| 1 |
/* |
| 2 |
* // Copyright (c) Radzivon Bartoshyk 3/2025. All rights reserved. |
| 3 |
* // |
| 4 |
* // Redistribution and use in source and binary forms, with or without modification, |
| 5 |
* // are permitted provided that the following conditions are met: |
| 6 |
* // |
| 7 |
* // 1. Redistributions of source code must retain the above copyright notice, this |
| 8 |
* // list of conditions and the following disclaimer. |
| 9 |
* // |
| 10 |
* // 2. Redistributions in binary form must reproduce the above copyright notice, |
| 11 |
* // this list of conditions and the following disclaimer in the documentation |
| 12 |
* // and/or other materials provided with the distribution. |
| 13 |
* // |
| 14 |
* // 3. Neither the name of the copyright holder nor the names of its |
| 15 |
* // contributors may be used to endorse or promote products derived from |
| 16 |
* // this software without specific prior written permission. |
| 17 |
* // |
| 18 |
* // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" |
| 19 |
* // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 20 |
* // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE |
| 21 |
* // DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE |
| 22 |
* // FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 23 |
* // DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR |
| 24 |
* // SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER |
| 25 |
* // CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, |
| 26 |
* // OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 27 |
* // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 28 |
*/ |
| 29 |
use crate::conversions::neon::rgb_xyz_q2_13::{split_by_twos, split_by_twos_mut}; |
| 30 |
use crate::conversions::rgbxyz::TransformMatrixShaperV; |
| 31 |
use crate::transform::PointeeSizeExpressible; |
| 32 |
use crate::{CmsError, Layout, TransformExecutor}; |
| 33 |
use num_traits::AsPrimitive; |
| 34 |
use std::arch::aarch64::*; |
| 35 |
|
| 36 |
#[repr(align(16), C)] |
| 37 |
pub(crate) struct NeonAlignedU16(pub(crate) [u16; 8]); |
| 38 |
|
| 39 |
#[repr(align(16), C)] |
| 40 |
pub(crate) struct NeonAlignedF32(pub(crate) [f32; 4]); |
| 41 |
|
| 42 |
pub(crate) struct TransformShaperRgbNeon< |
| 43 |
T: Clone + PointeeSizeExpressible + Copy + Default + 'static, |
| 44 |
const SRC_LAYOUT: u8, |
| 45 |
const DST_LAYOUT: u8, |
| 46 |
> { |
| 47 |
pub(crate) profile: TransformMatrixShaperV<T>, |
| 48 |
pub(crate) bit_depth: usize, |
| 49 |
pub(crate) gamma_lut: usize, |
| 50 |
