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(
85        &self,
86        src: &[T],
87        dst: &mut [T],
88        interpolator: Box<dyn AvxMdInterpolationQ0_15 + Send + Sync>,
89    ) {
90        unsafe {
91            let src_cn = Layout::from(SRC_LAYOUT);
92            let src_channels = src_cn.channels();
93
94            let dst_cn = Layout::from(DST_LAYOUT);
95            let dst_channels = dst_cn.channels();
96
97            let f_value_scale = _mm_set1_ps(1. / ((1 << 14i32) - 1) as f32);
98            let max_value = ((1u32 << BIT_DEPTH) - 1).as_();
99            let v_max_scale = if T::FINITE {
100                _mm_set1_epi16(((1i32 << BIT_DEPTH) - 1) as i16)
101            } else {
102                _mm_set1_epi16(((1i32 << 14i32) - 1) as i16)
103            };
104
105            for (src, dst) in src
106                .chunks_exact(src_channels)
107                .zip(dst.chunks_exact_mut(dst_channels))
108            {
109                let x = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
110                    src[src_cn.r_i()],
111                );
112                let y = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
113                    src[src_cn.g_i()],
114                );
115                let z = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
116                    src[src_cn.b_i()],
117                );
118
119                let a = if src_channels == 4 {
120                    src[src_cn.a_i()]
121                } else {
122                    max_value
123                };
124
125                let v = interpolator.inter3_sse(
126                    &self.lut,
127                    x.as_(),
128                    y.as_(),
129                    z.as_(),
130                    self.weights.as_slice(),
131                );
132                if T::FINITE {
133                    let mut o = _mm_max_epi16(v.v, _mm_setzero_si128());
134                    o = _mm_min_epi16(o, v_max_scale);
135                    let x = _mm_extract_epi16::<0>(o);
136                    let y = _mm_extract_epi16::<1>(o);
137                    let z = _mm_extract_epi16::<2>(o);
138
139                    dst[dst_cn.r_i()] = (x as u32).as_();
140                    dst[dst_cn.g_i()] = (y as u32).as_();
141                    dst[dst_cn.b_i()] = (z as u32).as_();
142                } else {
143                    let mut r = _mm_cvtepi32_ps(_mm_cvtepi16_epi32(v.v));
144                    r = _mm_mul_ps(r, f_value_scale);
145                    dst[dst_cn.r_i()] = f32::from_bits(_mm_extract_ps::<0>(r) as u32).as_();
146                    dst[dst_cn.g_i()] = f32::from_bits(_mm_extract_ps::<1>(r) as u32).as_();
147                    dst[dst_cn.b_i()] = f32::from_bits(_mm_extract_ps::<2>(r) as u32).as_();
148                }
149                if dst_channels == 4 {
150                    dst[dst_cn.a_i()] = a;
151                }
152            }
153        }
154    }
155}
156
157impl<
158    T: Copy + AsPrimitive<f32> + Default + PointeeSizeExpressible,
159    U: AsPrimitive<usize>,
160    const SRC_LAYOUT: u8,
161    const DST_LAYOUT: u8,
162    const GRID_SIZE: usize,
163    const BIT_DEPTH: usize,
164    const BINS: usize,
165    const BARYCENTRIC_BINS: usize,
166> TransformExecutor<T>
167    for TransformLut3x3AvxQ0_15<
168        T,
169        U,
170        SRC_LAYOUT,
171        DST_LAYOUT,
172        GRID_SIZE,
173        BIT_DEPTH,
174        BINS,
175        BARYCENTRIC_BINS,
176    >
177where
178    f32: AsPrimitive<T>,
179    u32: AsPrimitive<T>,
180    (): LutBarycentricReduction<T, U>,
181{
182    fn transform(&self, src: &[T], dst: &mut [T]) -> Result<(), CmsError> {
183        let src_cn = Layout::from(SRC_LAYOUT);
184        let src_channels = src_cn.channels();
185
186        let dst_cn = Layout::from(DST_LAYOUT);
187        let dst_channels = dst_cn.channels();
188        if src.len() % src_channels != 0 {
189            return Err(CmsError::LaneMultipleOfChannels);
190        }
191        if dst.len() % dst_channels != 0 {
192            return Err(CmsError::LaneMultipleOfChannels);
193        }
194        let src_chunks = src.len() / src_channels;
195        let dst_chunks = dst.len() / dst_channels;
196        if src_chunks != dst_chunks {
197            return Err(CmsError::LaneSizeMismatch);
198        }
199
200        unsafe {
201            if self.color_space == DataColorSpace::Lab
202                || (self.is_linear && self.color_space == DataColorSpace::Rgb)
203                || self.color_space == DataColorSpace::Xyz
204            {
205                self.transform_chunk(src, dst, Box::new(TrilinearAvxQ0_15::<GRID_SIZE> {}));
206            } else {
207                match self.interpolation_method {
208                    #[cfg(feature = "options")]
209                    InterpolationMethod::Tetrahedral => {
210                        self.transform_chunk(
211                            src,
212                            dst,
213                            Box::new(TetrahedralAvxQ0_15::<GRID_SIZE> {}),
214                        );
215                    }
216                    #[cfg(feature = "options")]
217                    InterpolationMethod::Pyramid => {
218                        self.transform_chunk(src, dst, Box::new(PyramidalAvxQ0_15::<GRID_SIZE> {}));
219                    }
220                    #[cfg(feature = "options")]
221                    InterpolationMethod::Prism => {
222                        self.transform_chunk(src, dst, Box::new(PrismaticAvxQ0_15::<GRID_SIZE> {}));
223                    }
224                    InterpolationMethod::Linear => {
225                        self.transform_chunk(src, dst, Box::new(TrilinearAvxQ0_15::<GRID_SIZE> {}));
226                    }
227                }
228            }
229        }
230        Ok(())
231    }
232}
moxcms/conversions/avx/t_lut3_to_3_q0_15.rs

moxcms/conversions/avx/
t_lut3_to_3_q0_15.rs
