moxcms/conversions/sse/
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::interpolator::BarycentricWeight;
31use crate::conversions::sse::interpolator_q0_15::*;
32use crate::transform::PointeeSizeExpressible;
33use crate::{CmsError, DataColorSpace, InterpolationMethod, Layout, TransformExecutor};
34use num_traits::AsPrimitive;
35#[cfg(target_arch = "x86")]
36use std::arch::x86::*;
37#[cfg(target_arch = "x86_64")]
38use std::arch::x86_64::*;
39use std::marker::PhantomData;
40
41pub(crate) struct TransformLut3x3SseQ0_15<
42    T,
43    U,
44    const SRC_LAYOUT: u8,
45    const DST_LAYOUT: u8,
46    const GRID_SIZE: usize,
47    const BIT_DEPTH: usize,
48    const BINS: usize,
49    const BARYCENTRIC_BINS: usize,
50> {
51    pub(crate) lut: Vec<SseAlignedI16x4>,
52    pub(crate) _phantom: PhantomData<T>,
53    pub(crate) _phantom2: PhantomData<U>,
54    pub(crate) interpolation_method: InterpolationMethod,
55    pub(crate) weights: Box<[BarycentricWeight<i16>; BINS]>,
56    pub(crate) color_space: DataColorSpace,
57    pub(crate) is_linear: bool,
58}
59
60impl<
61    T: Copy + AsPrimitive<f32> + Default + PointeeSizeExpressible,
62    U: AsPrimitive<usize>,
63    const SRC_LAYOUT: u8,
64    const DST_LAYOUT: u8,
65    const GRID_SIZE: usize,
66    const BIT_DEPTH: usize,
67    const BINS: usize,
68    const BARYCENTRIC_BINS: usize,
69>
70    TransformLut3x3SseQ0_15<
71        T,
72        U,
73        SRC_LAYOUT,
74        DST_LAYOUT,
75        GRID_SIZE,
76        BIT_DEPTH,
77        BINS,
78        BARYCENTRIC_BINS,
79    >
80where
81    f32: AsPrimitive<T>,
82    u32: AsPrimitive<T>,
83    (): LutBarycentricReduction<T, U>,
84{
85    #[allow(unused_unsafe)]
86    #[target_feature(enable = "sse4.1")]
87    unsafe fn transform_chunk<'b, Interpolator: SseMdInterpolationQ0_15<'b, GRID_SIZE>>(
88        &'b self,
89        src: &[T],
90        dst: &mut [T],
91    ) {
92        unsafe {
93            let src_cn = Layout::from(SRC_LAYOUT);
94            let src_channels = src_cn.channels();
95
96            let dst_cn = Layout::from(DST_LAYOUT);
97            let dst_channels = dst_cn.channels();
98
99            let f_value_scale = _mm_set1_ps(1. / ((1 << 14i32) - 1) as f32);
100            let max_value = ((1u32 << BIT_DEPTH) - 1).as_();
101            let v_max_scale = if T::FINITE {
102                _mm_set1_epi16(((1i32 << BIT_DEPTH) - 1) as i16)
103            } else {
104                _mm_set1_epi16(((1i32 << 14i32) - 1) as i16)
105            };
106
107            for (src, dst) in src
108                .chunks_exact(src_channels)
109                .zip(dst.chunks_exact_mut(dst_channels))
110            {
111                let x = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
112                    src[src_cn.r_i()],
113                );
114                let y = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
115                    src[src_cn.g_i()],
116                );
117                let z = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
118                    src[src_cn.b_i()],
119                );
120
121                let a = if src_channels == 4 {
122                    src[src_cn.a_i()]
123                } else {
124                    max_value
125                };
126
127                let tetrahedral = Interpolator::new(&self.lut);
128                let v = tetrahedral.inter3_sse(x, y, z, &self.weights);
129                if T::FINITE {
130                    let mut o = _mm_max_epi16(v.v, _mm_setzero_si128());
131                    o = _mm_min_epi16(o, v_max_scale);
132                    let x = _mm_extract_epi16::<0>(o);
133                    let y = _mm_extract_epi16::<1>(o);