} |
| 51 |
|
| 52 |
impl< |
| 53 |
T: Clone + PointeeSizeExpressible + Copy + Default + 'static, |
| 54 |
const SRC_LAYOUT: u8, |
| 55 |
const DST_LAYOUT: u8, |
| 56 |
> TransformExecutor<T> for TransformShaperRgbNeon<T, SRC_LAYOUT, DST_LAYOUT> |
| 57 |
where |
| 58 |
u32: AsPrimitive<T>, |
| 59 |
{ |
| 60 |
fn transform(&self, src: &[T], dst: &mut [T]) -> Result<(), CmsError> { |
| 61 |
let src_cn = Layout::from(SRC_LAYOUT); |
| 62 |
let dst_cn = Layout::from(DST_LAYOUT); |
| 63 |
let src_channels = src_cn.channels(); |
| 64 |
let dst_channels = dst_cn.channels(); |
| 65 |
|
| 66 |
let mut temporary0 = NeonAlignedU16([0; 8]); |
| 67 |
let mut temporary1 = NeonAlignedU16([0; 8]); |
| 68 |
let mut temporary2 = NeonAlignedU16([0; 8]); |
| 69 |
let mut temporary3 = NeonAlignedU16([0; 8]); |
| 70 |
|
| 71 |
if src.len() / src_channels != dst.len() / dst_channels { |
| 72 |
return Err(CmsError::LaneSizeMismatch); |
| 73 |
} |
| 74 |
if src.len() % src_channels != 0 { |
| 75 |
return Err(CmsError::LaneMultipleOfChannels); |
| 76 |
} |
| 77 |
if dst.len() % dst_channels != 0 { |
| 78 |
return Err(CmsError::LaneMultipleOfChannels); |
| 79 |
} |
| 80 |
|
| 81 |
let t = self.profile.adaptation_matrix.transpose(); |
| 82 |
let scale = (self.gamma_lut - 1) as f32; |
| 83 |
let max_colors: T = ((1 << self.bit_depth) - 1).as_(); |
| 84 |
|
| 85 |
// safety precondition for linearization table |
| 86 |
if T::FINITE { |
| 87 |
let cap = (1 << self.bit_depth) - 1; |
| 88 |
assert!(self.profile.r_linear.len() >= cap); |
| 89 |
assert!(self.profile.g_linear.len() >= cap); |
| 90 |
assert!(self.profile.b_linear.len() >= cap); |
| 91 |
} else { |
| 92 |
assert!(self.profile.r_linear.len() >= T::NOT_FINITE_LINEAR_TABLE_SIZE); |
| 93 |
assert!(self.profile.g_linear.len() >= T::NOT_FINITE_LINEAR_TABLE_SIZE); |
| 94 |
assert!(self.profile.b_linear.len() >= T::NOT_FINITE_LINEAR_TABLE_SIZE); |
| 95 |
} |
| 96 |
|
| 97 |
let r_lin = &self.profile.r_linear; |
| 98 |
let g_lin = &self.profile.g_linear; |
| 99 |
let b_lin = &self.profile.b_linear; |
| 100 |
|
| 101 |
let (src_chunks, src_remainder) = split_by_twos(src, src_channels); |
| 102 |
let (dst_chunks, dst_remainder) = split_by_twos_mut(dst, dst_channels); |
| 103 |
|
| 104 |
unsafe { |
| 105 |
let m0 = vld1q_f32([t.v[0][0], t.v[0][1], t.v[0][2], 0.].as_ptr()); |
| 106 |
let m1 = vld1q_f32([t.v[1][0], t.v[1][1], t.v[1][2], 0.].as_ptr()); |
| 107 |
let m2 = vld1q_f32([t.v[2][0], t.v[2][1], t.v[2][2], 0.].as_ptr()); |
| 108 |
|
| 109 |
let v_scale = vdupq_n_f32(scale); |
