yuv 0.8.13

/*
 * Copyright (c) Radzivon Bartoshyk, 10/2024. 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::internals::ProcessedOffset;
use crate::neon::utils::*;
use crate::yuv_support::{
    CbCrForwardTransform, YuvChromaRange, YuvChromaSubsampling, YuvSourceChannels,
};
use crate::{YuvBytesPacking, YuvEndianness};
use std::arch::aarch64::*;

pub(crate) unsafe fn neon_rgba_to_yuv_p16<
    const ORIGIN_CHANNELS: u8,
    const SAMPLING: u8,
    const ENDIANNESS: u8,
    const BYTES_POSITION: u8,
    const PRECISION: i32,
    const BIT_DEPTH: usize,
>(
    transform: &CbCrForwardTransform<i32>,
    range: &YuvChromaRange,
    y_plane: &mut [u16],
    u_plane: &mut [u16],
    v_plane: &mut [u16],
    rgba: &[u16],
    start_cx: usize,
    start_ux: usize,
    width: usize,
) -> ProcessedOffset {
    let chroma_subsampling: YuvChromaSubsampling = SAMPLING.into();
    let source_channels: YuvSourceChannels = ORIGIN_CHANNELS.into();
    let _endianness: YuvEndianness = ENDIANNESS.into();
    let bytes_position: YuvBytesPacking = BYTES_POSITION.into();
    let channels = source_channels.get_channels_count();

    let rounding_const_bias: u32 = (1 << (PRECISION - 1)) - 1;
    let bias_y = range.bias_y * (1 << PRECISION) + rounding_const_bias;
    let bias_uv = range.bias_uv * (1 << PRECISION) + rounding_const_bias;

    let y_ptr = y_plane.as_mut_ptr();
    let u_ptr = u_plane.as_mut_ptr();
    let v_ptr = v_plane.as_mut_ptr();

    let y_bias = vdupq_n_u32(bias_y);
    let uv_bias = vdupq_n_u32(bias_uv);

    let weights_arr: [u16; 8] = [
        transform.yr as u16,
        transform.yg as u16,
        transform.yb as u16,
        transform.cb_r.unsigned_abs() as u16,
        transform.cb_g.unsigned_abs() as u16,
        transform.cb_b.unsigned_abs() as u16,
        transform.cr_r.unsigned_abs() as u16,
        transform.cr_g.unsigned_abs() as u16,
    ];
    let v_weights = vld1q_u16(weights_arr.as_ptr());
    let v_cr_b = vdupq_n_u16(transform.cr_b.unsigned_abs() as u16);

    let mut cx = start_cx;
    let mut ux = start_ux;

    let i_bias_y = vdupq_n_u16(range.bias_y as u16);
    let i_cap_uv = vdupq_n_u16(range.bias_y as u16 + range.range_uv as u16);

    while cx + 8 <= width {
        let src_ptr = rgba.get_unchecked(cx * channels..);
        let (r_values, g_values, b_values) =
            neon_vld_rgb16_for_yuv::<ORIGIN_CHANNELS>(src_ptr.as_ptr());

        let mut y_vl = vdotl_laneq_u16_x3::<PRECISION, 0, 1, 2>(
            y_bias, r_values, g_values, b_values, v_weights,
        );

        if bytes_position == YuvBytesPacking::MostSignificantBytes {
            y_vl = vtomsbq_u16::<BIT_DEPTH>(y_vl);
        }

        #[cfg(all(feature = "big_endian", target_endian = "little"))]
        if _endianness == YuvEndianness::BigEndian {
            y_vl = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(y_vl)));
        }

        vst1q_u16(y_ptr.add(cx), y_vl);

        if chroma_subsampling == YuvChromaSubsampling::Yuv444 {
            let mut cb_h = vmlal_high_laneq_u16::<5>(uv_bias, b_values, v_weights);
            let mut cb_l = vmlal_laneq_u16::<5>(uv_bias, vget_low_u16(b_values), v_weights);
            let mut cr_h = vmlal_high_laneq_u16::<6>(uv_bias, r_values, v_weights);
            let mut cr_l = vmlal_laneq_u16::<6>(uv_bias, vget_low_u16(r_values), v_weights);

