pic-scale 0.7.5

/*
 * Copyright (c) Radzivon Bartoshyk 01/2025. All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without modification,
 * are permitted provided that the following conditions are met:
 *
 * 1.  Redistributions of source code must retain the above copyright notice, this
 * list of conditions and the following disclaimer.
 *
 * 2.  Redistributions in binary form must reproduce the above copyright notice,
 * this list of conditions and the following disclaimer in the documentation
 * and/or other materials provided with the distribution.
 *
 * 3.  Neither the name of the copyright holder nor the names of its
 * contributors may be used to endorse or promote products derived from
 * this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

use crate::avx2::utils::{_mm_dot16_avx_epi32, _mm256_dot16_avx_epi32};
use crate::filter_weights::FilterWeights;
use crate::support::{PRECISION, ROUNDING_CONST};
use std::arch::x86_64::*;

#[inline(always)]
fn acc_1_dot<const D: bool>(
    start_x: usize,
    src: &[u16],
    w0: __m128i,
    store: __m128i,
    shuffle: __m128i,
) -> __m128i {
    unsafe {
        const CN: usize = 4;
        let src_ptr = src.get_unchecked((start_x * CN)..);
        let rgba_pixel = _mm_loadu_si64(src_ptr.as_ptr() as *const u8);
        _mm_dot16_avx_epi32::<D>(store, _mm_shuffle_epi8(rgba_pixel, shuffle), w0)
    }
}

#[inline(always)]
fn acc_2_dot<const D: bool>(
    start_x: usize,
    src: &[u16],
    w0: __m128i,
    store: __m128i,
    shuffle: __m128i,
) -> __m128i {
    unsafe {
        const CN: usize = 4;
        let src_ptr = src.get_unchecked((start_x * CN)..);
        let rgba_pixel = _mm_loadu_si128(src_ptr.as_ptr().cast());
        _mm_dot16_avx_epi32::<D>(store, _mm_shuffle_epi8(rgba_pixel, shuffle), w0)
    }
}

#[inline(always)]
fn acc_4_dot<const D: bool>(
    start_x: usize,
    src: &[u16],
    w0: __m256i,
    store: __m256i,
    shuffle: __m256i,
) -> __m256i {
    unsafe {
        const CN: usize = 4;
        let src_ptr = src.get_unchecked((start_x * CN)..);
        let rgba_pixel = _mm256_loadu_si256(src_ptr.as_ptr().cast());
        _mm256_dot16_avx_epi32::<D>(store, _mm256_shuffle_epi8(rgba_pixel, shuffle), w0)
    }
}

#[inline(always)]
fn acc_8_dot<const D: bool>(
    start_x: usize,
    src: &[u16],
    w0: __m256i,
    w1: __m256i,
    store: __m256i,
    shuffle: __m256i,
) -> __m256i {
    unsafe {
        const CN: usize = 4;
        let src_ptr = src.get_unchecked((start_x * CN)..);
        let rgba_pixel0 = _mm256_loadu_si256(src_ptr.as_ptr().cast());
        let rgba_pixel1 = _mm256_loadu_si256(src_ptr.get_unchecked(16..).as_ptr().cast());

        let p0 = _mm256_dot16_avx_epi32::<D>(store, _mm256_shuffle_epi8(rgba_pixel0, shuffle), w0);
        _mm256_dot16_avx_epi32::<D>(p0, _mm256_shuffle_epi8(rgba_pixel1, shuffle), w1)
    }
}

pub(crate) fn convolve_horizontal_rgba_avx_rows_4_u16(
    src: &[u16],
    src_stride: usize,
    dst: &mut [u16],
    dst_stride: usize,
    filter_weights: &FilterWeights<i16>,
    bit_depth: u32,
) {
    unsafe {
        convolve_horizontal_rgba_avx_rows_4_lb_a(
            src,
            src_stride,
            dst,
            dst_stride,
            filter_weights,
            bit_depth,
        )
    }
}

#[cfg(feature = "avx512")]
pub(crate) fn convolve_horizontal_rgba_avx_rows_4_u16_vnni(
    src: &[u16],
    src_stride: usize,
    dst: &mut [u16],
    dst_stride: usize,
    filter_weights: &FilterWeights<i16>,
    bit_depth: u32,
) {
    unsafe {
        convolve_horizontal_rgba_avx_rows_4_lb_vn(
            src,
            src_stride,
            dst,
            dst_stride,
            filter_weights,
            bit_depth,
        )
    }
}