| 110 |
|
| 111 |
let rnd = vdupq_n_f32(0.5); |
| 112 |
|
| 113 |
if !src_chunks.is_empty() { |
| 114 |
let (src0, src1) = src_chunks.split_at(src_chunks.len() / 2); |
| 115 |
let (dst0, dst1) = dst_chunks.split_at_mut(dst_chunks.len() / 2); |
| 116 |
let mut src_iter0 = src0.chunks_exact(src_channels * 2); |
| 117 |
let mut src_iter1 = src1.chunks_exact(src_channels * 2); |
| 118 |
|
| 119 |
let (mut r0, mut g0, mut b0, mut a0); |
| 120 |
let (mut r1, mut g1, mut b1, mut a1); |
| 121 |
let (mut r2, mut g2, mut b2, mut a2); |
| 122 |
let (mut r3, mut g3, mut b3, mut a3); |
| 123 |
|
| 124 |
if let (Some(src0), Some(src1)) = (src_iter0.next(), src_iter1.next()) { |
| 125 |
let r0p = r_lin.get_unchecked(src0[src_cn.r_i()]._as_usize()); |
| 126 |
let g0p = g_lin.get_unchecked(src0[src_cn.g_i()]._as_usize()); |
| 127 |
let b0p = b_lin.get_unchecked(src0[src_cn.b_i()]._as_usize()); |
| 128 |
|
| 129 |
let r1p = r_lin.get_unchecked(src0[src_cn.r_i() + src_channels]._as_usize()); |
| 130 |
let g1p = g_lin.get_unchecked(src0[src_cn.g_i() + src_channels]._as_usize()); |
| 131 |
let b1p = b_lin.get_unchecked(src0[src_cn.b_i() + src_channels]._as_usize()); |
| 132 |
|
| 133 |
let r2p = r_lin.get_unchecked(src1[src_cn.r_i()]._as_usize()); |
| 134 |
let g2p = g_lin.get_unchecked(src1[src_cn.g_i()]._as_usize()); |
| 135 |
let b2p = b_lin.get_unchecked(src1[src_cn.b_i()]._as_usize()); |
| 136 |
|
| 137 |
let r3p = r_lin.get_unchecked(src1[src_cn.r_i() + src_channels]._as_usize()); |
| 138 |
let g3p = g_lin.get_unchecked(src1[src_cn.g_i() + src_channels]._as_usize()); |
| 139 |
let b3p = b_lin.get_unchecked(src1[src_cn.b_i() + src_channels]._as_usize()); |
| 140 |
|
| 141 |
r0 = vld1q_dup_f32(r0p); |
| 142 |
g0 = vld1q_dup_f32(g0p); |
| 143 |
b0 = vld1q_dup_f32(b0p); |
| 144 |
|
| 145 |
r1 = vld1q_dup_f32(r1p); |
| 146 |
g1 = vld1q_dup_f32(g1p); |
| 147 |
b1 = vld1q_dup_f32(b1p); |
| 148 |
|
| 149 |
r2 = vld1q_dup_f32(r2p); |
| 150 |
g2 = vld1q_dup_f32(g2p); |
| 151 |
b2 = vld1q_dup_f32(b2p); |
| 152 |
|
| 153 |
r3 = vld1q_dup_f32(r3p); |
| 154 |
g3 = vld1q_dup_f32(g3p); |
| 155 |
b3 = vld1q_dup_f32(b3p); |
| 156 |
|
| 157 |
a0 = if src_channels == 4 { |
| 158 |
src0[src_cn.a_i()] |
| 159 |
} else { |
| 160 |
max_colors |
| 161 |
}; |
| 162 |
|
| 163 |
a1 = if src_channels == 4 { |
| 164 |
src0[src_cn.a_i() + src_channels] |
| 165 |
} else { |
| 166 |
max_colors |
| 167 |
}; |
| 168 |
|
| 169 |
a2 = if src_channels == 4 { |
| 170 |
src1[src_cn.a_i()] |
| 171 |
} else { |
| 172 |
max_colors |
| 173 |
