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    #[inline(never)]
88    unsafe fn transform_chunk(
89        &self,
90        src: &[T],
91        dst: &mut [T],
92        interpolator: Box<dyn SseMdInterpolationQ0_15 + Send + Sync>,
93    ) {
94        unsafe {
95            let src_cn = Layout::from(SRC_LAYOUT);
96            let src_channels = src_cn.channels();
97
98            let dst_cn = Layout::from(DST_LAYOUT);
99            let dst_channels = dst_cn.channels();
100
101            let f_value_scale = _mm_set1_ps(1. / ((1 << 14i32) - 1) as f32);
102            let max_value = ((1u32 << BIT_DEPTH) - 1).as_();
103            let v_max_scale = if T::FINITE {
104                _mm_set1_epi16(((1i32 << BIT_DEPTH) - 1) as i16)
105            } else {
106                _mm_set1_epi16(((1i32 << 14i32) - 1) as i16)
107            };
108
109            for (src, dst) in src
110                .chunks_exact(src_channels)
111                .zip(dst.chunks_exact_mut(dst_channels))
112            {
113                let x = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
114                    src[src_cn.r_i()],
115                );
116                let y = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
117                    src[src_cn.g_i()],
118                );
119                let z = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
120                    src[src_cn.b_i()],
121                );
122
123                let a = if src_channels == 4 {
124                    src[src_cn.a_i()]
125                } else {
126                    max_value
127                };
128
129                let v = interpolator.inter3_sse(
130                    &self.lut,
131                    x.as_(),
132                    y.as_(),
133                    z.as_(),
134                    self.weights.as_slice(),
135                );
136                if T::FINITE {
137                    let mut o = _mm_max_epi16(v.v, _mm_setzero_si128());
138                    o = _mm_min_epi16(o, v_max_scale);
139                    let x = _mm_extract_epi16::<0>(o);
140                    let y = _mm_extract_epi16::<1>(o);
141                    let z = _mm_extract_epi16::<2>(o);
142
143                    dst[dst_cn.r_i()] = (x as u32).as_();
144                    dst[dst_cn.g_i()] = (y as u32).as_();
145                    dst[dst_cn.b_i()] = (z as u32).as_();
146                } else {
147                    let mut r = _mm_cvtepi32_ps(_mm_cvtepi16_epi32(v.v));
148                    r = _mm_mul_ps(r, f_value_scale);
149                    dst[dst_cn.r_i()] = f32::from_bits(_mm_extract_ps::<0>(r) as u32).as_();
150                    dst[dst_cn.g_i()] = f32::from_bits(_mm_extract_ps::<1>(r) as u32).as_();
151                    dst[dst_cn.b_i()] = f32::from_bits(_mm_extract_ps::<2>(r) as u32).as_();
152                }
153                if dst_channels == 4 {
154                    dst[dst_cn.a_i()] = a;
155                }
156            }
157        }
158    }
159}
160
161impl<
162    T: Copy + AsPrimitive<f32> + Default + PointeeSizeExpressible,
163    U: AsPrimitive<usize>,
164    const SRC_LAYOUT: u8,
165    const DST_LAYOUT: u8,
166    const GRID_SIZE: usize,
167    const BIT_DEPTH: usize,
168    const BINS: usize,
169    const BARYCENTRIC_BINS: usize,
170> TransformExecutor<T>
171    for TransformLut3x3SseQ0_15<
172        T,
173        U,
174        SRC_LAYOUT,
175        DST_LAYOUT,
176        GRID_SIZE,
177        BIT_DEPTH,
178        BINS,
179        BARYCENTRIC_BINS,
180    >
181where
182    f32: AsPrimitive<T>,
183    u32: AsPrimitive<T>,
184    (): LutBarycentricReduction<T, U>,
185{
186    fn transform(&self, src: &[T], dst: &mut [T]) -> Result<(), CmsError> {
187        let src_cn = Layout::from(SRC_LAYOUT);
188        let src_channels = src_cn.channels();
189
190        let dst_cn = Layout::from(DST_LAYOUT);
191        let dst_channels = dst_cn.channels();
192        if src.len() % src_channels != 0 {
193            return Err(CmsError::LaneMultipleOfChannels);
194        }
195        if dst.len() % dst_channels != 0 {
196            return Err(CmsError::LaneMultipleOfChannels);
197        }
198        let src_chunks = src.len() / src_channels;
199        let dst_chunks = dst.len() / dst_channels;
200        if src_chunks != dst_chunks {
201            return Err(CmsError::LaneSizeMismatch);
202        }
203
204        unsafe {
205            if self.color_space == DataColorSpace::Lab
206                || (self.is_linear && self.color_space == DataColorSpace::Rgb)
207                || self.color_space == DataColorSpace::Xyz
208            {
209                self.transform_chunk(src, dst, Box::new(TrilinearSseQ0_15::<GRID_SIZE> {}));
210            } else {
211                match self.interpolation_method {
212                    #[cfg(feature = "options")]
213                    InterpolationMethod::Tetrahedral => {
214                        self.transform_chunk(
215                            src,
216                            dst,
217                            Box::new(TetrahedralSseQ0_15::<GRID_SIZE> {}),
218                        );
219                    }
220                    #[cfg(feature = "options")]
221                    InterpolationMethod::Pyramid => {
222                        self.transform_chunk(src, dst, Box::new(PyramidalSseQ0_15::<GRID_SIZE> {}));
223                    }
224                    #[cfg(feature = "options")]
225                    InterpolationMethod::Prism => {
226                        self.transform_chunk(src, dst, Box::new(PrismaticSseQ0_15::<GRID_SIZE> {}));
227                    }
228                    InterpolationMethod::Linear => {
229                        self.transform_chunk(src, dst, Box::new(TrilinearSseQ0_15::<GRID_SIZE> {}));
230                    }
231                }
232            }
233        }
234        Ok(())
235    }
236}
moxcms/conversions/sse/t_lut3_to_3_q0_15.rs

moxcms/conversions/sse/
t_lut3_to_3_q0_15.rs
