colconv 0.1.0

use super::*;

// ---- Ship 9b RGBA/BGRA shuffles -----------------------------------------

fn pseudo_random_rgba(width: usize) -> std::vec::Vec<u8> {
  let n = width * 4;
  let mut out = std::vec::Vec::with_capacity(n);
  let mut state: u32 = 0x6CCD_5C7B;
  for _ in 0..n {
    state = state.wrapping_mul(1_664_525).wrapping_add(1_013_904_223);
    out.push((state >> 8) as u8);
  }
  out
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn rgba_to_rgb_neon_matches_scalar_widths() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u8; w * 3];
    let mut out_neon = std::vec![0u8; w * 3];
    scalar::rgba_to_rgb_row(&input, &mut out_scalar, w);
    unsafe {
      rgba_to_rgb_row(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
    // Direct semantic check: every output triple = first 3 of input quad.
    for x in 0..w {
      assert_eq!(out_neon[x * 3], input[x * 4], "R width {w} px {x}");
      assert_eq!(out_neon[x * 3 + 1], input[x * 4 + 1], "G width {w} px {x}");
      assert_eq!(out_neon[x * 3 + 2], input[x * 4 + 2], "B width {w} px {x}");
    }
  }
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn bgra_to_rgba_neon_matches_scalar_widths() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u8; w * 4];
    let mut out_neon = std::vec![0u8; w * 4];
    scalar::bgra_to_rgba_row(&input, &mut out_scalar, w);
    unsafe {
      bgra_to_rgba_row(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
    // R↔B swap, G + alpha unchanged.
    for x in 0..w {
      assert_eq!(out_neon[x * 4], input[x * 4 + 2], "R width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 1], input[x * 4 + 1], "G width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 2], input[x * 4], "B width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 3], input[x * 4 + 3], "A width {w} px {x}");
    }
  }
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn bgra_to_rgb_neon_matches_scalar_widths() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u8; w * 3];
    let mut out_neon = std::vec![0u8; w * 3];
    scalar::bgra_to_rgb_row(&input, &mut out_scalar, w);
    unsafe {
      bgra_to_rgb_row(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
    // R↔B swap, G unchanged, alpha dropped.
    for x in 0..w {
      assert_eq!(out_neon[x * 3], input[x * 4 + 2], "R width {w} px {x}");
      assert_eq!(out_neon[x * 3 + 1], input[x * 4 + 1], "G width {w} px {x}");
      assert_eq!(out_neon[x * 3 + 2], input[x * 4], "B width {w} px {x}");
    }
  }
}

// ---- Ship 9c leading-alpha shuffles -----------------------------------

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn argb_to_rgb_neon_matches_scalar_widths() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u8; w * 3];
    let mut out_neon = std::vec![0u8; w * 3];
    scalar::argb_to_rgb_row(&input, &mut out_scalar, w);
    unsafe {
      argb_to_rgb_row(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
    // Drop leading alpha — output triple = input bytes 1, 2, 3.
    for x in 0..w {
      assert_eq!(out_neon[x * 3], input[x * 4 + 1], "R width {w} px {x}");
      assert_eq!(out_neon[x * 3 + 1], input[x * 4 + 2], "G width {w} px {x}");
      assert_eq!(out_neon[x * 3 + 2], input[x * 4 + 3], "B width {w} px {x}");
    }
  }
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn abgr_to_rgb_neon_matches_scalar_widths() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u8; w * 3];
    let mut out_neon = std::vec![0u8; w * 3];
    scalar::abgr_to_rgb_row(&input, &mut out_scalar, w);
    unsafe {
      abgr_to_rgb_row(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
    // Reversed inner three bytes — output = input bytes 3, 2, 1.
    for x in 0..w {
      assert_eq!(out_neon[x * 3], input[x * 4 + 3], "R width {w} px {x}");
      assert_eq!(out_neon[x * 3 + 1], input[x * 4 + 2], "G width {w} px {x}");
      assert_eq!(out_neon[x * 3 + 2], input[x * 4 + 1], "B width {w} px {x}");
    }
  }
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn argb_to_rgba_neon_matches_scalar_widths() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u8; w * 4];
    let mut out_neon = std::vec![0u8; w * 4];
    scalar::argb_to_rgba_row(&input, &mut out_scalar, w);
    unsafe {
      argb_to_rgba_row(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
    // Rotate alpha to trailing — output = input bytes 1, 2, 3, 0.
    for x in 0..w {
      assert_eq!(out_neon[x * 4], input[x * 4 + 1], "R width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 1], input[x * 4 + 2], "G width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 2], input[x * 4 + 3], "B width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 3], input[x * 4], "A width {w} px {x}");
    }
  }
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn abgr_to_rgba_neon_matches_scalar_widths() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u8; w * 4];
    let mut out_neon = std::vec![0u8; w * 4];
    scalar::abgr_to_rgba_row(&input, &mut out_scalar, w);
    unsafe {
      abgr_to_rgba_row(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
    // Full byte reverse — output = input bytes 3, 2, 1, 0.
    for x in 0..w {
      assert_eq!(out_neon[x * 4], input[x * 4 + 3], "R width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 1], input[x * 4 + 2], "G width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 2], input[x * 4 + 1], "B width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 3], input[x * 4], "A width {w} px {x}");
    }
  }
}

