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 */
29#![cfg(feature = "sse_luts")]
30use crate::conversions::LutBarycentricReduction;
31use crate::conversions::interpolator::BarycentricWeight;
32use crate::conversions::sse::interpolator_q0_15::*;
33use crate::transform::PointeeSizeExpressible;
34use crate::{CmsError, DataColorSpace, InterpolationMethod, Layout, TransformExecutor};
35use num_traits::AsPrimitive;
36#[cfg(target_arch = "x86")]
37use std::arch::x86::*;
38#[cfg(target_arch = "x86_64")]
39use std::arch::x86_64::*;
40use std::marker::PhantomData;
41
42pub(crate) struct TransformLut3x3SseQ0_15<
43    T,
44    U,
45    const SRC_LAYOUT: u8,
46    const DST_LAYOUT: u8,
47    const GRID_SIZE: usize,
48    const BIT_DEPTH: usize,
49    const BINS: usize,
50    const BARYCENTRIC_BINS: usize,
51> {
52    pub(crate) lut: Vec<SseAlignedI16x4>,
53    pub(crate) _phantom: PhantomData<T>,
54    pub(crate) _phantom2: PhantomData<U>,
55    pub(crate) interpolation_method: InterpolationMethod,
56    pub(crate) weights: Box<[BarycentricWeight<i16>; BINS]>,
57    pub(crate) color_space: DataColorSpace,
58    pub(crate) is_linear: bool,
59}
60
61impl<
62    T: Copy + AsPrimitive<f32> + Default + PointeeSizeExpressible,
63    U: AsPrimitive<usize>,
64    const SRC_LAYOUT: u8,
65    const DST_LAYOUT: u8,
66    const GRID_SIZE: usize,
67    const BIT_DEPTH: usize,
68    const BINS: usize,
69    const BARYCENTRIC_BINS: usize,
70>
71    TransformLut3x3SseQ0_15<
72        T,
73        U,
74        SRC_LAYOUT,
75        DST_LAYOUT,
76        GRID_SIZE,
77        BIT_DEPTH,
78        BINS,
79        BARYCENTRIC_BINS,
80    >
81where
82    f32: AsPrimitive<T>,
83    u32: AsPrimitive<T>,
84    (): LutBarycentricReduction<T, U>,
85{
86    #[allow(unused_unsafe)]
87    #[target_feature(enable = "sse4.1")]
88    #[inline(never)]
89    unsafe fn transform_chunk(
90        &self,
91        src: &[T],
92        dst: &mut [T],
93        interpolator: Box<dyn SseMdInterpolationQ0_15 + Send + Sync>,
94    ) {
95        unsafe {
96            let src_cn = Layout::from(SRC_LAYOUT);
97            let src_channels = src_cn.channels();
98
99            let dst_cn = Layout::from(DST_LAYOUT);
100            let dst_channels = dst_cn.channels();
101
102            let f_value_scale = _mm_set1_ps(1. / ((1 << 14i32) - 1) as f32);
103            let max_value = ((1u32 << BIT_DEPTH) - 1).as_();
104            let v_max_scale = if T::FINITE {
105                _mm_set1_epi16(((1i32 << BIT_DEPTH) - 1) as i16)
106            } else {
107                _mm_set1_epi16(((1i32 << 14i32) - 1) as i16)
108            };
109
110            for (src, dst) in src
111                .chunks_exact(src_channels)
112                .zip(dst.chunks_exact_mut(dst_channels))
113            {
114                let x = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
115                    src[src_cn.r_i()],
116                );
117                let y = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
118                    src[src_cn.g_i()],
119                );
120                let z = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
121                    src[src_cn.b_i()],
122                );
123
124                let a = if src_channels == 4 {
125                    src[src_cn.a_i()]
126                } else {
127                    max_value
128                };
129
130                let v = interpolator.inter3_sse(
131                    &self.lut,
132                    x.as_(),
133                    y.as_(),
134                    z.as_(),
135                    self.weights.as_slice(),
136                );
137                if T::FINITE {
138                    let mut o = _mm_max_epi16(v.v, _mm_setzero_si128());
139                    o = _mm_min_epi16(o, v_max_scale);
140                    let x = _mm_extract_epi16::<0>(o);