}; |
| 174 |
|
| 175 |
a3 = if src_channels == 4 { |
| 176 |
src1[src_cn.a_i() + src_channels] |
| 177 |
} else { |
| 178 |
max_colors |
| 179 |
}; |
| 180 |
} else { |
| 181 |
r0 = vdupq_n_f32(0.); |
| 182 |
g0 = vdupq_n_f32(0.); |
| 183 |
b0 = vdupq_n_f32(0.); |
| 184 |
r1 = vdupq_n_f32(0.); |
| 185 |
g1 = vdupq_n_f32(0.); |
| 186 |
b1 = vdupq_n_f32(0.); |
| 187 |
r2 = vdupq_n_f32(0.); |
| 188 |
g2 = vdupq_n_f32(0.); |
| 189 |
b2 = vdupq_n_f32(0.); |
| 190 |
r3 = vdupq_n_f32(0.); |
| 191 |
g3 = vdupq_n_f32(0.); |
| 192 |
b3 = vdupq_n_f32(0.); |
| 193 |
a0 = max_colors; |
| 194 |
a1 = max_colors; |
| 195 |
a2 = max_colors; |
| 196 |
a3 = max_colors; |
| 197 |
} |
| 198 |
|
| 199 |
for (((src0, src1), dst0), dst1) in src_iter0 |
| 200 |
.zip(src_iter1) |
| 201 |
.zip(dst0.chunks_exact_mut(dst_channels * 2)) |
| 202 |
.zip(dst1.chunks_exact_mut(dst_channels * 2)) |
| 203 |
{ |
| 204 |
let v0_0 = vmulq_f32(r0, m0); |
| 205 |
let v0_1 = vmulq_f32(r1, m0); |
| 206 |
let v0_2 = vmulq_f32(r2, m0); |
| 207 |
let v0_3 = vmulq_f32(r3, m0); |
| 208 |
|
| 209 |
let v1_0 = vfmaq_f32(v0_0, g0, m1); |
| 210 |
let v1_1 = vfmaq_f32(v0_1, g1, m1); |
| 211 |
let v1_2 = vfmaq_f32(v0_2, g2, m1); |
| 212 |
let v1_3 = vfmaq_f32(v0_3, g3, m1); |
| 213 |
|
| 214 |
let mut vr0 = vfmaq_f32(v1_0, b0, m2); |
| 215 |
let mut vr1 = vfmaq_f32(v1_1, b1, m2); |
| 216 |
let mut vr2 = vfmaq_f32(v1_2, b2, m2); |
| 217 |
let mut vr3 = vfmaq_f32(v1_3, b3, m2); |
| 218 |
|
| 219 |
vr0 = vfmaq_f32(rnd, vr0, v_scale); |
| 220 |
vr1 = vfmaq_f32(rnd, vr1, v_scale); |
| 221 |
vr2 = vfmaq_f32(rnd, vr2, v_scale); |
| 222 |
vr3 = vfmaq_f32(rnd, vr3, v_scale); |
| 223 |
|
| 224 |
vr0 = vminq_f32(vr0, v_scale); |
| 225 |
vr1 = vminq_f32(vr1, v_scale); |
| 226 |
vr2 = vminq_f32(vr2, v_scale); |
| 227 |
vr3 = vminq_f32(vr3, v_scale); |
| 228 |
|
| 229 |
let zx0 = vcvtq_u32_f32(vr0); |
| 230 |
let zx1 = vcvtq_u32_f32(vr1); |
| 231 |
let zx2 = vcvtq_u32_f32(vr2); |
| 232 |
let zx3 = vcvtq_u32_f32(vr3); |
| 233 |
vst1q_u32(temporary0.0.as_mut_ptr() as *mut _, zx0); |
| 234 |
vst1q_u32(temporary1.0.as_mut_ptr() as *mut _, zx1); |
| 235 |
vst1q_u32(temporary2.0.as_mut_ptr() as *mut _, zx2); |
| 236 |
vst1q_u32(temporary3.0.as_mut_ptr() as *mut _, zx3); |
| 237 |
|
| 238 |
let r0p = r_lin.get_unchecked(src0[src_cn.r_i()]._as_usize()); |
| 239 |
let g0p = g_lin.get_unchecked(src0[src_cn.g_i()]._as_usize()); |
| 240 |
let b0p = b_lin.get_unchecked(src0[src_cn.b_i()]._as_usize()); |
| 241 |
|
| 242 |
let r1p = r_lin.get_unchecked(src0[src_cn.r_i() + src_channels]._as_usize()); |
