moxcms/conversions/sse/
lut4_to_3.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::lut_transforms::Lut4x3Factory;
32use crate::conversions::sse::interpolator::*;
33use crate::conversions::sse::interpolator_q0_15::SseAlignedI16x4;
34use crate::conversions::sse::lut4_to_3_q0_15::TransformLut4To3SseQ0_15;
35use crate::transform::PointeeSizeExpressible;
36use crate::{
37    BarycentricWeightScale, CmsError, DataColorSpace, InterpolationMethod, Layout,
38    TransformExecutor, TransformOptions,
39};
40use num_traits::AsPrimitive;
41#[cfg(target_arch = "x86")]
42use std::arch::x86::*;
43#[cfg(target_arch = "x86_64")]
44use std::arch::x86_64::*;
45use std::marker::PhantomData;
46
47struct TransformLut4To3Sse<
48    T,
49    U,
50    const LAYOUT: u8,
51    const GRID_SIZE: usize,
52    const BIT_DEPTH: usize,
53    const BINS: usize,
54    const BARYCENTRIC_BINS: usize,
55> {
56    lut: Vec<SseAlignedF32>,
57    _phantom: PhantomData<T>,
58    _phantom1: PhantomData<U>,
59    interpolation_method: InterpolationMethod,
60    weights: Box<[BarycentricWeight<f32>; BINS]>,
61    color_space: DataColorSpace,
62    is_linear: bool,
63}
64
65impl<
66    T: Copy + AsPrimitive<f32> + Default + PointeeSizeExpressible,
67    U: AsPrimitive<usize>,
68    const LAYOUT: u8,
69    const GRID_SIZE: usize,
70    const BIT_DEPTH: usize,
71    const BINS: usize,
72    const BARYCENTRIC_BINS: usize,
73> TransformLut4To3Sse<T, U, LAYOUT, GRID_SIZE, BIT_DEPTH, BINS, BARYCENTRIC_BINS>
74where
75    f32: AsPrimitive<T>,
76    u32: AsPrimitive<T>,
77    (): LutBarycentricReduction<T, U>,
78{
79    #[allow(unused_unsafe)]
80    #[target_feature(enable = "sse4.1")]
81    unsafe fn transform_chunk<'b, Interpolator: SseMdInterpolation<'b, GRID_SIZE>>(
82        &'b self,
83        src: &[T],
84        dst: &mut [T],
85    ) {
86        let cn = Layout::from(LAYOUT);
87        let channels = cn.channels();
88        let grid_size = GRID_SIZE as i32;
89        let grid_size3 = grid_size * grid_size * grid_size;
90
91        let value_scale = unsafe { _mm_set1_ps(((1 << BIT_DEPTH) - 1) as f32) };
92        let max_value = ((1 << BIT_DEPTH) - 1u32).as_();
93
94        for (src, dst) in src.chunks_exact(4).zip(dst.chunks_exact_mut(channels)) {
95            let c = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
96                src[0],
97            );
98            let m = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
99                src[1],
100            );
101            let y = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
102                src[2],
103            );
104            let k = <() as LutBarycentricReduction<T, U>>::reduce::<BIT_DEPTH, BARYCENTRIC_BINS>(
105                src[3],
106            );
107
108            let k_weights = self.weights[k.as_()];
109
110            let w: i32 = k_weights.x;
111            let w_n: i32 = k_weights.x_n;
112            let t: f32 = k_weights.w;
113
114            let table1 = &self.lut[(w * grid_size3) as usize..];
115            let table2 = &self.lut[(w_n * grid_size3) as usize..];
116
117            let tetrahedral1 = Interpolator::new(table1);
118            let tetrahedral2 = Interpolator::new(table2);
119            let a0 = tetrahedral1.inter3_sse(c, m, y, &self.weights).v;
120            let b0 = tetrahedral2.inter3_sse(c, m, y, &self.weights).v;
121
122            if T::FINITE {
123                unsafe {
124                    let t0 = _mm_set1_ps(t);
125                    let ones = _mm_set1_ps(1f32);
126                    let hp = _mm_mul_ps(a0, _mm_sub_ps(ones, t0));
127                    let mut v = _mm_add_ps(_mm_mul_ps(b0, t0), hp);
128                    v = _mm_max_ps(v, _mm_setzero_ps());
129                    v = _mm_mul_ps(v, value_scale);
130                    v = _mm_min_ps(v, value_scale);
131                    let jvz = _mm_cvtps_epi32(v);
132
133                    let x = _mm_extract_epi32::<0>(jvz);
