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 */
29#![cfg(feature = "avx_luts")]
30use crate::conversions::LutBarycentricReduction;
31use crate::conversions::avx::interpolator_q0_15::*;
32use crate::conversions::interpolator::BarycentricWeight;
33use crate::transform::PointeeSizeExpressible;
34use crate::{CmsError, DataColorSpace, InterpolationMethod, Layout, TransformExecutor};
35use num_traits::AsPrimitive;
36use std::arch::x86_64::*;
37use std::marker::PhantomData;
38
39pub(crate) struct TransformLut3x3AvxQ0_15<
40    T,
41    U,
42    const SRC_LAYOUT: u8,
43    const DST_LAYOUT: u8,
44    const GRID_SIZE: usize,
45    const BIT_DEPTH: usize,
46    const BINS: usize,
47    const BARYCENTRIC_BINS: usize,
48> {
49    pub(crate) lut: Vec<AvxAlignedI16>,
50    pub(crate) _phantom: PhantomData<T>,
51    pub(crate) _phantom2: PhantomData<U>,
52    pub(crate) interpolation_method: InterpolationMethod,
53    pub(crate) weights: Box<[BarycentricWeight<i16>; BINS]>,
54    pub(crate) color_space: DataColorSpace,
55    pub(crate) is_linear: bool,
56}
57
58impl<
59    T: Copy + AsPrimitive<f32> + Default + PointeeSizeExpressible,
60    U: AsPrimitive<usize>,
61    const SRC_LAYOUT: u8,
62    const DST_LAYOUT: u8,
63    const GRID_SIZE: usize,
64    const BIT_DEPTH: usize,
65    const BINS: usize,
66    const BARYCENTRIC_BINS: usize,
67>
68    TransformLut3x3AvxQ0_15<
69        T,
70        U,
71        SRC_LAYOUT,
72        DST_LAYOUT,
73        GRID_SIZE,
74        BIT_DEPTH,
75        BINS,
76        BARYCENTRIC_BINS,
77    >
78where
79    f32: AsPrimitive<T>,
80    u32: AsPrimitive<T>,
81    (): LutBarycentricReduction<T, U>,
82{
83    #[allow(unused_unsafe)]
84    #[target_feature(enable = "avx2")]
85    unsafe fn transform_chunk(
86        &self,
87        src: &[T],
88        dst: &mut [T],
89        interpolator: Box<dyn AvxMdInterpolationQ0_15 + Send + Sync>,
90    ) {
91        unsafe {
92            let src_cn = Layout::from(SRC_LAYOUT);
93            let src_channels = src_cn.channels();
94
95            let dst_cn = Layout::from(DST_LAYOUT);
96            let dst_channels = dst_cn.channels();
97
98            let f_value_scale = _mm_set1_ps(1. / ((1 << 14i32) - 1) as f32);
99            let max_value = ((1u32 << BIT_DEPTH) - 1).as_();
100            let v_max_scale = if T::FINITE {
101                _mm_set1_epi16(((1i32 << BIT_DEPTH) - 1) as i16)
102            } else {
103                _mm_set1_epi16(((1i32 << 14i32) - 1) as i16)
104            };
105
106            for (src, dst) in src
107                .chunks_exact(src_channels)
108                .zip(dst.chunks_exact_mut(dst_channels))
109            {
110                let x = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
111                    src[src_cn.r_i()],
112                );
113                let y = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
114                    src[src_cn.g_i()],
115                );
116                let z = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
117                    src[src_cn.b_i()],
118                );
119
120                let a = if src_channels == 4 {
121                    src[src_cn.a_i()]
122                } else {
123                    max_value
124                };
125
126                let v = interpolator.inter3_sse(
127                    &self.lut,
128                    x.as_(),
129                    y.as_(),
130                    z.as_(),
131                    self.weights.as_slice(),
132                );
133                if T::FINITE {
134                    let mut o = _mm_max_epi16(v.v, _mm_setzero_si128());
135                    o = _mm_min_epi16(o, v_max_scale);
136                    let x = _mm_extract_epi16::<0>(o);
137                    let y = _mm_extract_epi16::<1>(o);