#[cfg(feature = "avx512")]
#[target_feature(enable = "avxvnni", enable = "avx2")]
fn convolve_horizontal_rgba_avx_rows_4_lb_vn(
    src: &[u16],
    src_stride: usize,
    dst: &mut [u16],
    dst_stride: usize,
    filter_weights: &FilterWeights<i16>,
    bit_depth: u32,
) {
    let unit = Row4ExecutionHandler::<true>::default();
    unit.pass(src, src_stride, dst, dst_stride, filter_weights, bit_depth);
}

#[target_feature(enable = "avx2")]
fn convolve_horizontal_rgba_avx_rows_4_lb_a(
    src: &[u16],
    src_stride: usize,
    dst: &mut [u16],
    dst_stride: usize,
    filter_weights: &FilterWeights<i16>,
    bit_depth: u32,
) {
    let unit = Row4ExecutionHandler::<false>::default();
    unit.pass(src, src_stride, dst, dst_stride, filter_weights, bit_depth);
}

#[derive(Copy, Clone, Default)]
struct Row4ExecutionHandler<const D: bool> {}

impl<const D: bool> Row4ExecutionHandler<D> {
    #[inline(always)]
    fn acc_1_dot_avx(
        &self,
        start_x: usize,
        src0: &[u16],
        src1: &[u16],
        w0: __m256i,
        store: __m256i,
        shuffle: __m256i,
    ) -> __m256i {
        unsafe {
            const CN: usize = 4;

            let src_ptr0 = src0.get_unchecked((start_x * CN)..);
            let src_ptr1 = src1.get_unchecked((start_x * CN)..);

            let rgba_pixel0 = _mm_loadu_si64(src_ptr0.as_ptr().cast());
            let rgba_pixel1 = _mm_loadu_si64(src_ptr1.as_ptr().cast());

            let rgba_pixel =
                _mm256_inserti128_si256::<1>(_mm256_castsi128_si256(rgba_pixel0), rgba_pixel1);

            _mm256_dot16_avx_epi32::<D>(store, _mm256_shuffle_epi8(rgba_pixel, shuffle), w0)
        }
    }

    #[inline(always)]
    fn acc_2_dot_avx(
        &self,
        start_x: usize,
        src0: &[u16],
        src1: &[u16],
        w0: __m256i,
        store: __m256i,
        shuffle: __m256i,
    ) -> __m256i {
        unsafe {
            const CN: usize = 4;

            let src_ptr0 = src0.get_unchecked((start_x * CN)..);
            let src_ptr1 = src1.get_unchecked((start_x * CN)..);

            let rgba_pixel0 = _mm_loadu_si128(src_ptr0.as_ptr().cast());
            let rgba_pixel1 = _mm_loadu_si128(src_ptr1.as_ptr().cast());

            let rgba_pixel =
                _mm256_inserti128_si256::<1>(_mm256_castsi128_si256(rgba_pixel0), rgba_pixel1);

            _mm256_dot16_avx_epi32::<D>(store, _mm256_shuffle_epi8(rgba_pixel, shuffle), w0)
        }
    }

    #[inline(always)]
    fn pass(
        &self,
        src: &[u16],
        src_stride: usize,
        dst: &mut [u16],
        dst_stride: usize,
        filter_weights: &FilterWeights<i16>,
        bit_depth: u32,
    ) {
        unsafe {
            const CN: usize = 4;

            let v_max_colors = _mm256_set1_epi16((1 << bit_depth) - 1);

            let shuffle_weights_table = _mm256_setr_epi8(
                0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3,
                0, 1, 2, 3,
            );

            let shuffle_1_table = _mm256_setr_epi8(
                0, 1, -1, -1, 2, 3, -1, -1, 4, 5, -1, -1, 6, 7, -1, -1, 0, 1, -1, -1, 2, 3, -1, -1,
                4, 5, -1, -1, 6, 7, -1, -1,
            );

            let permute_avx_weights = _mm256_setr_epi32(0, 2, 0, 0, 1, 3, 1, 1);

            let a_shuffle_2_table = _mm256_setr_epi8(
                0, 1, 8, 9, 2, 3, 10, 11, 4, 5, 12, 13, 6, 7, 14, 15, 0, 1, 8, 9, 2, 3, 10, 11, 4,
                5, 12, 13, 6, 7, 14, 15,
            );

            let a_shuffle_weights_table = _mm256_setr_epi8(
                0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3,
                0, 1, 2, 3,
            );

