS]; |
| 133 |
|
| 134 |
ycgco_execute_limited::<V, PRECISION, CHANNELS, BIT_DEPTH>( |
| 135 |
dst0.try_into().unwrap(), |
| 136 |
y_value0, |
| 137 |
cg_value, |
| 138 |
co_value, |
| 139 |
scale_coef, |
| 140 |
); |
| 141 |
|
| 142 |
let y_value1 = y_src[1].as_() - bias_y; |
| 143 |
|
| 144 |
let dst1 = &mut rgb_dst[CHANNELS..2 * CHANNELS]; |
| 145 |
|
| 146 |
ycgco_execute_limited::<V, PRECISION, CHANNELS, BIT_DEPTH>( |
| 147 |
dst1.try_into().unwrap(), |
| 148 |
y_value1, |
| 149 |
cg_value, |
| 150 |
co_value, |
| 151 |
scale_coef, |
| 152 |
); |
| 153 |
} |
| 154 |
|
| 155 |
// Process remainder if width is odd. |
| 156 |
if image.width & 1 != 0 { |
| 157 |
let y_left = y_plane.chunks_exact(2).remainder(); |
| 158 |
let rgb_chunks = rgba |
| 159 |
.chunks_exact_mut(CHANNELS * 2) |
| 160 |
.into_remainder() |
| 161 |
.chunks_exact_mut(CHANNELS); |
| 162 |
let u_iter = u_plane.iter().rev(); |
| 163 |
let v_iter = v_plane.iter().rev(); |
| 164 |
|
| 165 |
for (((y_src, u_src), v_src), rgb_dst) in |
| 166 |
y_left.iter().zip(u_iter).zip(v_iter).zip(rgb_chunks) |
| 167 |
{ |
| 168 |
let y_value = y_src.as_() - bias_y; |
| 169 |
let cg_value = u_src.as_() - bias_uv; |
| 170 |
let co_value = v_src.as_() - bias_uv; |
| 171 |
|
| 172 |
ycgco_execute_limited::<V, PRECISION, CHANNELS, BIT_DEPTH>( |
| 173 |
rgb_dst.try_into().unwrap(), |
| 174 |
y_value, |
| 175 |
cg_value, |
| 176 |
co_value, |
| 177 |
scale_coef, |
| 178 |
); |
| 179 |
} |
| 180 |
} |
| 181 |
} |
| 182 |
|
| 183 |
/// Converts YCgCo 444 planar format to Rgba |
| 184 |
/// |
| 185 |
/// # Arguments |
| 186 |
/// |
| 187 |
/// * `image`: see [YuvPlanarImage] |
| 188 |
/// * `rgba`: RGB image layout |
| 189 |
/// * `range`: see [YuvIntensityRange] |
| 190 |
/// |
| 191 |
fn ycgco444_to_rgbx_impl< |
| 192 |
V: Copy + AsPrimitive<i32> + 'static + Sized, |
| 193 |
const CHANNELS: usize, |
| 194 |
const BIT_DEPTH: usize, |
| 195 |
>( |
| 196 |
image: YuvPlanarImage<V>, |
| 197 |
rgba: &mut [V], |
| 198 |
yuv_range: YuvIntensityRange, |
| 199 |
) -> Result<(), ImageError> |
| 200 |
where |
| 201 |
i32: AsPrimitive<V>, |
| 202 |
{ |
| 203 |
assert!( |
| 204 |
CHANNELS == 3 || CHANNELS == 4, |
| 205 |
"YUV 4:4:4 -> RGB is implemented only on 3 and 4 channels" |
| 206 |
); |
| 207 |
assert!( |
| 208 |
(8..=16).contains(&BIT_DEPTH), |
| 209 |
"Invalid bit depth is provided" |
| 210 |
); |
| 211 |
assert!( |
| 212 |
if BIT_DEPTH > 8 { |
| 213 |
size_of::<V>() == 2 |
| 214 |
} else { |
| 215 |
size_of::<V>() == 1 |
| 216 |
}, |
| 217 |
"Unsupported bit depth and data type combination" |
| 218 |
); |
| 219 |
|
| 220 |
let y_plane = image.y_plane; |
| 221 |
let u_plane = image.u_plane; |
| 222 |
let v_plane = image.v_plane; |
| 223 |
let y_stride = image.y_stride; |
| 224 |
let u_stride = image.u_stride; |
| 225 |
let v_stride = image.v_stride; |
| 226 |
let height = image.height; |
| 227 |
let width = image.width; |
| 228 |
|
| 229 |
check_yuv_plane_preconditions(y_plane, PlaneDefinition::Y, y_stride, height)?; |
