pic-scale 0.7.6

/*
 * Copyright (c) Radzivon Bartoshyk. 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::filter_weights::FilterWeights;
use crate::support::{PRECISION, ROUNDING_CONST};
#[cfg(target_arch = "x86")]
use std::arch::x86::*;
#[cfg(target_arch = "x86_64")]
use std::arch::x86_64::*;

pub(crate) fn convolve_horizontal_rgba_vnni_row_4(
    src: &[u8],
    src_stride: usize,
    dst: &mut [u8],
    dst_stride: usize,
    filter_weights: &FilterWeights<i16>,
    _: u32,
) {
    unsafe {
        convolve_horizontal_rgba_vnni_row_4_impl(src, src_stride, dst, dst_stride, filter_weights);
    }
}

#[target_feature(enable = "avxvnni", enable = "avx2")]
fn convolve_horizontal_rgba_vnni_row_4_impl(
    src: &[u8],
    src_stride: usize,
    dst: &mut [u8],
    dst_stride: usize,
    filter_weights: &FilterWeights<i16>,
) {
    unsafe {
        const CN: usize = 4;

        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 = _mm256_setr_epi8(
            0, -1, -1, -1, 1, -1, -1, -1, 2, -1, -1, -1, 3, -1, -1, -1, 0, -1, -1, -1, 1, -1, -1,
            -1, 2, -1, -1, -1, 3, -1, -1, -1,
        );

        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 a_shuffle_weights_table_hi = _mm256_setr_epi8(
            4, 5, 6, 7, 4, 5, 6, 7, 4, 5, 6, 7, 4, 5, 6, 7, 4, 5, 6, 7, 4, 5, 6, 7, 4, 5, 6, 7, 4,
            5, 6, 7,
        );
        let a_shuffle_2_table_hi = _mm256_setr_epi8(
            8, -1, 12, -1, 9, -1, 13, -1, 10, -1, 14, -1, 11, -1, 15, -1, 8, -1, 12, -1, 9, -1, 13,
            -1, 10, -1, 14, -1, 11, -1, 15, -1,
        );

        let a_shuffle_2_table = _mm256_setr_epi8(
            0, -1, 4, -1, 1, -1, 5, -1, 2, -1, 6, -1, 3, -1, 7, -1, 0, -1, 4, -1, 1, -1, 5, -1, 2,
            -1, 6, -1, 3, -1, 7, -1,
        );

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

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

        let init256f = _mm256_set1_epi32(ROUNDING_CONST);

        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 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 store_0 = init256f;
            let mut store_1 = init256f;

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

            if bounds.size > 4 {
                let mut store_avx0 = vld;
                let mut store_avx1 = vld;
                let mut store_avx2 = vld;
                let mut store_avx3 = vld;

                while jx + 8 < bounds.size {
                    let w_ptr = weights.get_unchecked(jx..);
                    let bounds_start = bounds.start + jx;

                    let weights = _mm256_permute4x64_epi64::<0x50>(_mm256_castsi128_si256(
                        _mm_loadu_si128(w_ptr.as_ptr() as *const _),
                    ));
                    let weight01 = _mm256_shuffle_epi8(weights, a_shuffle_weights_table);
                    let weight23 = _mm256_shuffle_epi8(weights, a_shuffle_weights_table_hi);

                    let rgb_pixel0 = _mm256_loadu_si256(
                        src0.get_unchecked((bounds_start * CN)..).as_ptr() as *const _,
                    );
                    let rgb_pixel1 = _mm256_loadu_si256(
                        src1.get_unchecked((bounds_start * CN)..).as_ptr() as *const _,
                    );
                    let rgb_pixel2 = _mm256_loadu_si256(
                        src2.get_unchecked((bounds_start * CN)..).as_ptr() as *const _,
                    );
                    let rgb_pixel3 = _mm256_loadu_si256(
                        src3.get_unchecked((bounds_start * CN)..).as_ptr() as *const _,
                    );