134                    let y = _mm_extract_epi32::<1>(jvz);
135                    let z = _mm_extract_epi32::<2>(jvz);
136
137                    dst[cn.r_i()] = (x as u32).as_();
138                    dst[cn.g_i()] = (y as u32).as_();
139                    dst[cn.b_i()] = (z as u32).as_();
140                }
141            } else {
142                unsafe {
143                    let t0 = _mm_set1_ps(t);
144                    let ones = _mm_set1_ps(1f32);
145                    let hp = _mm_mul_ps(a0, _mm_sub_ps(ones, t0));
146                    let v = _mm_add_ps(_mm_mul_ps(b0, t0), hp);
147
148                    dst[cn.r_i()] = f32::from_bits(_mm_extract_ps::<0>(v) as u32).as_();
149                    dst[cn.g_i()] = f32::from_bits(_mm_extract_ps::<1>(v) as u32).as_();
150                    dst[cn.b_i()] = f32::from_bits(_mm_extract_ps::<2>(v) as u32).as_();
151                }
152            }
153            if channels == 4 {
154                dst[cn.a_i()] = max_value;
155            }
156        }
157    }
158}
159
160impl<
161    T: Copy + AsPrimitive<f32> + Default + PointeeSizeExpressible,
162    U: AsPrimitive<usize>,
163    const LAYOUT: u8,
164    const GRID_SIZE: usize,
165    const BIT_DEPTH: usize,
166    const BINS: usize,
167    const BARYCENTRIC_BINS: usize,
168> TransformExecutor<T>
169    for TransformLut4To3Sse<T, U, LAYOUT, GRID_SIZE, BIT_DEPTH, BINS, BARYCENTRIC_BINS>
170where
171    f32: AsPrimitive<T>,
172    u32: AsPrimitive<T>,
173    (): LutBarycentricReduction<T, U>,
174{
175    fn transform(&self, src: &[T], dst: &mut [T]) -> Result<(), CmsError> {
176        let cn = Layout::from(LAYOUT);
177        let channels = cn.channels();
178        if src.len() % 4 != 0 {
179            return Err(CmsError::LaneMultipleOfChannels);
180        }
181        if dst.len() % channels != 0 {
182            return Err(CmsError::LaneMultipleOfChannels);
183        }
184        let src_chunks = src.len() / 4;
185        let dst_chunks = dst.len() / channels;
186        if src_chunks != dst_chunks {
187            return Err(CmsError::LaneSizeMismatch);
188        }
189
190        unsafe {
191            if self.color_space == DataColorSpace::Lab
192                || (self.is_linear && self.color_space == DataColorSpace::Rgb)
193                || self.color_space == DataColorSpace::Xyz
194            {
195                self.transform_chunk::<TrilinearSse<GRID_SIZE>>(src, dst);
196            } else {
197                match self.interpolation_method {
198                    #[cfg(feature = "options")]
199                    InterpolationMethod::Tetrahedral => {
200                        self.transform_chunk::<TetrahedralSse<GRID_SIZE>>(src, dst);
201                    }
202                    #[cfg(feature = "options")]
203                    InterpolationMethod::Pyramid => {
204                        self.transform_chunk::<PyramidalSse<GRID_SIZE>>(src, dst);
205                    }
206                    #[cfg(feature = "options")]
207                    InterpolationMethod::Prism => {
208                        self.transform_chunk::<PrismaticSse<GRID_SIZE>>(src, dst);
209                    }
210                    InterpolationMethod::Linear => {
211                        self.transform_chunk::<TrilinearSse<GRID_SIZE>>(src, dst);
212                    }
213                }
214            }
215        }
216
217        Ok(())
218    }
219}
220
221pub(crate) struct SseLut4x3Factory {}
222
223impl Lut4x3Factory for SseLut4x3Factory {
224    fn make_transform_4x3<
225        T: Copy + AsPrimitive<f32> + Default + PointeeSizeExpressible + 'static + Send + Sync,
226        const LAYOUT: u8,
227        const GRID_SIZE: usize,
228        const BIT_DEPTH: usize,
229    >(
230        lut: Vec<f32>,
231        options: TransformOptions,
232        color_space: DataColorSpace,
233        is_linear: bool,
234    ) -> Box<dyn TransformExecutor<T> + Sync + Send>
235    where
236        f32: AsPrimitive<T>,