            let permute_avx_weights_hi = _mm256_setr_epi32(2, 2, 2, 2, 3, 3, 3, 3);

            let (row0_ref, rest) = dst.split_at_mut(dst_stride);
            let (row1_ref, rest) = rest.split_at_mut(dst_stride);
            let (row2_ref, row3_ref) = rest.split_at_mut(dst_stride);

            let init256 = _mm256_setr_epi32(
                ROUNDING_CONST,
                ROUNDING_CONST,
                ROUNDING_CONST,
                ROUNDING_CONST,
                0,
                0,
                0,
                0,
            );

            let iter_row0 = row0_ref.as_chunks_mut::<CN>().0.iter_mut();
            let iter_row1 = row1_ref.as_chunks_mut::<CN>().0.iter_mut();
            let iter_row2 = row2_ref.as_chunks_mut::<CN>().0.iter_mut();
            let iter_row3 = row3_ref.as_chunks_mut::<CN>().0.iter_mut();

            for (((((chunk0, chunk1), chunk2), chunk3), &bounds), weights) in iter_row0
                .zip(iter_row1)
                .zip(iter_row2)
                .zip(iter_row3)
                .zip(filter_weights.bounds.iter())
                .zip(
                    filter_weights
                        .weights
                        .chunks_exact(filter_weights.aligned_size),
                )
            {
                let mut jx = 0usize;

                let mut astore_0 = init256;
                let mut astore_1 = init256;
                let mut astore_2 = init256;
                let mut astore_3 = init256;

                let src0 = src;
                let src1 = src0.get_unchecked(src_stride..);
                let src2 = src1.get_unchecked(src_stride..);
                let src3 = src2.get_unchecked(src_stride..);

                let bounds_size = bounds.size;

                while jx + 8 <= bounds_size {
                    let w_ptr = weights.get_unchecked(jx..);
                    let wl = _mm256_castsi128_si256(_mm_loadu_si128(w_ptr.as_ptr().cast()));
                    let w0 = _mm256_shuffle_epi8(
                        _mm256_permutevar8x32_epi32(wl, permute_avx_weights),
                        a_shuffle_weights_table,
                    );
                    let w1 = _mm256_shuffle_epi8(
                        _mm256_permutevar8x32_epi32(wl, permute_avx_weights_hi),
                        a_shuffle_weights_table,
                    );
                    let bounds_start = bounds.start + jx;
                    astore_0 =
                        acc_8_dot::<D>(bounds_start, src0, w0, w1, astore_0, a_shuffle_2_table);
                    astore_1 =
                        acc_8_dot::<D>(bounds_start, src1, w0, w1, astore_1, a_shuffle_2_table);
                    astore_2 =
                        acc_8_dot::<D>(bounds_start, src2, w0, w1, astore_2, a_shuffle_2_table);
                    astore_3 =
                        acc_8_dot::<D>(bounds_start, src3, w0, w1, astore_3, a_shuffle_2_table);
                    jx += 8;
                }

                while jx + 4 <= bounds_size {
                    let bounds_start = bounds.start + jx;
                    let w_ptr = weights.get_unchecked(jx..);
                    let w0 = _mm256_shuffle_epi8(
                        _mm256_permutevar8x32_epi32(
                            _mm256_castsi128_si256(_mm_loadu_si64(w_ptr.as_ptr().cast())),
                            permute_avx_weights,
                        ),
                        a_shuffle_weights_table,
                    );
                    astore_0 = acc_4_dot::<D>(bounds_start, src0, w0, astore_0, a_shuffle_2_table);
                    astore_1 = acc_4_dot::<D>(bounds_start, src1, w0, astore_1, a_shuffle_2_table);
                    astore_2 = acc_4_dot::<D>(bounds_start, src2, w0, astore_2, a_shuffle_2_table);
                    astore_3 = acc_4_dot::<D>(bounds_start, src3, w0, astore_3, a_shuffle_2_table);
                    jx += 4;
                }

                let mut store_0 = _mm256_add_epi32(
                    _mm256_permute2x128_si256::<0x20>(astore_0, astore_1),
                    _mm256_permute2x128_si256::<0x31>(astore_0, astore_1),
                );

                let mut store_1 = _mm256_add_epi32(
                    _mm256_permute2x128_si256::<0x20>(astore_2, astore_3),
                    _mm256_permute2x128_si256::<0x31>(astore_2, astore_3),
                );