| 230 |
check_yuv_plane_preconditions(u_plane, PlaneDefinition::U, u_stride, height)?; |
| 231 |
check_yuv_plane_preconditions(v_plane, PlaneDefinition::V, v_stride, height)?; |
| 232 |
|
| 233 |
check_rgb_preconditions(rgba, image.width * CHANNELS, height)?; |
| 234 |
|
| 235 |
let range = yuv_range.get_yuv_range(BIT_DEPTH as u32); |
| 236 |
const PRECISION: i32 = 13; |
| 237 |
|
| 238 |
let bias_y = range.bias_y as i32; |
| 239 |
let bias_uv = range.bias_uv as i32; |
| 240 |
|
| 241 |
let rgb_stride = width * CHANNELS; |
| 242 |
|
| 243 |
let y_iter = y_plane.chunks_exact(y_stride); |
| 244 |
let rgb_iter = rgba.chunks_exact_mut(rgb_stride); |
| 245 |
let u_iter = u_plane.chunks_exact(u_stride); |
| 246 |
let v_iter = v_plane.chunks_exact(v_stride); |
| 247 |
|
| 248 |
let max_value: i32 = (1 << BIT_DEPTH) - 1; |
| 249 |
|
| 250 |
// All branches on generic const will be optimized out. |
| 251 |
for (((y_src, u_src), v_src), rgb) in y_iter.zip(u_iter).zip(v_iter).zip(rgb_iter) { |
| 252 |
let rgb_chunks = rgb.chunks_exact_mut(CHANNELS); |
| 253 |
match yuv_range { |
| 254 |
YuvIntensityRange::Tv => { |
| 255 |
let y_coef = |
| 256 |
((max_value as f32 / range.range_y as f32) * (1 << PRECISION) as f32) as i32; |
| 257 |
for (((y_src, u_src), v_src), rgb_dst) in |
| 258 |
y_src.iter().zip(u_src).zip(v_src).zip(rgb_chunks) |
| 259 |
{ |
| 260 |
let y_value = y_src.as_() - bias_y; |
| 261 |
let cg_value = u_src.as_() - bias_uv; |
| 262 |
let co_value = v_src.as_() - bias_uv; |
| 263 |
|
| 264 |
ycgco_execute_limited::<V, PRECISION, CHANNELS, BIT_DEPTH>( |
| 265 |
rgb_dst.try_into().unwrap(), |
| 266 |
y_value, |
| 267 |
cg_value, |
| 268 |
co_value, |
| 269 |
y_coef, |
| 270 |
); |
| 271 |
} |
| 272 |
} |
| 273 |
YuvIntensityRange::Pc => { |
| 274 |
for (((y_src, u_src), v_src), rgb_dst) in |
| 275 |
y_src.iter().zip(u_src).zip(v_src).zip(rgb_chunks) |
| 276 |
{ |
| 277 |
let y_value = y_src.as_() - bias_y; |
| 278 |
let cg_value = u_src.as_() - bias_uv; |
| 279 |
let co_value = v_src.as_() - bias_uv; |
| 280 |
|
| 281 |
ycgco_execute_full::<V, PRECISION, CHANNELS, BIT_DEPTH>( |
| 282 |
rgb_dst.try_into().unwrap(), |
| 283 |
y_value, |
| 284 |
cg_value, |
| 285 |
co_value, |
| 286 |
); |
| 287 |
} |
| 288 |
} |
| 289 |
} |
| 290 |
} |
| 291 |
|
| 292 |
Ok(()) |
| 293 |
} |
| 294 |
|
| 295 |
macro_rules! define_ycgco_half_chroma { |
| 296 |
($name: ident, $invoker: ident, $storage: ident, $cn: expr, $bp: expr, $description: expr) => { |
| 297 |
#[doc = concat!($description, " |
| 298 |
|
| 299 |
# Arguments |
| 300 |
|
| 301 |
* `image`: see [YuvPlanarImage] |
| 302 |
* `rgb`: RGB image layout |
| 303 |
* `range`: see [YuvIntensityRange] |
| 304 |
* `matrix`: see [YuvStandardMatrix]")] |
| 305 |
pub(crate) fn $name( |
| 306 |
image: YuvPlanarImage<$storage>, |
| 307 |
rgb: &mut [$storage], |
| 308 |
range: YuvIntensityRange, |
| 309 |
) -> Result<(), ImageError> { |
| 310 |
const P: i32 = 13; |
| 311 |
$invoker::<$storage, HalvedRowHandler<$storage>, P, $cn, $bp>( |