            cb_h = vmlsl_high_laneq_u16::<4>(cb_h, g_values, v_weights);
            cb_l = vmlsl_laneq_u16::<4>(cb_l, vget_low_u16(g_values), v_weights);
            cr_h = vmlsl_high_laneq_u16::<7>(cr_h, g_values, v_weights);
            cr_l = vmlsl_laneq_u16::<7>(cr_l, vget_low_u16(g_values), v_weights);

            cb_h = vmlsl_high_laneq_u16::<3>(cb_h, r_values, v_weights);
            cb_l = vmlsl_laneq_u16::<3>(cb_l, vget_low_u16(r_values), v_weights);
            cr_h = vmlsl_high_laneq_u16::<0>(cr_h, b_values, v_cr_b);
            cr_l = vmlsl_laneq_u16::<0>(cr_l, vget_low_u16(b_values), v_cr_b);

            let cb_cc = vcombine_u16(
                vqshrn_n_u32::<PRECISION>(cb_l),
                vqshrn_n_u32::<PRECISION>(cb_h),
            );
            let cr_cc = vcombine_u16(
                vqshrn_n_u32::<PRECISION>(cr_l),
                vqshrn_n_u32::<PRECISION>(cr_h),
            );

            let cb_max = vmaxq_u16(cb_cc, i_bias_y);
            let cr_max = vmaxq_u16(cr_cc, i_bias_y);

            let mut cb_vl = vminq_u16(cb_max, i_cap_uv);
            let mut cr_vl = vminq_u16(cr_max, i_cap_uv);

            if bytes_position == YuvBytesPacking::MostSignificantBytes {
                cb_vl = vtomsbq_u16::<BIT_DEPTH>(cb_vl);
                cr_vl = vtomsbq_u16::<BIT_DEPTH>(cr_vl);
            }

            #[cfg(all(feature = "big_endian", target_endian = "little"))]
            if _endianness == YuvEndianness::BigEndian {
                cb_vl = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(cb_vl)));
                cr_vl = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(cr_vl)));
            }

            vst1q_u16(u_ptr.add(ux), cb_vl);
            vst1q_u16(v_ptr.add(ux), cr_vl);

            ux += 8;
        } else {
            let r1l = vpaddlq_u16(r_values);
            let g1l = vpaddlq_u16(g_values);
            let b1l = vpaddlq_u16(b_values);

            let r1 = vrshrn_n_u32::<1>(r1l);
            let g1 = vrshrn_n_u32::<1>(g1l);
            let b1 = vrshrn_n_u32::<1>(b1l);

            let mut cb_h = vmlal_laneq_u16::<5>(uv_bias, b1, v_weights);
            let mut cr_h = vmlal_laneq_u16::<6>(uv_bias, r1, v_weights);
            cb_h = vmlsl_laneq_u16::<4>(cb_h, g1, v_weights);
            cr_h = vmlsl_laneq_u16::<7>(cr_h, g1, v_weights);
            cb_h = vmlsl_laneq_u16::<3>(cb_h, r1, v_weights);
            cr_h = vmlsl_laneq_u16::<0>(cr_h, b1, v_cr_b);

            let qcb = vqshrn_n_u32::<PRECISION>(cb_h);
            let qcr = vqshrn_n_u32::<PRECISION>(cr_h);

            let cb_max = vmax_u16(qcb, vget_low_u16(i_bias_y));
            let cr_max = vmax_u16(qcr, vget_low_u16(i_bias_y));

            let mut cb_s = vmin_u16(cb_max, vget_low_u16(i_cap_uv));
            let mut cr_s = vmin_u16(cr_max, vget_low_u16(i_cap_uv));

            if bytes_position == YuvBytesPacking::MostSignificantBytes {
                cb_s = vtomsb_u16::<BIT_DEPTH>(cb_s);
                cr_s = vtomsb_u16::<BIT_DEPTH>(cr_s);
            }

