| 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 |
use crate::conversions::rgbxyz_fixed::TransformMatrixShaperFpOptVec; |
| 30 |
use crate::conversions::sse::rgb_xyz::SseAlignedU16; |
| 31 |
use crate::conversions::sse::rgb_xyz_q2_13::_xmm_load_epi32; |
| 32 |
use crate::transform::PointeeSizeExpressible; |
| 33 |
use crate::{CmsError, Layout, TransformExecutor}; |
| 34 |
use num_traits::AsPrimitive; |
| 35 |
#[cfg(target_arch = "x86")] |
| 36 |
use std::arch::x86::*; |
| 37 |
#[cfg(target_arch = "x86_64")] |
| 38 |
use std::arch::x86_64::*; |
| 39 |
|
| 40 |
pub(crate) struct TransformShaperQ2_13OptSse< |
| 41 |
T: Copy, |
| 42 |
const SRC_LAYOUT: u8, |
| 43 |
const DST_LAYOUT: u8, |
| 44 |
const PRECISION: i32, |
| 45 |
> { |
| 46 |
pub(crate) profile: TransformMatrixShaperFpOptVec<i32, i16, T>, |
| 47 |
pub(crate) bit_depth: usize, |
| 48 |
pub(crate) gamma_lut: usize, |
| 49 |
} |
| 50 |
|
| 51 |
impl< |
| 52 |
T: Copy + PointeeSizeExpressible + 'static, |
| 53 |
const SRC_LAYOUT: u8, |
| 54 |
const DST_LAYOUT: u8, |
| 55 |
const PRECISION: i32, |
| 56 |
> TransformShaperQ2_13OptSse<T, SRC_LAYOUT, DST_LAYOUT, PRECISION> |
| 57 |
where |
| 58 |
u32: AsPrimitive<T>, |
| 59 |
{ |
| 60 |
#[target_feature(enable = "sse4.1")] |
| 61 |
unsafe fn transform_impl(&self, src: &[T], dst: &mut [T]) -> Result<(), CmsError> { |
| 62 |
let src_cn = Layout::from(SRC_LAYOUT); |
| 63 |
let dst_cn = Layout::from(DST_LAYOUT); |
| 64 |
let src_channels = src_cn.channels(); |
| 65 |
let dst_channels = dst_cn.channels(); |
| 66 |
|
| 67 |
let mut temporary = SseAlignedU16([0; 8]); |
| 68 |
|
| 69 |
if src.len() / src_channels != dst.len() / dst_channels { |
| 70 |
return Err(CmsError::LaneSizeMismatch); |
| 71 |
} |
| 72 |
if src.len() % src_channels != 0 { |
| 73 |
return Err(CmsError::LaneMultipleOfChannels); |
| 74 |
} |
| 75 |
if dst.len() % dst_channels != 0 { |
| 76 |
return Err(CmsError::LaneMultipleOfChannels); |
| 77 |
} |
| 78 |
|
| 79 |
let t = self.profile.adaptation_matrix.transpose(); |
| 80 |
|
| 81 |
let max_colors = ((1 << self.bit_depth) - 1).as_(); |
| 82 |
|
| 83 |
unsafe { |
| 84 |
let m0 = _mm_setr_epi16( |
| 85 |
t.v[0][0], t.v[1][0], t.v[0][1], t.v[1][1], t.v[0][2], t.v[1][2], 0, 0, |
| 86 |
); |
| 87 |
let m2 = _mm_setr_epi16(t.v[2][0], 1, t.v[2][1], 1, t.v[2][2], 1, 0, 0); |
| 88 |
|
| 89 |
let rnd_val = ((1i32 << (PRECISION - 1)) as i16).to_ne_bytes(); |
| 90 |
let rnd = _mm_set1_epi32(i32::from_ne_bytes([0, 0, rnd_val[0], rnd_val[1]])); |
| 91 |
|
| 92 |
let v_max_value = _mm_set1_epi32(self.gamma_lut as i32 - 1); |