134                    let z = _mm_extract_epi16::<2>(o);
135
136                    dst[dst_cn.r_i()] = (x as u32).as_();
137                    dst[dst_cn.g_i()] = (y as u32).as_();
138                    dst[dst_cn.b_i()] = (z as u32).as_();
139                } else {
140                    let mut r = _mm_cvtepi32_ps(_mm_cvtepi16_epi32(v.v));
141                    r = _mm_mul_ps(r, f_value_scale);
142                    dst[dst_cn.r_i()] = f32::from_bits(_mm_extract_ps::<0>(r) as u32).as_();
143                    dst[dst_cn.g_i()] = f32::from_bits(_mm_extract_ps::<1>(r) as u32).as_();
144                    dst[dst_cn.b_i()] = f32::from_bits(_mm_extract_ps::<2>(r) as u32).as_();
145                }
146                if dst_channels == 4 {
147                    dst[dst_cn.a_i()] = a;
148                }
149            }
150        }
151    }
152}
153
154impl<
155    T: Copy + AsPrimitive<f32> + Default + PointeeSizeExpressible,
156    U: AsPrimitive<usize>,
157    const SRC_LAYOUT: u8,
158    const DST_LAYOUT: u8,
159    const GRID_SIZE: usize,
160    const BIT_DEPTH: usize,
161    const BINS: usize,
162    const BARYCENTRIC_BINS: usize,
163> TransformExecutor<T>
164    for TransformLut3x3SseQ0_15<
165        T,
166        U,
167        SRC_LAYOUT,
168        DST_LAYOUT,
169        GRID_SIZE,
170        BIT_DEPTH,
171        BINS,
172        BARYCENTRIC_BINS,
173    >
174where
175    f32: AsPrimitive<T>,
176    u32: AsPrimitive<T>,
177    (): LutBarycentricReduction<T, U>,
178{
179    fn transform(&self, src: &[T], dst: &mut [T]) -> Result<(), CmsError> {
180        let src_cn = Layout::from(SRC_LAYOUT);
181        let src_channels = src_cn.channels();
182
183        let dst_cn = Layout::from(DST_LAYOUT);
184        let dst_channels = dst_cn.channels();
185        if src.len() % src_channels != 0 {
186            return Err(CmsError::LaneMultipleOfChannels);
187        }
188        if dst.len() % dst_channels != 0 {
189            return Err(CmsError::LaneMultipleOfChannels);
190        }
191        let src_chunks = src.len() / src_channels;
192        let dst_chunks = dst.len() / dst_channels;
193        if src_chunks != dst_chunks {
194            return Err(CmsError::LaneSizeMismatch);
195        }
196
197        unsafe {
198            if self.color_space == DataColorSpace::Lab
199                || (self.is_linear && self.color_space == DataColorSpace::Rgb)
200                || self.color_space == DataColorSpace::Xyz
201            {
202                self.transform_chunk::<TrilinearSseQ0_15<GRID_SIZE>>(src, dst);
203            } else {
204                match self.interpolation_method {
205                    #[cfg(feature = "options")]
206                    InterpolationMethod::Tetrahedral => {
207                        self.transform_chunk::<TetrahedralSseQ0_15<GRID_SIZE>>(src, dst);
208                    }
209                    #[cfg(feature = "options")]
210                    InterpolationMethod::Pyramid => {
211                        self.transform_chunk::<PyramidalSseQ0_15<GRID_SIZE>>(src, dst);
212                    }
213                    #[cfg(feature = "options")]
214                    InterpolationMethod::Prism => {
215                        self.transform_chunk::<PrismaticSseQ0_15<GRID_SIZE>>(src, dst);
216                    }
217                    InterpolationMethod::Linear => {
218                        self.transform_chunk::<TrilinearSseQ0_15<GRID_SIZE>>(src, dst);
219                    }
220                }
221            }
222        }
223        Ok(())
224    }
225}
moxcms/conversions/sse/t_lut3_to_3_q0_15.rs

moxcms/conversions/sse/
t_lut3_to_3_q0_15.rs