| 312 |
image, |
| 313 |
rgb, |
| 314 |
range, |
| 315 |
YuvStandardMatrix::Bt709, |
| 316 |
process_halved_chroma_row_cgco::<$storage, P, $cn, $bp>, |
| 317 |
) |
| 318 |
} |
| 319 |
}; |
| 320 |
} |
| 321 |
|
| 322 |
const RGBA_CN: usize = 4; |
| 323 |
|
| 324 |
define_ycgco_half_chroma!( |
| 325 |
ycgco420_to_rgba8, |
| 326 |
yuv420_to_rgbx_invoker, |
| 327 |
u8, |
| 328 |
RGBA_CN, |
| 329 |
8, |
| 330 |
"Converts YCgCo 420 8-bit planar format to Rgba 8-bit" |
| 331 |
); |
| 332 |
|
| 333 |
define_ycgco_half_chroma!( |
| 334 |
ycgco422_to_rgba8, |
| 335 |
yuv422_to_rgbx_invoker, |
| 336 |
u8, |
| 337 |
RGBA_CN, |
| 338 |
8, |
| 339 |
"Converts YCgCo 420 8-bit planar format to Rgba 8-bit" |
| 340 |
); |
| 341 |
|
| 342 |
define_ycgco_half_chroma!( |
| 343 |
ycgco420_to_rgba10, |
| 344 |
yuv420_to_rgbx_invoker, |
| 345 |
u16, |
| 346 |
RGBA_CN, |
| 347 |
10, |
| 348 |
"Converts YCgCo 420 10-bit planar format to Rgba 10-bit" |
| 349 |
); |
| 350 |
|
| 351 |
define_ycgco_half_chroma!( |
| 352 |
ycgco422_to_rgba10, |
| 353 |
yuv422_to_rgbx_invoker, |
| 354 |
u16, |
| 355 |
RGBA_CN, |
| 356 |
10, |
| 357 |
"Converts YCgCo 422 10-bit planar format to Rgba 10-bit" |
| 358 |
); |
| 359 |
|
| 360 |
define_ycgco_half_chroma!( |
| 361 |
ycgco420_to_rgba12, |
| 362 |
yuv420_to_rgbx_invoker, |
| 363 |
u16, |
| 364 |
RGBA_CN, |
| 365 |
12, |
| 366 |
"Converts YCgCo 420 12-bit planar format to Rgba 12-bit" |
| 367 |
); |
| 368 |
|
| 369 |
define_ycgco_half_chroma!( |
| 370 |
ycgco422_to_rgba12, |
| 371 |
yuv422_to_rgbx_invoker, |
| 372 |
u16, |
| 373 |
RGBA_CN, |
| 374 |
12, |
| 375 |
"Converts YCgCo 422 12-bit planar format to Rgba 12-bit" |
| 376 |
); |
| 377 |
|
| 378 |
macro_rules! define_ycgcg_full_chroma { |
| 379 |
($name: ident, $storage: ident, $cn: expr, $bp: expr, $description: expr) => { |
| 380 |
#[doc = concat!($description, " |
| 381 |
|
| 382 |
# Arguments |
| 383 |
|
| 384 |
* `image`: see [YuvPlanarImage] |
| 385 |
* `rgba`: RGB image layout |
| 386 |
* `range`: see [YuvIntensityRange] |
| 387 |
* `matrix`: see [YuvStandardMatrix] |
| 388 |
")] |
| 389 |
pub(crate) fn $name( |
| 390 |
image: YuvPlanarImage<$storage>, |
| 391 |
rgba: &mut [$storage], |
| 392 |
range: YuvIntensityRange, |
| 393 |
) -> Result<(), ImageError> { |
| 394 |
ycgco444_to_rgbx_impl::<$storage, $cn, $bp>(image, rgba, range) |
| 395 |
} |
| 396 |
}; |
| 397 |
} |
| 398 |
|
| 399 |
define_ycgcg_full_chroma!( |
| 400 |
ycgco444_to_rgba8, |
| 401 |
u8, |
| 402 |
RGBA_CN, |
| 403 |
8, |
| 404 |
"Converts YCgCo 444 planar format 8 bit-depth to Rgba 8 bit" |
| 405 |
); |
| 406 |
define_ycgcg_full_chroma!( |
| 407 |
ycgco444_to_rgba10, |
| 408 |
u16, |
| 409 |
RGBA_CN, |
| 410 |
10, |
| 411 |
"Converts YCgCo 444 planar format 10 bit-depth to Rgba 10 bit" |
| 412 |
); |
| 413 |
define_ycgcg_full_chroma!( |
| 414 |
ycgco444_to_rgba12, |
| 415 |
u16, |
| 416 |
RGBA_CN, |
| 417 |
12, |
| 418 |
"Converts YCgCo 444 planar format 12 bit-depth to Rgba 12 bit" |
| 419 |
); |