// ---- Ship 9d padding-byte shuffles -----------------------------------

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn xrgb_to_rgba_neon_matches_scalar_widths() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u8; w * 4];
    let mut out_neon = std::vec![0u8; w * 4];
    scalar::xrgb_to_rgba_row(&input, &mut out_scalar, w);
    unsafe {
      xrgb_to_rgba_row(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
    // Drop leading byte; alpha forced to 0xFF.
    for x in 0..w {
      assert_eq!(out_neon[x * 4], input[x * 4 + 1], "R width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 1], input[x * 4 + 2], "G width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 2], input[x * 4 + 3], "B width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 3], 0xFF, "A width {w} px {x}");
    }
  }
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn rgbx_to_rgba_neon_matches_scalar_widths() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u8; w * 4];
    let mut out_neon = std::vec![0u8; w * 4];
    scalar::rgbx_to_rgba_row(&input, &mut out_scalar, w);
    unsafe {
      rgbx_to_rgba_row(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
    for x in 0..w {
      assert_eq!(out_neon[x * 4], input[x * 4], "R width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 1], input[x * 4 + 1], "G width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 2], input[x * 4 + 2], "B width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 3], 0xFF, "A width {w} px {x}");
    }
  }
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn xbgr_to_rgba_neon_matches_scalar_widths() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u8; w * 4];
    let mut out_neon = std::vec![0u8; w * 4];
    scalar::xbgr_to_rgba_row(&input, &mut out_scalar, w);
    unsafe {
      xbgr_to_rgba_row(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
    for x in 0..w {
      assert_eq!(out_neon[x * 4], input[x * 4 + 3], "R width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 1], input[x * 4 + 2], "G width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 2], input[x * 4 + 1], "B width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 3], 0xFF, "A width {w} px {x}");
    }
  }
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn bgrx_to_rgba_neon_matches_scalar_widths() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u8; w * 4];
    let mut out_neon = std::vec![0u8; w * 4];
    scalar::bgrx_to_rgba_row(&input, &mut out_scalar, w);
    unsafe {
      bgrx_to_rgba_row(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
    for x in 0..w {
      assert_eq!(out_neon[x * 4], input[x * 4 + 2], "R width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 1], input[x * 4 + 1], "G width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 2], input[x * 4], "B width {w} px {x}");
      assert_eq!(out_neon[x * 4 + 3], 0xFF, "A width {w} px {x}");
    }
  }
}

