moxcms/conversions/avx/
t_lut3_to_3_q0_15.rs

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 */
29use crate::conversions::LutBarycentricReduction;
30use crate::conversions::avx::interpolator_q0_15::*;
31use crate::conversions::interpolator::BarycentricWeight;
32use crate::transform::PointeeSizeExpressible;
33use crate::{CmsError, DataColorSpace, InterpolationMethod, Layout, TransformExecutor};
34use num_traits::AsPrimitive;
35use std::arch::x86_64::*;
36use std::marker::PhantomData;
37
38pub(crate) struct TransformLut3x3AvxQ0_15<
39    T,
40    U,
41    const SRC_LAYOUT: u8,
42    const DST_LAYOUT: u8,
43    const GRID_SIZE: usize,
44    const BIT_DEPTH: usize,
45    const BINS: usize,
46    const BARYCENTRIC_BINS: usize,
47> {
48    pub(crate) lut: Vec<AvxAlignedI16>,
49    pub(crate) _phantom: PhantomData<T>,
50    pub(crate) _phantom2: PhantomData<U>,
51    pub(crate) interpolation_method: InterpolationMethod,
52    pub(crate) weights: Box<[BarycentricWeight<i16>; BINS]>,
53    pub(crate) color_space: DataColorSpace,
54    pub(crate) is_linear: bool,
55}
56
57impl<
58    T: Copy + AsPrimitive<f32> + Default + PointeeSizeExpressible,
59    U: AsPrimitive<usize>,
60    const SRC_LAYOUT: u8,
61    const DST_LAYOUT: u8,
62    const GRID_SIZE: usize,
63    const BIT_DEPTH: usize,
64    const BINS: usize,
65    const BARYCENTRIC_BINS: usize,
66>
67    TransformLut3x3AvxQ0_15<
68        T,
69        U,
70        SRC_LAYOUT,
71        DST_LAYOUT,
72        GRID_SIZE,
73        BIT_DEPTH,
74        BINS,
75        BARYCENTRIC_BINS,
76    >
77where
78    f32: AsPrimitive<T>,
79    u32: AsPrimitive<T>,
80    (): LutBarycentricReduction<T, U>,
81{
82    #[allow(unused_unsafe)]
83    #[target_feature(enable = "avx2")]
84    unsafe fn transform_chunk<'b, Interpolator: AvxMdInterpolationQ0_15<'b, GRID_SIZE>>(
85        &'b self,
86        src: &[T],
87        dst: &mut [T],
88    ) {
89        unsafe {
90            let src_cn = Layout::from(SRC_LAYOUT);
91            let src_channels = src_cn.channels();
92
93            let dst_cn = Layout::from(DST_LAYOUT);
94            let dst_channels = dst_cn.channels();
95
96            let f_value_scale = _mm_set1_ps(1. / ((1 << 14i32) - 1) as f32);
97            let max_value = ((1u32 << BIT_DEPTH) - 1).as_();
98            let v_max_scale = if T::FINITE {
99                _mm_set1_epi16(((1i32 << BIT_DEPTH) - 1) as i16)
100            } else {
101                _mm_set1_epi16(((1i32 << 14i32) - 1) as i16)
102            };
103
104            for (src, dst) in src
105                .chunks_exact(src_channels)
106                .zip(dst.chunks_exact_mut(dst_channels))
107            {
108                let x = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
109                    src[src_cn.r_i()],
110                );
111                let y = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
112                    src[src_cn.g_i()],
113                );
114                let z = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
115                    src[src_cn.b_i()],
116                );
117
118                let a = if src_channels == 4 {
119                    src[src_cn.a_i()]
120                } else {
121                    max_value
122                };
123
124                let tetrahedral = Interpolator::new(&self.lut);
125                let v = tetrahedral.inter3_sse(x, y, z, &self.weights);
126                if T::FINITE {
127                    let mut o = _mm_max_epi16(v.v, _mm_setzero_si128());
128                    o = _mm_min_epi16(o, v_max_scale);
129                    let x = _mm_extract_epi16::<0>(o);
130                    let y = _mm_extract_epi16::<1>(o);