                    let hi0 = _mm256_shuffle_epi8(rgb_pixel0, a_shuffle_2_table_hi);
                    let lo0 = _mm256_shuffle_epi8(rgb_pixel0, a_shuffle_2_table);
                    let hi1 = _mm256_shuffle_epi8(rgb_pixel1, a_shuffle_2_table_hi);
                    let lo1 = _mm256_shuffle_epi8(rgb_pixel1, a_shuffle_2_table);
                    let hi2 = _mm256_shuffle_epi8(rgb_pixel2, a_shuffle_2_table_hi);
                    let lo2 = _mm256_shuffle_epi8(rgb_pixel2, a_shuffle_2_table);
                    let hi3 = _mm256_shuffle_epi8(rgb_pixel3, a_shuffle_2_table_hi);
                    let lo3 = _mm256_shuffle_epi8(rgb_pixel3, a_shuffle_2_table);

                    store_avx0 = _mm256_dpwssd_avx_epi32(store_avx0, lo0, weight01);
                    store_avx0 = _mm256_dpwssd_avx_epi32(store_avx0, hi0, weight23);

                    store_avx1 = _mm256_dpwssd_avx_epi32(store_avx1, lo1, weight01);
                    store_avx1 = _mm256_dpwssd_avx_epi32(store_avx1, hi1, weight23);

                    store_avx2 = _mm256_dpwssd_avx_epi32(store_avx2, lo2, weight01);
                    store_avx2 = _mm256_dpwssd_avx_epi32(store_avx2, hi2, weight23);

                    store_avx3 = _mm256_dpwssd_avx_epi32(store_avx3, lo3, weight01);
                    store_avx3 = _mm256_dpwssd_avx_epi32(store_avx3, hi3, weight23);

                    jx += 8;
                }

                while jx + 4 < bounds.size {
                    let w_ptr = weights.get_unchecked(jx..);
                    let bounds_start = bounds.start + jx;

                    let weights = _mm256_permutevar8x32_epi32(
                        _mm256_castsi128_si256(_mm_loadu_si64(w_ptr.as_ptr() as *const u8)),
                        permute_avx_weights,
                    );
                    let weight01 = _mm256_shuffle_epi8(weights, a_shuffle_weights_table);

                    let rgb_pixel_0 = _mm_loadu_si128(
                        src0.get_unchecked((bounds_start * CN)..).as_ptr() as *const _,
                    );
                    let rgb_pixel_1 = _mm_loadu_si128(
                        src1.get_unchecked((bounds_start * CN)..).as_ptr() as *const _,
                    );
                    let rgb_pixel_2 = _mm_loadu_si128(
                        src2.get_unchecked((bounds_start * CN)..).as_ptr() as *const _,
                    );
                    let rgb_pixel_3 = _mm_loadu_si128(
                        src3.get_unchecked((bounds_start * CN)..).as_ptr() as *const _,
                    );

                    let rgb_pixel0 =
                        _mm256_permute4x64_epi64::<0x50>(_mm256_castsi128_si256(rgb_pixel_0));
                    let rgb_pixel1 =
                        _mm256_permute4x64_epi64::<0x50>(_mm256_castsi128_si256(rgb_pixel_1));
                    let rgb_pixel2 =
                        _mm256_permute4x64_epi64::<0x50>(_mm256_castsi128_si256(rgb_pixel_2));
                    let rgb_pixel3 =
                        _mm256_permute4x64_epi64::<0x50>(_mm256_castsi128_si256(rgb_pixel_3));

                    let lo0 = _mm256_shuffle_epi8(rgb_pixel0, a_shuffle_2_table);
                    let lo1 = _mm256_shuffle_epi8(rgb_pixel1, a_shuffle_2_table);
                    let lo2 = _mm256_shuffle_epi8(rgb_pixel2, a_shuffle_2_table);
                    let lo3 = _mm256_shuffle_epi8(rgb_pixel3, a_shuffle_2_table);