                while jx + 2 <= bounds_size {
                    let w_ptr = weights.get_unchecked(jx..);
                    let bounds_start = bounds.start + jx;
                    let ww0 = _mm_loadu_si32(w_ptr.as_ptr().cast());
                    let w0 = _mm256_shuffle_epi8(
                        _mm256_inserti128_si256::<1>(_mm256_castsi128_si256(ww0), ww0),
                        shuffle_weights_table,
                    );
                    store_0 = self.acc_2_dot_avx(
                        bounds_start,
                        src0,
                        src1,
                        w0,
                        store_0,
                        a_shuffle_2_table,
                    );
                    store_1 = self.acc_2_dot_avx(
                        bounds_start,
                        src2,
                        src3,
                        w0,
                        store_1,
                        a_shuffle_2_table,
                    );
                    jx += 2;
                }

                while jx < bounds_size {
                    let w_ptr = weights.get_unchecked(jx);
                    let bounds_start = bounds.start + jx;
                    let ww0 = _mm_set1_epi16(*w_ptr);
                    let w0 = _mm256_shuffle_epi8(
                        _mm256_inserti128_si256::<1>(_mm256_castsi128_si256(ww0), ww0),
                        shuffle_weights_table,
                    );
                    store_0 =
                        self.acc_1_dot_avx(bounds_start, src0, src1, w0, store_0, shuffle_1_table);
                    store_1 =
                        self.acc_1_dot_avx(bounds_start, src2, src3, w0, store_1, shuffle_1_table);
                    jx += 1;
                }

                store_0 = _mm256_srai_epi32::<PRECISION>(store_0);
                store_1 = _mm256_srai_epi32::<PRECISION>(store_1);

                store_0 = _mm256_packus_epi32(store_0, store_0);
                store_1 = _mm256_packus_epi32(store_1, store_1);

                let v_st0 = _mm256_min_epi16(store_0, v_max_colors);
                let v_st1 = _mm256_min_epi16(store_1, v_max_colors);

                _mm_storeu_si64(chunk0.as_mut_ptr().cast(), _mm256_castsi256_si128(v_st0));
                _mm_storeu_si64(
                    chunk1.as_mut_ptr().cast(),
                    _mm256_extracti128_si256::<1>(v_st0),
                );
                _mm_storeu_si64(chunk2.as_mut_ptr().cast(), _mm256_castsi256_si128(v_st1));
                _mm_storeu_si64(
                    chunk3.as_mut_ptr().cast(),
                    _mm256_extracti128_si256::<1>(v_st1),
                );
            }
        }
    }
}

pub(crate) fn convolve_horizontal_rgba_avx_u16lp_row(
    src: &[u16],
    dst: &mut [u16],
    filter_weights: &FilterWeights<i16>,
    bit_depth: u32,
) {
    unsafe {
        convolve_horizontal_rgba_avx_u16_row_avx(src, dst, filter_weights, bit_depth);
    }
}

#[cfg(feature = "avx512")]
pub(crate) fn convolve_horizontal_rgba_avx_u16lp_row_vnni(
    src: &[u16],
    dst: &mut [u16],
    filter_weights: &FilterWeights<i16>,
    bit_depth: u32,
) {
    unsafe { convolve_horizontal_rgba_avx_u16_row_vn(src, dst, filter_weights, bit_depth) }
}

#[cfg(feature = "avx512")]
#[target_feature(enable = "avxvnni", enable = "avx2")]
fn convolve_horizontal_rgba_avx_u16_row_vn(
    src: &[u16],
    dst: &mut [u16],
    filter_weights: &FilterWeights<i16>,
    bit_depth: u32,
) {
    let unit = OneRowExecutionUnit::<true>::default();
    unit.pass(src, dst, filter_weights, bit_depth);
}

#[target_feature(enable = "avx2")]
fn convolve_horizontal_rgba_avx_u16_row_avx(
    src: &[u16],
    dst: &mut [u16],
    filter_weights: &FilterWeights<i16>,
    bit_depth: u32,
) {
    let unit = OneRowExecutionUnit::<false>::default();
    unit.pass(src, dst, filter_weights, bit_depth);
}