| 93 |
|
| 94 |
// safety precondition for linearization table |
| 95 |
if T::FINITE { |
| 96 |
let cap = (1 << self.bit_depth) - 1; |
| 97 |
assert!(self.profile.linear.len() >= cap); |
| 98 |
} else { |
| 99 |
assert!(self.profile.linear.len() >= T::NOT_FINITE_LINEAR_TABLE_SIZE); |
| 100 |
} |
| 101 |
|
| 102 |
let lut_lin = &self.profile.linear; |
| 103 |
|
| 104 |
for (src, dst) in src |
| 105 |
.chunks_exact(src_channels) |
| 106 |
.zip(dst.chunks_exact_mut(dst_channels)) |
| 107 |
{ |
| 108 |
let rp = lut_lin.get_unchecked(src[src_cn.r_i()]._as_usize()); |
| 109 |
let gp = lut_lin.get_unchecked(src[src_cn.g_i()]._as_usize()); |
| 110 |
let bp = lut_lin.get_unchecked(src[src_cn.b_i()]._as_usize()); |
| 111 |
|
| 112 |
let mut r = _xmm_load_epi32(rp); |
| 113 |
let mut g = _xmm_load_epi32(gp); |
| 114 |
let mut b = _xmm_load_epi32(bp); |
| 115 |
let a = if src_channels == 4 { |
| 116 |
src[src_cn.a_i()] |
| 117 |
} else { |
| 118 |
max_colors |
| 119 |
}; |
| 120 |
|
| 121 |
r = _mm_shuffle_epi32::<0>(r); |
| 122 |
g = _mm_shuffle_epi32::<0>(g); |
| 123 |
b = _mm_shuffle_epi32::<0>(b); |
| 124 |
|
| 125 |
g = _mm_slli_epi32::<16>(g); |
| 126 |
|
| 127 |
let zrg0 = _mm_or_si128(r, g); |
| 128 |
let zbz0 = _mm_or_si128(b, rnd); |
| 129 |
|
| 130 |
let v0 = _mm_madd_epi16(zrg0, m0); |
| 131 |
let v1 = _mm_madd_epi16(zbz0, m2); |
| 132 |
|
| 133 |
let mut v = _mm_add_epi32(v0, v1); |
| 134 |
|
| 135 |
v = _mm_srai_epi32::<PRECISION>(v); |
| 136 |
v = _mm_max_epi32(v, _mm_setzero_si128()); |
| 137 |
v = _mm_min_epi32(v, v_max_value); |
| 138 |
|
| 139 |
_mm_store_si128(temporary.0.as_mut_ptr() as *mut _, v); |
| 140 |
|
| 141 |
dst[dst_cn.r_i()] = self.profile.gamma[temporary.0[0] as usize]; |
| 142 |
dst[dst_cn.g_i()] = self.profile.gamma[temporary.0[2] as usize]; |
| 143 |
dst[dst_cn.b_i()] = self.profile.gamma[temporary.0[4] as usize]; |
| 144 |
if dst_channels == 4 { |
| 145 |
dst[dst_cn.a_i()] = a; |
| 146 |
} |
| 147 |
} |
| 148 |
} |
| 149 |
|
| 150 |
Ok(()) |
| 151 |
} |
| 152 |
} |
| 153 |
|
| 154 |
impl< |
| 155 |
T: Copy + PointeeSizeExpressible + 'static + Default, |
| 156 |
const SRC_LAYOUT: u8, |
| 157 |
const DST_LAYOUT: u8, |
| 158 |
const PRECISION: i32, |
| 159 |
> TransformExecutor<T> for TransformShaperQ2_13OptSse<T, SRC_LAYOUT, DST_LAYOUT, PRECISION> |
| 160 |
where |
| 161 |
u32: AsPrimitive<T>, |
| 162 |
{ |
| 163 |
fn transform(&self, src: &[T], dst: &mut [T]) -> Result<(), CmsError> { |
| 164 |
unsafe { self.transform_impl(src, dst) } |
| 165 |
} |
| 166 |
} |