                    store_avx0 = _mm256_dpwssd_avx_epi32(store_avx0, lo0, weight01);
                    store_avx1 = _mm256_dpwssd_avx_epi32(store_avx1, lo1, weight01);
                    store_avx2 = _mm256_dpwssd_avx_epi32(store_avx2, lo2, weight01);
                    store_avx3 = _mm256_dpwssd_avx_epi32(store_avx3, lo3, weight01);

                    jx += 4;
                }

                store_0 = _mm256_add_epi16(
                    _mm256_permute2x128_si256::<0x20>(store_avx0, store_avx1),
                    _mm256_permute2x128_si256::<0x31>(store_avx0, store_avx1),
                );

                store_1 = _mm256_add_epi16(
                    _mm256_permute2x128_si256::<0x20>(store_avx2, store_avx3),
                    _mm256_permute2x128_si256::<0x31>(store_avx2, store_avx3),
                );
            }

            while jx + 2 < bounds.size {
                let w_ptr = weights.get_unchecked(jx..);
                let bounds_start = bounds.start + jx;

                let w01 = _mm_shuffle_epi8(
                    _mm_loadu_si32(w_ptr.as_ptr() as *const _),
                    shuffle_weights_table,
                );

                let weight01 = _mm256_inserti128_si256::<1>(_mm256_castsi128_si256(w01), w01);

                let rgb_pixel_0 =
                    _mm_loadu_si64(src0.get_unchecked((bounds_start * CN)..).as_ptr());
                let rgb_pixel_1 =
                    _mm_loadu_si64(src1.get_unchecked((bounds_start * CN)..).as_ptr());
                let rgb_pixel_2 =
                    _mm_loadu_si64(src2.get_unchecked((bounds_start * CN)..).as_ptr());
                let rgb_pixel_3 =
                    _mm_loadu_si64(src3.get_unchecked((bounds_start * CN)..).as_ptr());

                let px0 =
                    _mm256_inserti128_si256::<1>(_mm256_castsi128_si256(rgb_pixel_0), rgb_pixel_1);
                let px1 =
                    _mm256_inserti128_si256::<1>(_mm256_castsi128_si256(rgb_pixel_2), rgb_pixel_3);

                let lo_0 = _mm256_shuffle_epi8(px0, a_shuffle_2_table);
                let lo_1 = _mm256_shuffle_epi8(px1, a_shuffle_2_table);

                store_0 = _mm256_dpwssd_avx_epi32(store_0, lo_0, weight01);
                store_1 = _mm256_dpwssd_avx_epi32(store_1, lo_1, weight01);

                jx += 2;
            }

            while jx < bounds.size {
                let w_ptr = weights.get_unchecked(jx..);

                let w0 = _mm_shuffle_epi8(
                    _mm_loadu_si16(w_ptr.as_ptr() as *const _),
                    shuffle_weights_table,
                );

                let weight0 = _mm256_inserti128_si256::<1>(_mm256_castsi128_si256(w0), w0);

                let bounds_start = bounds.start + jx;

                let src_ptr0 = src0.get_unchecked((bounds_start * CN)..);
                let src_ptr1 = src1.get_unchecked((bounds_start * CN)..);
                let src_ptr2 = src2.get_unchecked((bounds_start * CN)..);
                let src_ptr3 = src3.get_unchecked((bounds_start * CN)..);

                let rgba_pixel0 = _mm_loadu_si32(src_ptr0.as_ptr() as *const _);
                let rgba_pixel1 = _mm_loadu_si32(src_ptr1.as_ptr() as *const _);
                let rgba_pixel2 = _mm_loadu_si32(src_ptr2.as_ptr() as *const _);
                let rgba_pixel3 = _mm_loadu_si32(src_ptr3.as_ptr() as *const _);

