use super::xyb::rgb_to_xyb_f32x4_neon;
use crate::image::Image3F;
use crate::xyb::XybMatrix;
use std::arch::aarch64::*;
#[target_feature(enable = "neon")]
pub(crate) fn sampled_b_gradient_ratio_neon(linear: &Image3F, stride: usize) -> f32 {
let (w, h) = (linear.xsize(), linear.ysize());
let [rp, gp, bp] = std::array::from_fn(|c| linear.plane_data(c));
let (mut sum_y, mut sum_b) = (0.0, 0.0);
for y in (0..h).step_by(stride) {
for x in (0..w).step_by(stride) {
let i = y * w + x;
let right = if x + 1 < w { i + 1 } else { i };
let below = if y + 1 < h { i + w } else { i };
let gather = |p: &[f32]| unsafe { vld1q_f32([p[i], p[right], p[below], 0.0].as_ptr()) };
let (_, yv, bv) =
rgb_to_xyb_f32x4_neon::<true>(&XybMatrix::SPEC, gather(rp), gather(gp), gather(bp));
let bv = vsubq_f32(bv, yv);
let mut ys = [0.0; 4];
let mut bs = [0.0; 4];
unsafe {
vst1q_f32(ys.as_mut_ptr(), yv);
vst1q_f32(bs.as_mut_ptr(), bv);
}
if x + 1 < w {
sum_y += (ys[1] - ys[0]).abs();
sum_b += (bs[1] - bs[0]).abs();
}
if y + 1 < h {
sum_y += (ys[2] - ys[0]).abs();
sum_b += (bs[2] - bs[0]).abs();
}
}
}
if sum_y > f32::EPSILON {
sum_b / sum_y
} else {
0.0
}
}