#[derive(Copy, Clone, Default)]
struct OneRowExecutionUnit<const D: bool> {}

impl<const D: bool> OneRowExecutionUnit<D> {
    #[inline(always)]
    fn pass(
        &self,
        src: &[u16],
        dst: &mut [u16],
        filter_weights: &FilterWeights<i16>,
        bit_depth: u32,
    ) {
        unsafe {
            const CN: usize = 4;

            let v_max_colors = _mm_set1_epi16((1 << bit_depth) - 1);

            let shuffle_weights_table =
                _mm_setr_epi8(0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3);

            let shuffle_1_table =
                _mm_setr_epi8(0, 1, -1, -1, 2, 3, -1, -1, 4, 5, -1, -1, 6, 7, -1, -1);

            let shuffle_2_table =
                _mm_setr_epi8(0, 1, 8, 9, 2, 3, 10, 11, 4, 5, 12, 13, 6, 7, 14, 15);

            let permute_avx_weights = _mm256_setr_epi32(0, 0, 0, 0, 1, 1, 1, 1);

            let permute_avx_weights_hi = _mm256_setr_epi32(2, 2, 2, 2, 3, 3, 3, 3);

            let a_shuffle_2_table = _mm256_setr_epi8(
                0, 1, 8, 9, 2, 3, 10, 11, 4, 5, 12, 13, 6, 7, 14, 15, 0, 1, 8, 9, 2, 3, 10, 11, 4,
                5, 12, 13, 6, 7, 14, 15,
            );

            let a_shuffle_weights_table = _mm256_setr_epi8(
                0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3,
                0, 1, 2, 3,
            );

            for ((dst, bounds), weights) in dst
                .as_chunks_mut::<CN>()
                .0
                .iter_mut()
                .zip(filter_weights.bounds.iter())
                .zip(
                    filter_weights
                        .weights
                        .chunks_exact(filter_weights.aligned_size),
                )
            {
                let bounds_size = bounds.size;
                let mut jx = 0usize;
                let mut store = _mm256_setr_epi32(
                    ROUNDING_CONST,
                    ROUNDING_CONST,
                    ROUNDING_CONST,
                    ROUNDING_CONST,
                    0,
                    0,
                    0,
                    0,
                );

                while jx + 8 <= bounds_size {
                    let w_ptr = weights.get_unchecked(jx..);
                    let wl = _mm256_castsi128_si256(_mm_loadu_si128(w_ptr.as_ptr().cast()));
                    let w0 = _mm256_shuffle_epi8(
                        _mm256_permutevar8x32_epi32(wl, permute_avx_weights),
                        a_shuffle_weights_table,
                    );
                    let w1 = _mm256_shuffle_epi8(
                        _mm256_permutevar8x32_epi32(wl, permute_avx_weights_hi),
                        a_shuffle_weights_table,
                    );
                    let bounds_start = bounds.start + jx;
                    store = acc_8_dot::<D>(bounds_start, src, w0, w1, store, a_shuffle_2_table);
                    jx += 8;
                }

                while jx + 4 <= bounds_size {
                    let w_ptr = weights.get_unchecked(jx..);
                    let w0 = _mm256_shuffle_epi8(
                        _mm256_permutevar8x32_epi32(
                            _mm256_castsi128_si256(_mm_loadu_si64(w_ptr.as_ptr().cast())),
                            permute_avx_weights,
                        ),
                        a_shuffle_weights_table,
                    );
                    let bounds_start = bounds.start + jx;
                    store = acc_4_dot::<D>(bounds_start, src, w0, store, a_shuffle_2_table);
                    jx += 4;
                }

                let mut store = _mm_add_epi32(
                    _mm256_castsi256_si128(store),
                    _mm256_extracti128_si256::<1>(store),
                );

                while jx + 2 <= bounds_size {
                    let w_ptr = weights.get_unchecked(jx..);
                    let bounds_start = bounds.start + jx;
                    let w0 = _mm_shuffle_epi8(
                        _mm_loadu_si32(w_ptr.as_ptr().cast()),
                        shuffle_weights_table,
                    );
                    store = acc_2_dot::<D>(bounds_start, src, w0, store, shuffle_2_table);
                    jx += 2;
                }

                while jx < bounds_size {
                    let w_ptr = weights.get_unchecked(jx);
                    let w0 = _mm_shuffle_epi8(_mm_set1_epi16(*w_ptr), shuffle_weights_table);
                    let bounds_start = bounds.start + jx;
                    store = acc_1_dot::<D>(bounds_start, src, w0, store, shuffle_1_table);
                    jx += 1;
                }

                store = _mm_srai_epi32::<PRECISION>(store);

                let v_st = _mm_min_epi16(_mm_packus_epi32(store, store), v_max_colors);

                _mm_storeu_si64(dst.as_mut_ptr().cast(), v_st);
            }
        }
    }
}