            #[cfg(all(feature = "big_endian", target_endian = "little"))]
            if _endianness == YuvEndianness::BigEndian {
                cb_s = vreinterpret_u16_u8(vrev16_u8(vreinterpret_u8_u16(cb_s)));
                cr_s = vreinterpret_u16_u8(vrev16_u8(vreinterpret_u8_u16(cr_s)));
            }

            vst1_u16(u_ptr.add(ux), cb_s);
            vst1_u16(v_ptr.add(ux), cr_s);

            ux += 4;
        }

        cx += 8;
    }

    if cx < width {
        let diff = width - cx;
        assert!(diff <= 16);
        let mut src_buffer: [u16; 16 * 4] = [0; 16 * 4];
        let mut y_buffer: [u16; 16] = [0; 16];
        let mut u_buffer: [u16; 16] = [0; 16];
        let mut v_buffer: [u16; 16] = [0; 16];

        // Replicate last item to one more position for subsampling
        if chroma_subsampling != YuvChromaSubsampling::Yuv444 && diff % 2 != 0 {
            let lst = (width - 1) * channels;
            let last_items = rgba.get_unchecked(lst..(lst + channels));
            let dvb = diff * channels;
            let dst = src_buffer.get_unchecked_mut(dvb..(dvb + channels));
            for (dst, src) in dst.iter_mut().zip(last_items) {
                *dst = *src;
            }
        }

        std::ptr::copy_nonoverlapping(
            rgba.get_unchecked(cx * channels..).as_ptr(),
            src_buffer.as_mut_ptr().cast(),
            diff * channels,
        );

        let (r_values, g_values, b_values) =
            neon_vld_rgb16_for_yuv::<ORIGIN_CHANNELS>(src_buffer.as_ptr().cast());

        let mut y_vl = vdotl_laneq_u16_x3::<PRECISION, 0, 1, 2>(
            y_bias, r_values, g_values, b_values, v_weights,
        );

        if bytes_position == YuvBytesPacking::MostSignificantBytes {
            y_vl = vtomsbq_u16::<BIT_DEPTH>(y_vl);
        }

        #[cfg(all(feature = "big_endian", target_endian = "little"))]
        if _endianness == YuvEndianness::BigEndian {
            y_vl = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(y_vl)));
        }

        vst1q_u16(y_buffer.as_mut_ptr().cast(), y_vl);

        if chroma_subsampling == YuvChromaSubsampling::Yuv444 {
            let mut cb_h = vmlal_high_laneq_u16::<5>(uv_bias, b_values, v_weights);
            let mut cb_l = vmlal_laneq_u16::<5>(uv_bias, vget_low_u16(b_values), v_weights);
            let mut cr_h = vmlal_high_laneq_u16::<6>(uv_bias, r_values, v_weights);
            let mut cr_l = vmlal_laneq_u16::<6>(uv_bias, vget_low_u16(r_values), v_weights);

            cb_h = vmlsl_high_laneq_u16::<4>(cb_h, g_values, v_weights);
            cb_l = vmlsl_laneq_u16::<4>(cb_l, vget_low_u16(g_values), v_weights);
            cr_h = vmlsl_high_laneq_u16::<7>(cr_h, g_values, v_weights);
            cr_l = vmlsl_laneq_u16::<7>(cr_l, vget_low_u16(g_values), v_weights);

            cb_h = vmlsl_high_laneq_u16::<3>(cb_h, r_values, v_weights);
            cb_l = vmlsl_laneq_u16::<3>(cb_l, vget_low_u16(r_values), v_weights);
            cr_h = vmlsl_high_laneq_u16::<0>(cr_h, b_values, v_cr_b);
            cr_l = vmlsl_laneq_u16::<0>(cr_l, vget_low_u16(b_values), v_cr_b);

            let cb_cc = vcombine_u16(
                vqshrn_n_u32::<PRECISION>(cb_l),
                vqshrn_n_u32::<PRECISION>(cb_h),
            );
            let cr_cc = vcombine_u16(
                vqshrn_n_u32::<PRECISION>(cr_l),
                vqshrn_n_u32::<PRECISION>(cr_h),
            );

            let cb_max = vmaxq_u16(cb_cc, i_bias_y);
            let cr_max = vmaxq_u16(cr_cc, i_bias_y);

            let mut cb_vl = vminq_u16(cb_max, i_cap_uv);
            let mut cr_vl = vminq_u16(cr_max, i_cap_uv);

            if bytes_position == YuvBytesPacking::MostSignificantBytes {
                cb_vl = vtomsbq_u16::<BIT_DEPTH>(cb_vl);
                cr_vl = vtomsbq_u16::<BIT_DEPTH>(cr_vl);
            }