141                    let y = _mm_extract_epi16::<1>(o);
142                    let z = _mm_extract_epi16::<2>(o);
143
144                    dst[dst_cn.r_i()] = (x as u32).as_();
145                    dst[dst_cn.g_i()] = (y as u32).as_();
146                    dst[dst_cn.b_i()] = (z as u32).as_();
147                } else {
148                    let mut r = _mm_cvtepi32_ps(_mm_cvtepi16_epi32(v.v));
149                    r = _mm_mul_ps(r, f_value_scale);
150                    dst[dst_cn.r_i()] = f32::from_bits(_mm_extract_ps::<0>(r) as u32).as_();
151                    dst[dst_cn.g_i()] = f32::from_bits(_mm_extract_ps::<1>(r) as u32).as_();
152                    dst[dst_cn.b_i()] = f32::from_bits(_mm_extract_ps::<2>(r) as u32).as_();
153                }
154                if dst_channels == 4 {
155                    dst[dst_cn.a_i()] = a;
156                }
157            }
158        }
159    }
160}
161
162impl<
163    T: Copy + AsPrimitive<f32> + Default + PointeeSizeExpressible,
164    U: AsPrimitive<usize>,
165    const SRC_LAYOUT: u8,
166    const DST_LAYOUT: u8,
167    const GRID_SIZE: usize,
168    const BIT_DEPTH: usize,
169    const BINS: usize,
170    const BARYCENTRIC_BINS: usize,
171> TransformExecutor<T>
172    for TransformLut3x3SseQ0_15<
173        T,
174        U,
175        SRC_LAYOUT,
176        DST_LAYOUT,
177        GRID_SIZE,
178        BIT_DEPTH,
179        BINS,
180        BARYCENTRIC_BINS,
181    >
182where
183    f32: AsPrimitive<T>,
184    u32: AsPrimitive<T>,
185    (): LutBarycentricReduction<T, U>,
186{
187    fn transform(&self, src: &[T], dst: &mut [T]) -> Result<(), CmsError> {
188        let src_cn = Layout::from(SRC_LAYOUT);
189        let src_channels = src_cn.channels();
190
191        let dst_cn = Layout::from(DST_LAYOUT);
192        let dst_channels = dst_cn.channels();
193        if src.len() % src_channels != 0 {
194            return Err(CmsError::LaneMultipleOfChannels);
195        }
196        if dst.len() % dst_channels != 0 {
197            return Err(CmsError::LaneMultipleOfChannels);
198        }
199        let src_chunks = src.len() / src_channels;
200        let dst_chunks = dst.len() / dst_channels;
201        if src_chunks != dst_chunks {
202            return Err(CmsError::LaneSizeMismatch);
203        }
204
205        unsafe {
206            if self.color_space == DataColorSpace::Lab
207                || (self.is_linear && self.color_space == DataColorSpace::Rgb)
208                || self.color_space == DataColorSpace::Xyz
209            {
210                self.transform_chunk(src, dst, Box::new(TrilinearSseQ0_15::<GRID_SIZE> {}));
211            } else {
212                match self.interpolation_method {
213                    #[cfg(feature = "options")]
214                    InterpolationMethod::Tetrahedral => {
215                        self.transform_chunk(
216                            src,
217                            dst,
218                            Box::new(TetrahedralSseQ0_15::<GRID_SIZE> {}),
219                        );
220                    }
221                    #[cfg(feature = "options")]
222                    InterpolationMethod::Pyramid => {
223                        self.transform_chunk(src, dst, Box::new(PyramidalSseQ0_15::<GRID_SIZE> {}));
224                    }
225                    #[cfg(feature = "options")]
226                    InterpolationMethod::Prism => {
227                        self.transform_chunk(src, dst, Box::new(PrismaticSseQ0_15::<GRID_SIZE> {}));
228                    }
229                    InterpolationMethod::Linear => {
230                        self.transform_chunk(src, dst, Box::new(TrilinearSseQ0_15::<GRID_SIZE> {}));
231                    }
232                }
233            }
234        }
235        Ok(())
236    }
237}
moxcms/conversions/sse/t_lut3_to_3_q0_15.rs

moxcms/conversions/sse/
t_lut3_to_3_q0_15.rs
