Debian Codegraph @

.cargo/vendor/moxcms/src/conversions/neon/lut4_to_3_q0_15.rs

raw
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::LutBarycentricReduction;
30 use crate::conversions::interpolator::BarycentricWeight;
31 use crate::conversions::neon::interpolator_q0_15::*;
32 use crate::transform::PointeeSizeExpressible;
33 use crate::{CmsError, DataColorSpace, InterpolationMethod, Layout, TransformExecutor};
34 use num_traits::AsPrimitive;
35 use std::arch::aarch64::*;
36 use std::marker::PhantomData;
37
38 pub(crate) struct TransformLut4To3NeonQ0_15<
39 T,
40 U,
41 const LAYOUT: u8,
42 const GRID_SIZE: usize,
43 const BIT_DEPTH: usize,
44 const BINS: usize,
45 const BARYCENTRIC_BINS: usize,
46 > {
47 pub(crate) lut: Vec<NeonAlignedI16x4>,
48 pub(crate) _phantom: PhantomData<T>,
49 pub(crate) _phantom1: PhantomData<U>,
50 pub(crate) interpolation_method: InterpolationMethod,
51 pub(crate) weights: Box<[BarycentricWeight<i16>; BINS]>,
52 pub(crate) color_space: DataColorSpace,
53 pub(crate) is_linear: bool,
54 }
55
56 impl<
57 T: Copy + AsPrimitive<f32> + Default + PointeeSizeExpressible,
58 U: AsPrimitive<usize>,
59 const LAYOUT: u8,
60 const GRID_SIZE: usize,
61 const BIT_DEPTH: usize,
62 const BINS: usize,
63 const BARYCENTRIC_BINS: usize,
64 > TransformLut4To3NeonQ0_15<T, U, LAYOUT, GRID_SIZE, BIT_DEPTH, BINS, BARYCENTRIC_BINS>
65 where
66 f32: AsPrimitive<T>,
67 u32: AsPrimitive<T>,
68 (): LutBarycentricReduction<T, U>,
69 {
70 #[allow(unused_unsafe)]
71 #[target_feature(enable = "rdm")]
72 #[inline(never)]
73 unsafe fn transform_chunk(
74 &self,
75 src: &[T],
76 dst: &mut [T],
77 interpolator: Box<dyn NeonMdInterpolationQ0_15Double + Send + Sync>,
78 ) {
79 unsafe {
80 let cn = Layout::from(LAYOUT);
81 let channels = cn.channels();
82 let grid_size = GRID_SIZE as i32;
83 let grid_size3 = grid_size * grid_size * grid_size;
84
85 let f_value_scale = vdupq_n_f32(1. / ((1 << 14i32) - 1) as f32);
86 let max_value = ((1u32 << BIT_DEPTH) - 1).as_();
87 let v_max_scale = if T::FINITE {
88 vdup_n_s16(((1i32 << BIT_DEPTH) - 1) as i16)
89 } else {
90 vdup_n_s16(((1i32 << 14i32) - 1) as i16)
91 };
92
93 for (src, dst) in src.chunks_exact(4).zip(dst.chunks_exact_mut(channels)) {
94 let c = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
95 src[0],
96 );
97 let m = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
98 src[1],
99 );
100 let y = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
101 src[2],
102 );
103 let k = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
104 src[3],
105 );
106
107 let k_weights = self.weights[k.as_()];
108
109 let w: i32 = k_weights.x;
110 let w_n: i32 = k_weights.x_n;
111 let t: i16 = k_weights.w;
112
113 let table1 = &self.lut[(w * grid_size3) as usize..];
114 let table2 = &self.lut[(w_n * grid_size3) as usize..];
115
116 let (a0, b0) = interpolator.inter3_neon(