            #[cfg(all(feature = "big_endian", target_endian = "little"))]
            if _endianness == YuvEndianness::BigEndian {
                cb_vl = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(cb_vl)));
                cr_vl = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(cr_vl)));
            }

            vst1q_u16(u_buffer.as_mut_ptr().cast(), cb_vl);
            vst1q_u16(v_buffer.as_mut_ptr().cast(), cr_vl);
        } else {
            let r1l = vpaddlq_u16(r_values);
            let g1l = vpaddlq_u16(g_values);
            let b1l = vpaddlq_u16(b_values);

            let r1 = vrshrn_n_u32::<1>(r1l);
            let g1 = vrshrn_n_u32::<1>(g1l);
            let b1 = vrshrn_n_u32::<1>(b1l);

            let mut cb_h = vmlal_laneq_u16::<5>(uv_bias, b1, v_weights);
            let mut cr_h = vmlal_laneq_u16::<6>(uv_bias, r1, v_weights);
            cb_h = vmlsl_laneq_u16::<4>(cb_h, g1, v_weights);
            cr_h = vmlsl_laneq_u16::<7>(cr_h, g1, v_weights);
            cb_h = vmlsl_laneq_u16::<3>(cb_h, r1, v_weights);
            cr_h = vmlsl_laneq_u16::<0>(cr_h, b1, v_cr_b);

            let qcb = vqshrn_n_u32::<PRECISION>(cb_h);
            let qcr = vqshrn_n_u32::<PRECISION>(cr_h);

            let cb_max = vmax_u16(qcb, vget_low_u16(i_bias_y));
            let cr_max = vmax_u16(qcr, vget_low_u16(i_bias_y));

            let mut cb_s = vmin_u16(cb_max, vget_low_u16(i_cap_uv));
            let mut cr_s = vmin_u16(cr_max, vget_low_u16(i_cap_uv));

            if bytes_position == YuvBytesPacking::MostSignificantBytes {
                cb_s = vtomsb_u16::<BIT_DEPTH>(cb_s);
                cr_s = vtomsb_u16::<BIT_DEPTH>(cr_s);
            }

            #[cfg(all(feature = "big_endian", target_endian = "little"))]
            if _endianness == YuvEndianness::BigEndian {
                cb_s = vreinterpret_u16_u8(vrev16_u8(vreinterpret_u8_u16(cb_s)));
                cr_s = vreinterpret_u16_u8(vrev16_u8(vreinterpret_u8_u16(cr_s)));
            }

            vst1_u16(u_buffer.as_mut_ptr().cast(), cb_s);
            vst1_u16(v_buffer.as_mut_ptr().cast(), cr_s);
        }

        std::ptr::copy_nonoverlapping(
            y_buffer.as_ptr().cast(),
            y_plane.get_unchecked_mut(cx..).as_mut_ptr(),
            diff,
        );

        cx += diff;

        if chroma_subsampling == YuvChromaSubsampling::Yuv444 {
            std::ptr::copy_nonoverlapping(
                u_buffer.as_ptr().cast(),
                u_plane.get_unchecked_mut(ux..).as_mut_ptr(),
                diff,
            );
            std::ptr::copy_nonoverlapping(
                v_buffer.as_ptr().cast(),
                v_plane.get_unchecked_mut(ux..).as_mut_ptr(),
                diff,
            );

            ux += diff;
        } else if (chroma_subsampling == YuvChromaSubsampling::Yuv420)
            || (chroma_subsampling == YuvChromaSubsampling::Yuv422)
        {
            let hv = diff.div_ceil(2);
            std::ptr::copy_nonoverlapping(
                u_buffer.as_ptr().cast(),
                u_plane.get_unchecked_mut(ux..).as_mut_ptr(),
                hv,
            );
            std::ptr::copy_nonoverlapping(
                v_buffer.as_ptr().cast(),
                v_plane.get_unchecked_mut(ux..).as_mut_ptr(),
                hv,
            );