131                    let z = _mm_extract_epi16::<2>(o);
132
133                    dst[dst_cn.r_i()] = (x as u32).as_();
134                    dst[dst_cn.g_i()] = (y as u32).as_();
135                    dst[dst_cn.b_i()] = (z as u32).as_();
136                } else {
137                    let mut r = _mm_cvtepi32_ps(_mm_cvtepi16_epi32(v.v));
138                    r = _mm_mul_ps(r, f_value_scale);
139                    dst[dst_cn.r_i()] = f32::from_bits(_mm_extract_ps::<0>(r) as u32).as_();
140                    dst[dst_cn.g_i()] = f32::from_bits(_mm_extract_ps::<1>(r) as u32).as_();
141                    dst[dst_cn.b_i()] = f32::from_bits(_mm_extract_ps::<2>(r) as u32).as_();
142                }
143                if dst_channels == 4 {
144                    dst[dst_cn.a_i()] = a;
145                }
146            }
147        }
148    }
149}
150
151impl<
152    T: Copy + AsPrimitive<f32> + Default + PointeeSizeExpressible,
153    U: AsPrimitive<usize>,
154    const SRC_LAYOUT: u8,
155    const DST_LAYOUT: u8,
156    const GRID_SIZE: usize,
157    const BIT_DEPTH: usize,
158    const BINS: usize,
159    const BARYCENTRIC_BINS: usize,
160> TransformExecutor<T>
161    for TransformLut3x3AvxQ0_15<
162        T,
163        U,
164        SRC_LAYOUT,
165        DST_LAYOUT,
166        GRID_SIZE,
167        BIT_DEPTH,
168        BINS,
169        BARYCENTRIC_BINS,
170    >
171where
172    f32: AsPrimitive<T>,
173    u32: AsPrimitive<T>,
174    (): LutBarycentricReduction<T, U>,
175{
176    fn transform(&self, src: &[T], dst: &mut [T]) -> Result<(), CmsError> {
177        let src_cn = Layout::from(SRC_LAYOUT);
178        let src_channels = src_cn.channels();
179
180        let dst_cn = Layout::from(DST_LAYOUT);
181        let dst_channels = dst_cn.channels();
182        if src.len() % src_channels != 0 {
183            return Err(CmsError::LaneMultipleOfChannels);
184        }
185        if dst.len() % dst_channels != 0 {
186            return Err(CmsError::LaneMultipleOfChannels);
187        }
188        let src_chunks = src.len() / src_channels;
189        let dst_chunks = dst.len() / dst_channels;
190        if src_chunks != dst_chunks {
191            return Err(CmsError::LaneSizeMismatch);
192        }
193
194        unsafe {
195            if self.color_space == DataColorSpace::Lab
196                || (self.is_linear && self.color_space == DataColorSpace::Rgb)
197                || self.color_space == DataColorSpace::Xyz
198            {
199                self.transform_chunk::<TrilinearAvxQ0_15<GRID_SIZE>>(src, dst);
200            } else {
201                match self.interpolation_method {
202                    #[cfg(feature = "options")]
203                    InterpolationMethod::Tetrahedral => {
204                        self.transform_chunk::<TetrahedralAvxQ0_15<GRID_SIZE>>(src, dst);
205                    }
206                    #[cfg(feature = "options")]
207                    InterpolationMethod::Pyramid => {
208                        self.transform_chunk::<PyramidalAvxQ0_15<GRID_SIZE>>(src, dst);
209                    }
210                    #[cfg(feature = "options")]
211                    InterpolationMethod::Prism => {
212                        self.transform_chunk::<PrismaticAvxQ0_15<GRID_SIZE>>(src, dst);
213                    }
214                    InterpolationMethod::Linear => {
215                        self.transform_chunk::<TrilinearAvxQ0_15<GRID_SIZE>>(src, dst);
216                    }
217                }
218            }
219        }
220        Ok(())
221    }
222}
moxcms/conversions/avx/t_lut3_to_3_q0_15.rs

moxcms/conversions/avx/
t_lut3_to_3_q0_15.rs