                let px0 =
                    _mm256_inserti128_si256::<1>(_mm256_castsi128_si256(rgba_pixel0), rgba_pixel1);
                let px1 =
                    _mm256_inserti128_si256::<1>(_mm256_castsi128_si256(rgba_pixel2), rgba_pixel3);

                let lo0 = _mm256_shuffle_epi8(px0, shuffle_1_table);
                let lo1 = _mm256_shuffle_epi8(px1, shuffle_1_table);

                store_0 = _mm256_dpwssd_avx_epi32(store_0, lo0, weight0);
                store_1 = _mm256_dpwssd_avx_epi32(store_1, lo1, weight0);

                jx += 1;
            }

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

            let store_16_8_0 = _mm256_packus_epi32(store_0, store_0);
            let store_16_8_1 = _mm256_packus_epi32(store_1, store_1);

            let packed16_0 = _mm256_packus_epi16(store_16_8_0, store_16_8_0);
            let packed16_1 = _mm256_packus_epi16(store_16_8_1, store_16_8_1);

            _mm_storeu_si32(
                chunk0.as_mut_ptr() as *mut _,
                _mm256_castsi256_si128(packed16_0),
            );
            _mm_storeu_si32(
                chunk1.as_mut_ptr() as *mut _,
                _mm256_extracti128_si256::<1>(packed16_0),
            );
            _mm_storeu_si32(
                chunk2.as_mut_ptr() as *mut _,
                _mm256_castsi256_si128(packed16_1),
            );
            _mm_storeu_si32(
                chunk3.as_mut_ptr() as *mut _,
                _mm256_extracti128_si256::<1>(packed16_1),
            );
        }
    }
}

pub(crate) fn convolve_horizontal_rgba_vnni_row_1(
    src: &[u8],
    dst: &mut [u8],
    filter_weights: &FilterWeights<i16>,
    _: u32,
) {
    unsafe {
        convolve_horizontal_rgba_vnni_rows_one_impl(src, dst, filter_weights);
    }
}

#[target_feature(enable = "avxvnni", enable = "avx2")]
fn convolve_horizontal_rgba_vnni_rows_one_impl(
    src: &[u8],
    dst: &mut [u8],
    filter_weights: &FilterWeights<i16>,
) {
    unsafe {
        const CN: usize = 4;

        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_2_table = _mm_setr_epi8(0, -1, 4, -1, 1, -1, 5, -1, 2, -1, 6, -1, 3, -1, 7, -1);

        let shuffle_1_table =
            _mm_setr_epi8(0, -1, -1, -1, 1, -1, -1, -1, 2, -1, -1, -1, 3, -1, -1, -1);

        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 a_shuffle_weights_table_hi = _mm256_setr_epi8(
            4, 5, 6, 7, 4, 5, 6, 7, 4, 5, 6, 7, 4, 5, 6, 7, 4, 5, 6, 7, 4, 5, 6, 7, 4, 5, 6, 7, 4,
            5, 6, 7,
        );
        let a_shuffle_2_table_hi = _mm256_setr_epi8(
            8, -1, 12, -1, 9, -1, 13, -1, 10, -1, 14, -1, 11, -1, 15, -1, 8, -1, 12, -1, 9, -1, 13,
            -1, 10, -1, 14, -1, 11, -1, 15, -1,
        );

        let a_shuffle_2_table = _mm256_setr_epi8(
            0, -1, 4, -1, 1, -1, 5, -1, 2, -1, 6, -1, 3, -1, 7, -1, 0, -1, 4, -1, 1, -1, 5, -1, 2,
            -1, 6, -1, 3, -1, 7, -1,
        );

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

        let vld = _mm_set1_epi32(PRECISION);