| 243 |
let g1p = g_lin.get_unchecked(src0[src_cn.g_i() + src_channels]._as_usize()); |
| 244 |
let b1p = b_lin.get_unchecked(src0[src_cn.b_i() + src_channels]._as_usize()); |
| 245 |
|
| 246 |
let r2p = r_lin.get_unchecked(src1[src_cn.r_i()]._as_usize()); |
| 247 |
let g2p = g_lin.get_unchecked(src1[src_cn.g_i()]._as_usize()); |
| 248 |
let b2p = b_lin.get_unchecked(src1[src_cn.b_i()]._as_usize()); |
| 249 |
|
| 250 |
let r3p = r_lin.get_unchecked(src1[src_cn.r_i() + src_channels]._as_usize()); |
| 251 |
let g3p = g_lin.get_unchecked(src1[src_cn.g_i() + src_channels]._as_usize()); |
| 252 |
let b3p = b_lin.get_unchecked(src1[src_cn.b_i() + src_channels]._as_usize()); |
| 253 |
|
| 254 |
r0 = vld1q_dup_f32(r0p); |
| 255 |
g0 = vld1q_dup_f32(g0p); |
| 256 |
b0 = vld1q_dup_f32(b0p); |
| 257 |
|
| 258 |
r1 = vld1q_dup_f32(r1p); |
| 259 |
g1 = vld1q_dup_f32(g1p); |
| 260 |
b1 = vld1q_dup_f32(b1p); |
| 261 |
|
| 262 |
r2 = vld1q_dup_f32(r2p); |
| 263 |
g2 = vld1q_dup_f32(g2p); |
| 264 |
b2 = vld1q_dup_f32(b2p); |
| 265 |
|
| 266 |
r3 = vld1q_dup_f32(r3p); |
| 267 |
g3 = vld1q_dup_f32(g3p); |
| 268 |
b3 = vld1q_dup_f32(b3p); |
| 269 |
|
| 270 |
dst0[dst_cn.r_i()] = self.profile.r_gamma[temporary0.0[0] as usize]; |
| 271 |
dst0[dst_cn.g_i()] = self.profile.g_gamma[temporary0.0[2] as usize]; |
| 272 |
dst0[dst_cn.b_i()] = self.profile.b_gamma[temporary0.0[4] as usize]; |
| 273 |
if dst_channels == 4 { |
| 274 |
dst0[dst_cn.a_i()] = a0; |
| 275 |
} |
| 276 |
|
| 277 |
dst0[dst_cn.r_i() + dst_channels] = |
| 278 |
self.profile.r_gamma[temporary1.0[0] as usize]; |
| 279 |
dst0[dst_cn.g_i() + dst_channels] = |
| 280 |
self.profile.g_gamma[temporary1.0[2] as usize]; |
| 281 |
dst0[dst_cn.b_i() + dst_channels] = |
| 282 |
self.profile.b_gamma[temporary1.0[4] as usize]; |
| 283 |
if dst_channels == 4 { |
| 284 |
dst0[dst_cn.a_i() + dst_channels] = a1; |
| 285 |
} |
| 286 |
|
| 287 |
dst1[dst_cn.r_i()] = self.profile.r_gamma[temporary2.0[0] as usize]; |
| 288 |
dst1[dst_cn.g_i()] = self.profile.g_gamma[temporary2.0[2] as usize]; |
| 289 |
dst1[dst_cn.b_i()] = self.profile.b_gamma[temporary2.0[4] as usize]; |
| 290 |
if dst_channels == 4 { |
| 291 |
dst1[dst_cn.a_i()] = a2; |
| 292 |
} |
| 293 |
|
| 294 |
dst1[dst_cn.r_i() + dst_channels] = |
| 295 |
self.profile.r_gamma[temporary3.0[0] as usize]; |
| 296 |
dst1[dst_cn.g_i() + dst_channels] = |
| 297 |
self.profile.g_gamma[temporary3.0[2] as usize]; |
| 298 |
dst1[dst_cn.b_i() + dst_channels] = |
| 299 |