138                    let z = _mm_extract_epi16::<2>(o);
139
140                    dst[dst_cn.r_i()] = (x as u32).as_();
141                    dst[dst_cn.g_i()] = (y as u32).as_();
142                    dst[dst_cn.b_i()] = (z as u32).as_();
143                } else {
144                    let mut r = _mm_cvtepi32_ps(_mm_cvtepi16_epi32(v.v));
145                    r = _mm_mul_ps(r, f_value_scale);
146                    dst[dst_cn.r_i()] = f32::from_bits(_mm_extract_ps::<0>(r) as u32).as_();
147                    dst[dst_cn.g_i()] = f32::from_bits(_mm_extract_ps::<1>(r) as u32).as_();
148                    dst[dst_cn.b_i()] = f32::from_bits(_mm_extract_ps::<2>(r) as u32).as_();
149                }
150                if dst_channels == 4 {
151                    dst[dst_cn.a_i()] = a;
152                }
153            }
154        }
155    }
156}
157
158impl<
159    T: Copy + AsPrimitive<f32> + Default + PointeeSizeExpressible,
160    U: AsPrimitive<usize>,
161    const SRC_LAYOUT: u8,
162    const DST_LAYOUT: u8,
163    const GRID_SIZE: usize,
164    const BIT_DEPTH: usize,
165    const BINS: usize,
166    const BARYCENTRIC_BINS: usize,
167> TransformExecutor<T>
168    for TransformLut3x3AvxQ0_15<
169        T,
170        U,
171        SRC_LAYOUT,
172        DST_LAYOUT,
173        GRID_SIZE,
174        BIT_DEPTH,
175        BINS,
176        BARYCENTRIC_BINS,
177    >
178where
179    f32: AsPrimitive<T>,
180    u32: AsPrimitive<T>,
181    (): LutBarycentricReduction<T, U>,
182{
183    fn transform(&self, src: &[T], dst: &mut [T]) -> Result<(), CmsError> {
184        let src_cn = Layout::from(SRC_LAYOUT);
185        let src_channels = src_cn.channels();
186
187        let dst_cn = Layout::from(DST_LAYOUT);
188        let dst_channels = dst_cn.channels();
189        if src.len() % src_channels != 0 {
190            return Err(CmsError::LaneMultipleOfChannels);
191        }
192        if dst.len() % dst_channels != 0 {
193            return Err(CmsError::LaneMultipleOfChannels);
194        }
195        let src_chunks = src.len() / src_channels;
196        let dst_chunks = dst.len() / dst_channels;
197        if src_chunks != dst_chunks {
198            return Err(CmsError::LaneSizeMismatch);
199        }
200
201        unsafe {
202            if self.color_space == DataColorSpace::Lab
203                || (self.is_linear && self.color_space == DataColorSpace::Rgb)
204                || self.color_space == DataColorSpace::Xyz
205            {
206                self.transform_chunk(src, dst, Box::new(TrilinearAvxQ0_15::<GRID_SIZE> {}));
207            } else {
208                match self.interpolation_method {
209                    #[cfg(feature = "options")]
210                    InterpolationMethod::Tetrahedral => {
211                        self.transform_chunk(
212                            src,
213                            dst,
214                            Box::new(TetrahedralAvxQ0_15::<GRID_SIZE> {}),
215                        );
216                    }
217                    #[cfg(feature = "options")]
218                    InterpolationMethod::Pyramid => {
219                        self.transform_chunk(src, dst, Box::new(PyramidalAvxQ0_15::<GRID_SIZE> {}));
220                    }
221                    #[cfg(feature = "options")]
222                    InterpolationMethod::Prism => {
223                        self.transform_chunk(src, dst, Box::new(PrismaticAvxQ0_15::<GRID_SIZE> {}));
224                    }
225                    InterpolationMethod::Linear => {
226                        self.transform_chunk(src, dst, Box::new(TrilinearAvxQ0_15::<GRID_SIZE> {}));
227                    }
228                }
229            }
230        }
231        Ok(())
232    }
233}
moxcms/conversions/avx/t_lut3_to_3_q0_15.rs

moxcms/conversions/avx/
t_lut3_to_3_q0_15.rs
