pic-scale 0.7.8

/*
 * 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::neon::utils::{
    expand8_high_to_14, expand8_to_14, xvld1q_s16_x2, xvld1q_s16_x4, xvld1q_u8_x2,
};
use std::arch::aarch64::*;

#[inline]
#[target_feature(enable = "rdm")]
fn accumulate_32_horiz(store: int16x8_t, ptr: *const u8, weights: int16x8x4_t) -> int16x8_t {
    unsafe {
        let pixel_colors = xvld1q_u8_x2(ptr);
        let lo0 = expand8_to_14(vget_low_u8(pixel_colors.0));
        let hi0 = expand8_high_to_14(pixel_colors.0);
        let lo1 = expand8_to_14(vget_low_u8(pixel_colors.1));
        let hi1 = expand8_high_to_14(pixel_colors.1);
        let mut p = vqrdmlahq_s16(store, lo0, weights.0);
        p = vqrdmlahq_s16(p, hi0, weights.1);
        p = vqrdmlahq_s16(p, lo1, weights.2);
        vqrdmlahq_s16(p, hi1, weights.3)
    }
}

#[inline]
#[target_feature(enable = "rdm")]
fn accumulate_16_horiz(store: int16x8_t, ptr: *const u8, weights: int16x8x2_t) -> int16x8_t {
    unsafe {
        let pixel_colors = vld1q_u8(ptr);
        let lo = expand8_to_14(vget_low_u8(pixel_colors));
        let hi = expand8_high_to_14(pixel_colors);
        let p = vqrdmlahq_s16(store, lo, weights.0);
        vqrdmlahq_s16(p, hi, weights.1)
    }
}

#[inline]
#[target_feature(enable = "rdm")]
fn accumulate_8_horiz(store: int16x8_t, ptr: *const u8, weights: int16x8_t) -> int16x8_t {
    unsafe {
        let pixel_colors = vld1_u8(ptr);
        let lo = expand8_to_14(pixel_colors);
        vqrdmlahq_s16(store, lo, weights)
    }
}

#[inline]
#[target_feature(enable = "rdm")]
fn accumulate_4_horiz(store: int16x8_t, ptr: *const u8, weights: int16x8_t) -> int16x8_t {
    unsafe {
        let pixel_colors =
            vreinterpret_u8_u32(vld1_lane_u32::<0>(ptr as *const u32, vdup_n_u32(0)));
        let lo = expand8_to_14(pixel_colors);
        vqrdmlahq_s16(store, lo, weights)
    }
}

#[inline]
#[target_feature(enable = "rdm")]
fn accumulate_1_horiz(store: int16x8_t, ptr: *const u8, weights: int16x8_t) -> int16x8_t {
    unsafe {
        let pixel_colors = vld1_lane_u8::<0>(ptr, vdup_n_u8(0));
        let lo = expand8_to_14(pixel_colors);
        vqrdmlahq_s16(store, lo, weights)
    }
}

pub(crate) fn convolve_horizontal_plane_neon_rows_rdm_4_u8(
    src: &[u8],
    src_stride: usize,
    dst: &mut [u8],
    dst_stride: usize,
    filter_weights: &FilterWeights<i16>,
    _: u32,
) {
    unsafe {
        convolve_horizontal_plane_neon_rows_4_u8_impl(
            src,
            src_stride,
            dst,
            dst_stride,
            filter_weights,
        );
    }
}

#[target_feature(enable = "rdm")]
fn convolve_horizontal_plane_neon_rows_4_u8_impl(
    src: &[u8],
    src_stride: usize,
    dst: &mut [u8],
    dst_stride: usize,
    filter_weights: &FilterWeights<i16>,
) {
    unsafe {
        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.iter_mut();
        let iter_row1 = row1_ref.iter_mut();
        let iter_row2 = row2_ref.iter_mut();
        let iter_row3 = row3_ref.iter_mut();

        const PRECISION: i16 = 6;
        const ROUNDING_CONST: i16 = 1 << (PRECISION - 1);
        let base_val = {
            let j = vdupq_n_s16(0);
            vsetq_lane_s16::<0>(ROUNDING_CONST, j)
        };