self.profile.b_gamma[temporary3.0[4] as usize]; |
| 300 |
if dst_channels == 4 { |
| 301 |
dst1[dst_cn.a_i() + dst_channels] = a3; |
| 302 |
} |
| 303 |
|
| 304 |
a0 = if src_channels == 4 { |
| 305 |
src0[src_cn.a_i()] |
| 306 |
} else { |
| 307 |
max_colors |
| 308 |
}; |
| 309 |
|
| 310 |
a1 = if src_channels == 4 { |
| 311 |
src0[src_cn.a_i() + src_channels] |
| 312 |
} else { |
| 313 |
max_colors |
| 314 |
}; |
| 315 |
|
| 316 |
a2 = if src_channels == 4 { |
| 317 |
src1[src_cn.a_i()] |
| 318 |
} else { |
| 319 |
max_colors |
| 320 |
}; |
| 321 |
|
| 322 |
a3 = if src_channels == 4 { |
| 323 |
src1[src_cn.a_i() + src_channels] |
| 324 |
} else { |
| 325 |
max_colors |
| 326 |
}; |
| 327 |
} |
| 328 |
|
| 329 |
if let (Some(dst0), Some(dst1)) = ( |
| 330 |
dst0.chunks_exact_mut(dst_channels * 2).last(), |
| 331 |
dst1.chunks_exact_mut(dst_channels * 2).last(), |
| 332 |
) { |
| 333 |
let v0_0 = vmulq_f32(r0, m0); |
| 334 |
let v0_1 = vmulq_f32(r1, m0); |
| 335 |
let v0_2 = vmulq_f32(r2, m0); |
| 336 |
let v0_3 = vmulq_f32(r3, m0); |
| 337 |
|
| 338 |
let v1_0 = vfmaq_f32(v0_0, g0, m1); |
| 339 |
let v1_1 = vfmaq_f32(v0_1, g1, m1); |
| 340 |
let v1_2 = vfmaq_f32(v0_2, g2, m1); |
| 341 |
let v1_3 = vfmaq_f32(v0_3, g3, m1); |
| 342 |
|
| 343 |
let mut vr0 = vfmaq_f32(v1_0, b0, m2); |
| 344 |
let mut vr1 = vfmaq_f32(v1_1, b1, m2); |
| 345 |
let mut vr2 = vfmaq_f32(v1_2, b2, m2); |
| 346 |
let mut vr3 = vfmaq_f32(v1_3, b3, m2); |
| 347 |
|
| 348 |
vr0 = vfmaq_f32(rnd, vr0, v_scale); |
| 349 |
vr1 = vfmaq_f32(rnd, vr1, v_scale); |
| 350 |
vr2 = vfmaq_f32(rnd, vr2, v_scale); |
| 351 |
vr3 = vfmaq_f32(rnd, vr3, v_scale); |
| 352 |
|
| 353 |
vr0 = vminq_f32(vr0, v_scale); |
| 354 |
vr1 = vminq_f32(vr1, v_scale); |
| 355 |
vr2 = vminq_f32(vr2, v_scale); |
| 356 |
vr3 = vminq_f32(vr3, v_scale); |
| 357 |
|
| 358 |
let zx0 = vcvtq_u32_f32(vr0); |
| 359 |
let zx1 = vcvtq_u32_f32(vr1); |
| 360 |
let zx2 = vcvtq_u32_f32(vr2); |
| 361 |
let zx3 = vcvtq_u32_f32(vr3); |
| 362 |
vst1q_u32(temporary0.0.as_mut_ptr() as *mut _, zx0); |
| 363 |
vst1q_u32(temporary1.0.as_mut_ptr() as *mut _, zx1); |
| 364 |
vst1q_u32(temporary2.0.as_mut_ptr() as *mut _, zx2); |
| 365 |
vst1q_u32(temporary3.0.as_mut_ptr() as *mut _, zx3); |
| 366 |
|
| 367 |
dst0[dst_cn.r_i()] = self.profile.r_gamma[temporary0.0[0] as usize]; |
| 368 |
dst0[dst_cn.g_i()] = self.profile.g_gamma[temporary0.0[2] as usize]; |
| 369 |
dst0[dst_cn.b_i()] = self.profile.b_gamma[temporary0.0[4] as usize]; |
| 370 |
if dst_channels == 4 { |
| 371 |
dst0[dst_cn.a_i()] = a0; |