            ux += hv;
        }
    }

    ProcessedOffset { ux, cx }
}

#[cfg(feature = "rdm")]
#[target_feature(enable = "rdm")]
pub(crate) unsafe fn neon_rgba_to_yuv_p16_rdm<
    const ORIGIN_CHANNELS: u8,
    const SAMPLING: u8,
    const ENDIANNESS: u8,
    const BYTES_POSITION: u8,
    const PRECISION: i32,
    const BIT_DEPTH: usize,
>(
    transform: &CbCrForwardTransform<i32>,
    range: &YuvChromaRange,
    y_plane: &mut [u16],
    u_plane: &mut [u16],
    v_plane: &mut [u16],
    rgba: &[u16],
    start_cx: usize,
    start_ux: usize,
    width: usize,
) -> ProcessedOffset {
    let chroma_subsampling: YuvChromaSubsampling = SAMPLING.into();
    let source_channels: YuvSourceChannels = ORIGIN_CHANNELS.into();
    let _endianness: YuvEndianness = ENDIANNESS.into();
    let bytes_position: YuvBytesPacking = BYTES_POSITION.into();
    let channels = source_channels.get_channels_count();

    let bias_uv = range.bias_uv as i16;

    const SCALE: i32 = 2;

    let weights_arr: [i16; 8] = [
        transform.yr as i16,
        transform.yg as i16,
        transform.yb as i16,
        transform.cb_r as i16,
        transform.cb_g as i16,
        transform.cb_b as i16,
        transform.cr_r as i16,
        transform.cr_g as i16,
    ];
    let v_weights = vld1q_s16(weights_arr.as_ptr());
    let v_cr_b = vdupq_n_s16(transform.cr_b as i16);

    let i_bias_y = vdupq_n_s16(range.bias_y as i16 * (1 << 2) + (1 << (SCALE - 1)) - 1);
    let uv_bias = vdupq_n_s16(bias_uv * (1 << 2) + (1 << (SCALE - 1)) - 1);
    let i_cap_y = vdupq_n_s16(range.range_y as i16 + range.bias_y as i16);
    let i_cap_uv = vdupq_n_s16(range.bias_y as i16 + range.range_uv as i16);

    let mut cx = start_cx;
    let mut ux = start_ux;

    while cx + 8 <= width {
        let src_ptr = rgba.get_unchecked(cx * channels..);
        let (mut r_values, mut g_values, mut b_values) =
            neon_vld_rgb16_for_yuv::<ORIGIN_CHANNELS>(src_ptr.as_ptr());

        r_values = vexpandu_high_bp_by_2::<BIT_DEPTH>(r_values);
        g_values = vexpandu_high_bp_by_2::<BIT_DEPTH>(g_values);
        b_values = vexpandu_high_bp_by_2::<BIT_DEPTH>(b_values);

        let mut y_values =
            vqrdmlahq_laneq_s16::<0>(i_bias_y, vreinterpretq_s16_u16(r_values), v_weights);
        y_values = vqrdmlahq_laneq_s16::<1>(y_values, vreinterpretq_s16_u16(g_values), v_weights);
        y_values = vqrdmlahq_laneq_s16::<2>(y_values, vreinterpretq_s16_u16(b_values), v_weights);

        let mut y_vl = vreinterpretq_u16_s16(vminq_s16(vshrq_n_s16::<2>(y_values), i_cap_y));

        if bytes_position == YuvBytesPacking::MostSignificantBytes {
            y_vl = vtomsbq_u16::<BIT_DEPTH>(y_vl);
        }

        #[cfg(all(feature = "big_endian", target_endian = "little"))]
        if _endianness == YuvEndianness::BigEndian {
            y_vl = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(y_vl)));
        }

        vst1q_u16(y_plane.get_unchecked_mut(cx..).as_mut_ptr(), y_vl);