        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 store0 = base_val;
            let mut store1 = base_val;
            let mut store2 = base_val;
            let mut store3 = base_val;

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

            while jx + 32 <= bounds.size {
                let w_ptr = weights.get_unchecked(jx..);
                let weights = xvld1q_s16_x4(w_ptr.as_ptr());
                let bounds_start = bounds.start + jx;

                let src_ptr = src0.get_unchecked(bounds_start..);
                store0 = accumulate_32_horiz(store0, src_ptr.as_ptr(), weights);

                let src_ptr1 = src1.get_unchecked(bounds_start..);
                store1 = accumulate_32_horiz(store1, src_ptr1.as_ptr(), weights);

                let src_ptr2 = src2.get_unchecked(bounds_start..);
                store2 = accumulate_32_horiz(store2, src_ptr2.as_ptr(), weights);

                let src_ptr3 = src3.get_unchecked(bounds_start..);
                store3 = accumulate_32_horiz(store3, src_ptr3.as_ptr(), weights);

                jx += 32;
            }

            while jx + 16 <= bounds.size {
                let w_ptr = weights.get_unchecked(jx..);
                let weights = xvld1q_s16_x2(w_ptr.as_ptr());
                let bounds_start = bounds.start + jx;

                let src_ptr = src0.get_unchecked(bounds_start..);
                store0 = accumulate_16_horiz(store0, src_ptr.as_ptr(), weights);

                let src_ptr1 = src1.get_unchecked(bounds_start..);
                store1 = accumulate_16_horiz(store1, src_ptr1.as_ptr(), weights);

                let src_ptr2 = src2.get_unchecked(bounds_start..);
                store2 = accumulate_16_horiz(store2, src_ptr2.as_ptr(), weights);

                let src_ptr3 = src3.get_unchecked(bounds_start..);
                store3 = accumulate_16_horiz(store3, src_ptr3.as_ptr(), weights);

                jx += 16;
            }

            while jx + 8 <= bounds.size {
                let w_ptr = weights.get_unchecked(jx..);
                let weights = vld1q_s16(w_ptr.as_ptr());
                let bounds_start = bounds.start + jx;

                let src_ptr = src0.get_unchecked(bounds_start..);
                store0 = accumulate_8_horiz(store0, src_ptr.as_ptr(), weights);

                let src_ptr1 = src1.get_unchecked(bounds_start..);
                store1 = accumulate_8_horiz(store1, src_ptr1.as_ptr(), weights);

                let src_ptr2 = src2.get_unchecked(bounds_start..);
                store2 = accumulate_8_horiz(store2, src_ptr2.as_ptr(), weights);

                let src_ptr3 = src3.get_unchecked(bounds_start..);
                store3 = accumulate_8_horiz(store3, src_ptr3.as_ptr(), weights);

                jx += 8;
            }

            while jx + 4 <= bounds.size {
                let w_ptr = weights.get_unchecked(jx..);
                let weights = vcombine_s16(vld1_s16(w_ptr.as_ptr()), vdup_n_s16(0));
                let bounds_start = bounds.start + jx;

                let src_ptr = src0.get_unchecked(bounds_start..);
                store0 = accumulate_4_horiz(store0, src_ptr.as_ptr(), weights);

                let src_ptr1 = src1.get_unchecked(bounds_start..);
                store1 = accumulate_4_horiz(store1, src_ptr1.as_ptr(), weights);

                let src_ptr2 = src2.get_unchecked(bounds_start..);
                store2 = accumulate_4_horiz(store2, src_ptr2.as_ptr(), weights);

                let src_ptr3 = src3.get_unchecked(bounds_start..);
                store3 = accumulate_4_horiz(store3, src_ptr3.as_ptr(), weights);

                jx += 4;
            }

            while jx < bounds.size {
                let w_ptr = weights.get_unchecked(jx..);
                let weight = vcombine_s16(
                    vld1_lane_s16::<0>(w_ptr.as_ptr(), vdup_n_s16(0)),
                    vdup_n_s16(0),
                );
                let bounds_start = bounds.start + jx;