| 372 |
} |
| 373 |
|
| 374 |
dst0[dst_cn.r_i() + dst_channels] = |
| 375 |
self.profile.r_gamma[temporary1.0[0] as usize]; |
| 376 |
dst0[dst_cn.g_i() + dst_channels] = |
| 377 |
self.profile.g_gamma[temporary1.0[2] as usize]; |
| 378 |
dst0[dst_cn.b_i() + dst_channels] = |
| 379 |
self.profile.b_gamma[temporary1.0[4] as usize]; |
| 380 |
if dst_channels == 4 { |
| 381 |
dst0[dst_cn.a_i() + dst_channels] = a1; |
| 382 |
} |
| 383 |
|
| 384 |
dst1[dst_cn.r_i()] = self.profile.r_gamma[temporary2.0[0] as usize]; |
| 385 |
dst1[dst_cn.g_i()] = self.profile.g_gamma[temporary2.0[2] as usize]; |
| 386 |
dst1[dst_cn.b_i()] = self.profile.b_gamma[temporary2.0[4] as usize]; |
| 387 |
if dst_channels == 4 { |
| 388 |
dst1[dst_cn.a_i()] = a2; |
| 389 |
} |
| 390 |
|
| 391 |
dst1[dst_cn.r_i() + dst_channels] = |
| 392 |
self.profile.r_gamma[temporary3.0[0] as usize]; |
| 393 |
dst1[dst_cn.g_i() + dst_channels] = |
| 394 |
self.profile.g_gamma[temporary3.0[2] as usize]; |
| 395 |
dst1[dst_cn.b_i() + dst_channels] = |
| 396 |
self.profile.b_gamma[temporary3.0[4] as usize]; |
| 397 |
if dst_channels == 4 { |
| 398 |
dst1[dst_cn.a_i() + dst_channels] = a3; |
| 399 |
} |
| 400 |
} |
| 401 |
} |
| 402 |
|
| 403 |
for (src, dst) in src_remainder |
| 404 |
.chunks_exact(src_channels) |
| 405 |
.zip(dst_remainder.chunks_exact_mut(dst_channels)) |
| 406 |
{ |
| 407 |
let rp = r_lin.get_unchecked(src[src_cn.r_i()]._as_usize()); |
| 408 |
let gp = g_lin.get_unchecked(src[src_cn.g_i()]._as_usize()); |
| 409 |
let bp = b_lin.get_unchecked(src[src_cn.b_i()]._as_usize()); |
| 410 |
let r = vld1q_dup_f32(rp); |
| 411 |
let g = vld1q_dup_f32(gp); |
| 412 |
let b = vld1q_dup_f32(bp); |
| 413 |
let a = if src_channels == 4 { |
| 414 |
src[src_cn.a_i()] |
| 415 |
} else { |
| 416 |
max_colors |
| 417 |
}; |
| 418 |
|
| 419 |
let v0 = vmulq_f32(r, m0); |
| 420 |
let v1 = vfmaq_f32(v0, g, m1); |
| 421 |
let mut v = vfmaq_f32(v1, b, m2); |
| 422 |
|
| 423 |
v = vfmaq_f32(rnd, v, v_scale); |
| 424 |
v = vminq_f32(v, v_scale); |
| 425 |
|
| 426 |
let zx = vcvtq_u32_f32(v); |
| 427 |
vst1q_u32(temporary0.0.as_mut_ptr() as *mut _, zx); |
| 428 |
|
| 429 |
dst[dst_cn.r_i()] = self.profile.r_gamma[temporary0.0[0] as usize]; |
| 430 |
dst[dst_cn.g_i()] = self.profile.g_gamma[temporary0.0[2] as usize]; |
| 431 |
dst[dst_cn.b_i()] = self.profile.b_gamma[temporary0.0[4] as usize]; |
| 432 |
if dst_channels == 4 { |
| 433 |
dst[dst_cn.a_i()] = a; |
| 434 |
} |
| 435 |
} |
| 436 |
} |
| 437 |
|
| 438 |
Ok(()) |
| 439 |
} |
| 440 |
} |