117 table1,
118 table2,
119 c.as_(),
120 m.as_(),
121 y.as_(),
122 self.weights.as_slice(),
123 );
124 let (a0, b0) = (a0.v, b0.v);
125
126 let t0 = vdup_n_s16(t);
127 let hp = vqrdmlsh_s16(a0, a0, t0);
128 let mut v = vqrdmlah_s16(hp, b0, t0);
129
130 if T::FINITE {
131 v = vmax_s16(v, vdup_n_s16(0));
132 v = vmin_s16(v, v_max_scale);
133 dst[cn.r_i()] = (vget_lane_s16::<0>(v) as u32).as_();
134 dst[cn.g_i()] = (vget_lane_s16::<1>(v) as u32).as_();
135 dst[cn.b_i()] = (vget_lane_s16::<2>(v) as u32).as_();
136 } else {
137 let o = vcvtq_f32_s32(vmovl_s16(v));
138 let r = vmulq_f32(o, f_value_scale);
139 dst[cn.r_i()] = vgetq_lane_f32::<0>(r).as_();
140 dst[cn.g_i()] = vgetq_lane_f32::<1>(r).as_();
141 dst[cn.b_i()] = vgetq_lane_f32::<2>(r).as_();
142 }
143 if channels == 4 {
144 dst[cn.a_i()] = max_value;
145 }
146 }
147 }
148 }
149 }
150
151 impl<
152 T: Copy + AsPrimitive<f32> + Default + PointeeSizeExpressible,
153 U: AsPrimitive<usize>,
154 const LAYOUT: u8,
155 const GRID_SIZE: usize,
156 const BIT_DEPTH: usize,
157 const BINS: usize,
158 const BARYCENTRIC_BINS: usize,
159 > TransformExecutor<T>
160 for TransformLut4To3NeonQ0_15<T, U, LAYOUT, GRID_SIZE, BIT_DEPTH, BINS, BARYCENTRIC_BINS>
161 where
162 f32: AsPrimitive<T>,
163 u32: AsPrimitive<T>,
164 (): LutBarycentricReduction<T, U>,
165 {
166 fn transform(&self, src: &[T], dst: &mut [T]) -> Result<(), CmsError> {
167 let cn = Layout::from(LAYOUT);
168 let channels = cn.channels();
169 if src.len() % 4 != 0 {
170 return Err(CmsError::LaneMultipleOfChannels);
171 }
172 if dst.len() % channels != 0 {
173 return Err(CmsError::LaneMultipleOfChannels);
174 }
175 let src_chunks = src.len() / 4;
176 let dst_chunks = dst.len() / channels;
177 if src_chunks != dst_chunks {
178 return Err(CmsError::LaneSizeMismatch);
179 }
180
181 unsafe {
182 if self.color_space == DataColorSpace::Lab
183 || (self.is_linear && self.color_space == DataColorSpace::Rgb)
184 || self.color_space == DataColorSpace::Xyz
185 {
186 self.transform_chunk(src, dst, Box::new(TrilinearNeonQ0_15Double::<GRID_SIZE> {}));
187 } else {
188 match self.interpolation_method {
189 #[cfg(feature = "options")]
190 InterpolationMethod::Tetrahedral => {
191 self.transform_chunk(
192 src,
193 dst,
194 Box::new(TetrahedralNeonQ0_15Double::<GRID_SIZE> {}),
195 );
196 }
197 #[cfg(feature = "options")]
198 InterpolationMethod::Pyramid => {
199 self.transform_chunk(
200 src,
201 dst,
202 Box::new(PyramidalNeonQ0_15Double::<GRID_SIZE> {}),
203 );
204 }
205 #[cfg(feature = "options")]
206 InterpolationMethod::Prism => {
207 self.transform_chunk(
208 src,
209 dst,
210 Box::new(PrismaticNeonQ0_15Double::<GRID_SIZE> {}),
211 );
212 }
213 InterpolationMethod::Linear => {
214 self.transform_chunk(
215 src,
216 dst,
217 Box::new(TrilinearNeonQ0_15Double::<GRID_SIZE> {}),
218 );
219 }
220 }
221 }
222 }
223
224 Ok(())
225 }
226 }