        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 mut jx = 0usize;
            let mut store = vld;

            if bounds.size > 4 {
                let mut store_avx = _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 bounds_start = bounds.start + jx;

                    let weights = _mm256_permute4x64_epi64::<0x50>(_mm256_castsi128_si256(
                        _mm_loadu_si128(w_ptr.as_ptr() as *const _),
                    ));
                    let weight01 = _mm256_shuffle_epi8(weights, a_shuffle_weights_table);
                    let weight23 = _mm256_shuffle_epi8(weights, a_shuffle_weights_table_hi);

                    let src_ptr = src.get_unchecked((bounds_start * CN)..);

                    let rgb_pixel = _mm256_loadu_si256(src_ptr.as_ptr() as *const _);

                    let hi = _mm256_shuffle_epi8(rgb_pixel, a_shuffle_2_table_hi);
                    let lo = _mm256_shuffle_epi8(rgb_pixel, a_shuffle_2_table);

                    store_avx = _mm256_dpwssd_avx_epi32(store_avx, lo, weight01);
                    store_avx = _mm256_dpwssd_avx_epi32(store_avx, hi, weight23);

                    jx += 8;
                }

                while jx + 4 < bounds.size {
                    let w_ptr = weights.get_unchecked(jx..);
                    let bounds_start = bounds.start + jx;

                    let weights = _mm256_permutevar8x32_epi32(
                        _mm256_castsi128_si256(_mm_loadu_si64(w_ptr.as_ptr() as *const u8)),
                        permute_avx_weights,
                    );
                    let weight01 = _mm256_shuffle_epi8(weights, a_shuffle_weights_table);

                    let src_ptr = src.get_unchecked((bounds_start * CN)..);

                    let rgb_pixel = _mm256_permute4x64_epi64::<0x50>(_mm256_castsi128_si256(
                        _mm_loadu_si128(src_ptr.as_ptr() as *const __m128i),
                    ));

                    let lo = _mm256_shuffle_epi8(rgb_pixel, a_shuffle_2_table);

                    store_avx = _mm256_dpwssd_avx_epi32(store_avx, lo, weight01);
                    jx += 4;
                }

                store = _mm_add_epi32(
                    _mm256_castsi256_si128(store_avx),
                    _mm256_extracti128_si256::<1>(store_avx),
                );
            }

            while jx + 2 < bounds.size {
                let w_ptr = weights.get_unchecked(jx..);
                let bounds_start = bounds.start + jx;

                let weight01 = _mm_shuffle_epi8(
                    _mm_loadu_si32(w_ptr.as_ptr() as *const _),
                    shuffle_weights_table,
                );

                let src_ptr = src.get_unchecked((bounds_start * CN)..);

                let rgb_pixel = _mm_loadu_si64(src_ptr.as_ptr());
                let lo = _mm_shuffle_epi8(rgb_pixel, shuffle_2_table);
                store = _mm_dpwssd_avx_epi32(store, lo, weight01);

                jx += 2;
            }

            while jx < bounds.size {
                let w_ptr = weights.get_unchecked(jx..);
                let weight0 = _mm_shuffle_epi8(
                    _mm_loadu_si16(w_ptr.as_ptr() as *const _),
                    shuffle_weights_table,
                );

                let bounds_start = bounds.start + jx;

                const CN: usize = 4;
                let src_ptr = src.get_unchecked((bounds_start * CN)..);

                let src_ptr_32 = src_ptr.as_ptr() as *const i32;
                let rgba_pixel = _mm_loadu_si32(src_ptr_32 as *const _);

                let lo = _mm_shuffle_epi8(rgba_pixel, shuffle_1_table);

                store = _mm_dpwssd_avx_epi32(store, lo, weight0);

                jx += 1;
            }

            let store_0 = _mm_srai_epi32::<PRECISION>(store);

            let store_0 = _mm_packus_epi32(store_0, store_0);

            _mm_storeu_si32(
                dst.as_mut_ptr() as *mut _,
                _mm_packus_epi16(store_0, store_0),
            );
        }
    }
}