        if chroma_subsampling == YuvChromaSubsampling::Yuv444 {
            let mut cb_h =
                vqrdmlahq_laneq_s16::<3>(uv_bias, vreinterpretq_s16_u16(r_values), v_weights);
            let mut cr_h =
                vqrdmlahq_laneq_s16::<6>(uv_bias, vreinterpretq_s16_u16(r_values), v_weights);
            cb_h = vqrdmlahq_laneq_s16::<4>(cb_h, vreinterpretq_s16_u16(g_values), v_weights);
            cr_h = vqrdmlahq_laneq_s16::<7>(cr_h, vreinterpretq_s16_u16(g_values), v_weights);
            cb_h = vqrdmlahq_laneq_s16::<5>(cb_h, vreinterpretq_s16_u16(b_values), v_weights);
            cr_h = vqrdmlahq_laneq_s16::<0>(cr_h, vreinterpretq_s16_u16(b_values), v_cr_b);

            let cb_max = vmaxq_s16(vshrq_n_s16::<2>(cb_h), i_bias_y);
            let cr_max = vmaxq_s16(vshrq_n_s16::<2>(cr_h), i_bias_y);

            let mut cb_vl = vreinterpretq_u16_s16(vminq_s16(cb_max, i_cap_uv));
            let mut cr_vl = vreinterpretq_u16_s16(vminq_s16(cr_max, i_cap_uv));

            if bytes_position == YuvBytesPacking::MostSignificantBytes {
                cb_vl = vtomsbq_u16::<BIT_DEPTH>(cb_vl);
                cr_vl = vtomsbq_u16::<BIT_DEPTH>(cr_vl);
            }

            #[cfg(all(feature = "big_endian", target_endian = "little"))]
            if _endianness == YuvEndianness::BigEndian {
                cb_vl = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(cb_vl)));
                cr_vl = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(cr_vl)));
            }

            vst1q_u16(u_plane.get_unchecked_mut(ux..).as_mut_ptr(), cb_vl);
            vst1q_u16(v_plane.get_unchecked_mut(ux..).as_mut_ptr(), cr_vl);

            ux += 8;
        } else if (chroma_subsampling == YuvChromaSubsampling::Yuv420)
            || (chroma_subsampling == YuvChromaSubsampling::Yuv422)
        {
            let r1 = vreinterpret_s16_u16(vrshrn_n_u32::<1>(vpaddlq_u16(r_values)));
            let g1 = vreinterpret_s16_u16(vrshrn_n_u32::<1>(vpaddlq_u16(g_values)));
            let b1 = vreinterpret_s16_u16(vrshrn_n_u32::<1>(vpaddlq_u16(b_values)));

            let mut cbk = vqrdmlah_laneq_s16::<3>(vget_low_s16(uv_bias), r1, v_weights);
            let mut crk = vqrdmlah_laneq_s16::<6>(vget_low_s16(uv_bias), r1, v_weights);
            cbk = vqrdmlah_laneq_s16::<4>(cbk, g1, v_weights);
            crk = vqrdmlah_laneq_s16::<7>(crk, g1, v_weights);
            cbk = vqrdmlah_laneq_s16::<5>(cbk, b1, v_weights);
            crk = vqrdmlah_laneq_s16::<0>(crk, b1, v_cr_b);

            let cb_max = vmax_s16(vshr_n_s16::<2>(cbk), vget_low_s16(i_bias_y));
            let cr_max = vmax_s16(vshr_n_s16::<2>(crk), vget_low_s16(i_bias_y));

            let mut cb = vreinterpret_u16_s16(vmin_s16(cb_max, vget_low_s16(i_cap_uv)));
            let mut cr = vreinterpret_u16_s16(vmin_s16(cr_max, vget_low_s16(i_cap_uv)));

            if bytes_position == YuvBytesPacking::MostSignificantBytes {
                cb = vtomsb_u16::<BIT_DEPTH>(cb);
                cr = vtomsb_u16::<BIT_DEPTH>(cr);
            }

            #[cfg(all(feature = "big_endian", target_endian = "little"))]
            if _endianness == YuvEndianness::BigEndian {
                cb = vreinterpret_u16_u8(vrev16_u8(vreinterpret_u8_u16(cb)));
                cr = vreinterpret_u16_u8(vrev16_u8(vreinterpret_u8_u16(cr)));
            }

            vst1_u16(u_plane.get_unchecked_mut(ux..).as_mut_ptr(), cb);
            vst1_u16(v_plane.get_unchecked_mut(ux..).as_mut_ptr(), cr);

            ux += 4;
        }

        cx += 8;
    }

    if cx < width {
        let diff = width - cx;
        assert!(diff <= 8);
        let mut src_buffer: [u16; 8 * 4] = [0; 8 * 4];
        let mut y_buffer: [u16; 8] = [0; 8];
        let mut u_buffer: [u16; 8] = [0; 8];
        let mut v_buffer: [u16; 8] = [0; 8];