                let src_ptr = src0.get_unchecked(bounds_start..);
                store0 = accumulate_1_horiz(store0, src_ptr.as_ptr(), weight);

                let src_ptr1 = src1.get_unchecked(bounds_start..);
                store1 = accumulate_1_horiz(store1, src_ptr1.as_ptr(), weight);

                let src_ptr2 = src2.get_unchecked(bounds_start..);
                store2 = accumulate_1_horiz(store2, src_ptr2.as_ptr(), weight);

                let src_ptr3 = src3.get_unchecked(bounds_start..);
                store3 = accumulate_1_horiz(store3, src_ptr3.as_ptr(), weight);

                jx += 1;
            }

            let sums = vaddvq_s16(store0).max(0);
            let shifted = sums >> PRECISION;
            let value = shifted.min(255) as u8;
            *chunk0 = value;

            let sums = vaddvq_s16(store1).max(0);
            let shifted = sums >> PRECISION;
            let value = shifted.min(255) as u8;
            *chunk1 = value;

            let sums = vaddvq_s16(store2).max(0);
            let shifted = sums >> PRECISION;
            let value = shifted.min(255) as u8;
            *chunk2 = value;

            let sums = vaddvq_s16(store3).max(0);
            let shifted = sums >> PRECISION;
            let value = shifted.min(255) as u8;
            *chunk3 = value;
        }
    }
}

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

#[target_feature(enable = "rdm")]
fn convolve_horizontal_plane_neon_rdm_row_impl(
    src: &[u8],
    dst: &mut [u8],
    filter_weights: &FilterWeights<i16>,
) {
    unsafe {
        const PRECISION: i16 = 6;
        const ROUNDING_CONST: i16 = 1 << (PRECISION - 1);
        let base_val = {
            let j = vdupq_n_s16(0);
            vsetq_lane_s16::<0>(ROUNDING_CONST, j)
        };

        for ((dst, bounds), weights) in dst.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 = base_val;

            while jx + 32 <= bounds_size {
                let w_ptr = weights.get_unchecked(jx..);
                let weights = xvld1q_s16_x4(w_ptr.as_ptr());
                let bounds_start = bounds.start + jx;

                let src_ptr = src.get_unchecked(bounds_start..).as_ptr();
                store = accumulate_32_horiz(store, src_ptr, weights);

                jx += 32;
            }

            while jx + 16 <= bounds_size {
                let w_ptr = weights.get_unchecked(jx..);
                let weights = xvld1q_s16_x2(w_ptr.as_ptr());
                let bounds_start = bounds.start + jx;

                let src_ptr = src.get_unchecked(bounds_start..).as_ptr();
                store = accumulate_16_horiz(store, src_ptr, weights);

                jx += 16;
            }

            while jx + 8 <= bounds_size {
                let w_ptr = weights.get_unchecked(jx..);
                let weights = vld1q_s16(w_ptr.as_ptr());
                let bounds_start = bounds.start + jx;

                let src_ptr = src.get_unchecked(bounds_start..).as_ptr();
                store = accumulate_8_horiz(store, src_ptr, weights);

                jx += 8;
            }

            while jx + 4 <= bounds_size {
                let w_ptr = weights.get_unchecked(jx..);
                let weights = vcombine_s16(vld1_s16(w_ptr.as_ptr()), vdup_n_s16(0));
                let bounds_start = bounds.start + jx;

                let src_ptr = src.get_unchecked(bounds_start..).as_ptr();
                store = accumulate_4_horiz(store, src_ptr, weights);

                jx += 4;
            }

            while jx < bounds_size {
                let w_ptr = weights.get_unchecked(jx..);
                let weight = vcombine_s16(
                    vld1_lane_s16::<0>(w_ptr.as_ptr(), vdup_n_s16(0)),
                    vdup_n_s16(0),
                );
                let bounds_start = bounds.start + jx;
                let src_ptr = src.get_unchecked(bounds_start..).as_ptr();
                store = accumulate_1_horiz(store, src_ptr, weight);
                jx += 1;
            }

            let sums = vaddvq_s16(store).max(0);
            let shifted = sums >> PRECISION;
            let value = shifted.min(255) as u8;
            *dst = value;
        }
    }
}