237        u32: AsPrimitive<T>,
238        (): LutBarycentricReduction<T, u8>,
239        (): LutBarycentricReduction<T, u16>,
240    {
241        if options.prefer_fixed_point && BIT_DEPTH < 16 {
242            let q: f32 = if T::FINITE {
243                ((1i32 << BIT_DEPTH as i32) - 1) as f32
244            } else {
245                ((1i32 << 14i32) - 1) as f32
246            };
247            let lut = lut
248                .chunks_exact(3)
249                .map(|x| {
250                    SseAlignedI16x4([
251                        (x[0] * q).round() as i16,
252                        (x[1] * q).round() as i16,
253                        (x[2] * q).round() as i16,
254                        0,
255                    ])
256                })
257                .collect::<Vec<_>>();
258            return match options.barycentric_weight_scale {
259                BarycentricWeightScale::Low => Box::new(TransformLut4To3SseQ0_15::<
260                    T,
261                    u8,
262                    LAYOUT,
263                    GRID_SIZE,
264                    BIT_DEPTH,
265                    256,
266                    256,
267                > {
268                    lut,
269                    interpolation_method: options.interpolation_method,
270                    weights: BarycentricWeight::<i16>::create_ranged_256::<GRID_SIZE>(),
271                    _phantom: PhantomData,
272                    _phantom1: PhantomData,
273                    color_space,
274                    is_linear,
275                }),
276                #[cfg(feature = "options")]
277                BarycentricWeightScale::High => Box::new(TransformLut4To3SseQ0_15::<
278                    T,
279                    u16,
280                    LAYOUT,
281                    GRID_SIZE,
282                    BIT_DEPTH,
283                    65536,
284                    65536,
285                > {
286                    lut,
287                    interpolation_method: options.interpolation_method,
288                    weights: BarycentricWeight::<i16>::create_binned::<GRID_SIZE, 65536>(),
289                    _phantom: PhantomData,
290                    _phantom1: PhantomData,
291                    color_space,
292                    is_linear,
293                }),
294            };
295        }
296        let lut = lut
297            .chunks_exact(3)
298            .map(|x| SseAlignedF32([x[0], x[1], x[2], 0f32]))
299            .collect::<Vec<_>>();
300        match options.barycentric_weight_scale {
301            BarycentricWeightScale::Low => {
302                Box::new(
303                    TransformLut4To3Sse::<T, u8, LAYOUT, GRID_SIZE, BIT_DEPTH, 256, 256> {
304                        lut,
305                        _phantom: PhantomData,
306                        _phantom1: PhantomData,
307                        interpolation_method: options.interpolation_method,
308                        weights: BarycentricWeight::<f32>::create_ranged_256::<GRID_SIZE>(),
309                        color_space,
310                        is_linear,
311                    },
312                )
313            }
314            #[cfg(feature = "options")]
315            BarycentricWeightScale::High => {
316                Box::new(
317                    TransformLut4To3Sse::<T, u16, LAYOUT, GRID_SIZE, BIT_DEPTH, 65536, 65536> {
318                        lut,
319                        _phantom: PhantomData,
320                        _phantom1: PhantomData,
321                        interpolation_method: options.interpolation_method,
322                        weights: BarycentricWeight::<f32>::create_binned::<GRID_SIZE, 65536>(),
323                        color_space,
324                        is_linear,
325                    },
326                )
327            }
328        }
329    }
330}
moxcms/conversions/sse/lut4_to_3.rs

moxcms/conversions/sse/
lut4_to_3.rs