        // Replicate last item to one more position for subsampling
        if chroma_subsampling != YuvChromaSubsampling::Yuv444 && diff % 2 != 0 {
            let lst = (width - 1) * channels;
            let last_items = rgba.get_unchecked(lst..(lst + channels));
            let dvb = diff * channels;
            let dst = src_buffer.get_unchecked_mut(dvb..(dvb + channels));
            for (dst, src) in dst.iter_mut().zip(last_items) {
                *dst = *src;
            }
        }

        std::ptr::copy_nonoverlapping(
            rgba.get_unchecked(cx * channels..).as_ptr(),
            src_buffer.as_mut_ptr().cast(),
            diff * channels,
        );

        let (mut r_values, mut g_values, mut b_values) =
            neon_vld_rgb16_for_yuv::<ORIGIN_CHANNELS>(src_buffer.as_ptr().cast());

        r_values = vexpandu_high_bp_by_2::<BIT_DEPTH>(r_values);
        g_values = vexpandu_high_bp_by_2::<BIT_DEPTH>(g_values);
        b_values = vexpandu_high_bp_by_2::<BIT_DEPTH>(b_values);

        let mut y_values =
            vqrdmlahq_laneq_s16::<0>(i_bias_y, vreinterpretq_s16_u16(r_values), v_weights);
        y_values = vqrdmlahq_laneq_s16::<1>(y_values, vreinterpretq_s16_u16(g_values), v_weights);
        y_values = vqrdmlahq_laneq_s16::<2>(y_values, vreinterpretq_s16_u16(b_values), v_weights);

        let mut y_vl = vreinterpretq_u16_s16(vminq_s16(vshrq_n_s16::<2>(y_values), i_cap_y));

        if bytes_position == YuvBytesPacking::MostSignificantBytes {
            y_vl = vtomsbq_u16::<BIT_DEPTH>(y_vl);
        }

        #[cfg(all(feature = "big_endian", target_endian = "little"))]
        if _endianness == YuvEndianness::BigEndian {
            y_vl = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(y_vl)));
        }

        vst1q_u16(y_buffer.as_mut_ptr().cast(), y_vl);

        if chroma_subsampling == YuvChromaSubsampling::Yuv444 {
            let mut cb_h =
                vqrdmlahq_laneq_s16::<3>(uv_bias, vreinterpretq_s16_u16(r_values), v_weights);
            let mut cr_h =
                vqrdmlahq_laneq_s16::<6>(uv_bias, vreinterpretq_s16_u16(r_values), v_weights);
            cb_h = vqrdmlahq_laneq_s16::<4>(cb_h, vreinterpretq_s16_u16(g_values), v_weights);
            cr_h = vqrdmlahq_laneq_s16::<7>(cr_h, vreinterpretq_s16_u16(g_values), v_weights);
            cb_h = vqrdmlahq_laneq_s16::<5>(cb_h, vreinterpretq_s16_u16(b_values), v_weights);
            cr_h = vqrdmlahq_laneq_s16::<0>(cr_h, vreinterpretq_s16_u16(b_values), v_cr_b);

            let cb_max = vmaxq_s16(vshrq_n_s16::<2>(cb_h), i_bias_y);
            let cr_max = vmaxq_s16(vshrq_n_s16::<2>(cr_h), i_bias_y);

            let mut cb_vl = vreinterpretq_u16_s16(vminq_s16(cb_max, i_cap_uv));
            let mut cr_vl = vreinterpretq_u16_s16(vminq_s16(cr_max, i_cap_uv));

            if bytes_position == YuvBytesPacking::MostSignificantBytes {
                cb_vl = vtomsbq_u16::<BIT_DEPTH>(cb_vl);
                cr_vl = vtomsbq_u16::<BIT_DEPTH>(cr_vl);
            }

            #[cfg(all(feature = "big_endian", target_endian = "little"))]
            if _endianness == YuvEndianness::BigEndian {
                cb_vl = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(cb_vl)));
                cr_vl = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(cr_vl)));
            }

            vst1q_u16(u_buffer.as_mut_ptr().cast(), cb_vl);
            vst1q_u16(v_buffer.as_mut_ptr().cast(), cr_vl);
        } else if (chroma_subsampling == YuvChromaSubsampling::Yuv420)
            || (chroma_subsampling == YuvChromaSubsampling::Yuv422)
        {
            let r1 = vreinterpret_s16_u16(vrshrn_n_u32::<1>(vpaddlq_u16(r_values)));
            let g1 = vreinterpret_s16_u16(vrshrn_n_u32::<1>(vpaddlq_u16(g_values)));
            let b1 = vreinterpret_s16_u16(vrshrn_n_u32::<1>(vpaddlq_u16(b_values)));

            let mut cbk = vqrdmlah_laneq_s16::<3>(vget_low_s16(uv_bias), r1, v_weights);
            let mut crk = vqrdmlah_laneq_s16::<6>(vget_low_s16(uv_bias), r1, v_weights);
            cbk = vqrdmlah_laneq_s16::<4>(cbk, g1, v_weights);
            crk = vqrdmlah_laneq_s16::<7>(crk, g1, v_weights);
            cbk = vqrdmlah_laneq_s16::<5>(cbk, b1, v_weights);
            crk = vqrdmlah_laneq_s16::<0>(crk, b1, v_cr_b);

            let cb_max = vmax_s16(vshr_n_s16::<2>(cbk), vget_low_s16(i_bias_y));
            let cr_max = vmax_s16(vshr_n_s16::<2>(crk), vget_low_s16(i_bias_y));

            let mut cb = vreinterpret_u16_s16(vmin_s16(cb_max, vget_low_s16(i_cap_uv)));
            let mut cr = vreinterpret_u16_s16(vmin_s16(cr_max, vget_low_s16(i_cap_uv)));

            if bytes_position == YuvBytesPacking::MostSignificantBytes {
                cb = vtomsb_u16::<BIT_DEPTH>(cb);
                cr = vtomsb_u16::<BIT_DEPTH>(cr);
            }

            #[cfg(all(feature = "big_endian", target_endian = "little"))]
            if _endianness == YuvEndianness::BigEndian {
                cb = vreinterpret_u16_u8(vrev16_u8(vreinterpret_u8_u16(cb)));
                cr = vreinterpret_u16_u8(vrev16_u8(vreinterpret_u8_u16(cr)));
            }

            vst1_u16(u_buffer.as_mut_ptr().cast(), cb);
            vst1_u16(v_buffer.as_mut_ptr().cast(), cr);
        }

        std::ptr::copy_nonoverlapping(
            y_buffer.as_ptr().cast(),
            y_plane.get_unchecked_mut(cx..).as_mut_ptr(),
            diff,
        );

        cx += diff;

        if chroma_subsampling == YuvChromaSubsampling::Yuv444 {
            std::ptr::copy_nonoverlapping(
                u_buffer.as_ptr().cast(),
                u_plane.get_unchecked_mut(ux..).as_mut_ptr(),
                diff,
            );
            std::ptr::copy_nonoverlapping(
                v_buffer.as_ptr().cast(),
                v_plane.get_unchecked_mut(ux..).as_mut_ptr(),
                diff,
            );

            ux += diff;
        } else if (chroma_subsampling == YuvChromaSubsampling::Yuv420)
            || (chroma_subsampling == YuvChromaSubsampling::Yuv422)
        {
            let hv = diff.div_ceil(2);
            std::ptr::copy_nonoverlapping(
                u_buffer.as_ptr().cast(),
                u_plane.get_unchecked_mut(ux..).as_mut_ptr(),
                hv,
            );
            std::ptr::copy_nonoverlapping(
                v_buffer.as_ptr().cast(),
                v_plane.get_unchecked_mut(ux..).as_mut_ptr(),
                hv,
            );

            ux += hv;
        }
    }

    ProcessedOffset { ux, cx }
}