// ---- Ship 9e 10-bit packed RGB ---------------------------------------

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn x2rgb10_to_rgb_neon_matches_scalar_widths() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u8; w * 3];
    let mut out_neon = std::vec![0u8; w * 3];
    scalar::x2rgb10_to_rgb_row::<false>(&input, &mut out_scalar, w);
    unsafe {
      x2rgb10_to_rgb_row::<false>(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
  }
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn x2rgb10_to_rgba_neon_matches_scalar_widths() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u8; w * 4];
    let mut out_neon = std::vec![0u8; w * 4];
    scalar::x2rgb10_to_rgba_row::<false>(&input, &mut out_scalar, w);
    unsafe {
      x2rgb10_to_rgba_row::<false>(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
  }
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn x2rgb10_to_rgb_u16_neon_matches_scalar_widths() {
  for w in [1usize, 7, 8, 9, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u16; w * 3];
    let mut out_neon = std::vec![0u16; w * 3];
    scalar::x2rgb10_to_rgb_u16_row::<false>(&input, &mut out_scalar, w);
    unsafe {
      x2rgb10_to_rgb_u16_row::<false>(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
  }
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn x2bgr10_to_rgb_neon_matches_scalar_widths() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u8; w * 3];
    let mut out_neon = std::vec![0u8; w * 3];
    scalar::x2bgr10_to_rgb_row::<false>(&input, &mut out_scalar, w);
    unsafe {
      x2bgr10_to_rgb_row::<false>(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
  }
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn x2bgr10_to_rgba_neon_matches_scalar_widths() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u8; w * 4];
    let mut out_neon = std::vec![0u8; w * 4];
    scalar::x2bgr10_to_rgba_row::<false>(&input, &mut out_scalar, w);
    unsafe {
      x2bgr10_to_rgba_row::<false>(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
  }
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn x2bgr10_to_rgb_u16_neon_matches_scalar_widths() {
  for w in [1usize, 7, 8, 9, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut out_scalar = std::vec![0u16; w * 3];
    let mut out_neon = std::vec![0u16; w * 3];
    scalar::x2bgr10_to_rgb_u16_row::<false>(&input, &mut out_scalar, w);
    unsafe {
      x2bgr10_to_rgb_u16_row::<false>(&input, &mut out_neon, w);
    }
    assert_eq!(out_scalar, out_neon, "width {w}");
  }
}

// SIMD-level BE-vs-LE parity for X2RGB10 / X2BGR10.
//
// Both `<false>` and `<true>` exercise the real SIMD body (LE via plain
// `vld1q_u32`, BE via `load_endian_u32x4::<true>`'s extra `vrev32q_u8`).
// Width 33 crosses the 16-pixel SIMD boundary so the body runs at least
// twice and the scalar tail handles the leftover.

fn pseudo_random_x2_intended(width: usize, seed: u32) -> std::vec::Vec<u32> {
  let mut state = seed;
  (0..width)
    .map(|_| {
      state = state.wrapping_mul(1_664_525).wrapping_add(1_013_904_223);
      state
    })
    .collect()
}

fn make_le_be_pair_x2(intended: &[u32]) -> (std::vec::Vec<u8>, std::vec::Vec<u8>) {
  let le_bytes: std::vec::Vec<u8> = intended.iter().flat_map(|v| v.to_le_bytes()).collect();
  let be_bytes: std::vec::Vec<u8> = intended.iter().flat_map(|v| v.to_be_bytes()).collect();
  (le_bytes, be_bytes)
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn neon_x2rgb10_be_le_simd_parity_width33() {
  let intended = pseudo_random_x2_intended(33, 0xC0DE_BEEF);
  let (le, be) = make_le_be_pair_x2(&intended);

  let mut out_le = std::vec![0u8; 33 * 3];
  let mut out_be = std::vec![0u8; 33 * 3];
  unsafe {
    x2rgb10_to_rgb_row::<false>(&le, &mut out_le, 33);
    x2rgb10_to_rgb_row::<true>(&be, &mut out_be, 33);
  }
  assert_eq!(out_le, out_be, "x2rgb10→rgb SIMD BE/LE parity");

  let mut out_le = std::vec![0u8; 33 * 4];
  let mut out_be = std::vec![0u8; 33 * 4];
  unsafe {
    x2rgb10_to_rgba_row::<false>(&le, &mut out_le, 33);
    x2rgb10_to_rgba_row::<true>(&be, &mut out_be, 33);
  }
  assert_eq!(out_le, out_be, "x2rgb10→rgba SIMD BE/LE parity");

  let mut out_le = std::vec![0u16; 33 * 3];
  let mut out_be = std::vec![0u16; 33 * 3];
  unsafe {
    x2rgb10_to_rgb_u16_row::<false>(&le, &mut out_le, 33);
    x2rgb10_to_rgb_u16_row::<true>(&be, &mut out_be, 33);
  }
  assert_eq!(out_le, out_be, "x2rgb10→rgb_u16 SIMD BE/LE parity");
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn neon_x2bgr10_be_le_simd_parity_width33() {
  let intended = pseudo_random_x2_intended(33, 0xFEED_FACE);
  let (le, be) = make_le_be_pair_x2(&intended);

  let mut out_le = std::vec![0u8; 33 * 3];
  let mut out_be = std::vec![0u8; 33 * 3];
  unsafe {
    x2bgr10_to_rgb_row::<false>(&le, &mut out_le, 33);
    x2bgr10_to_rgb_row::<true>(&be, &mut out_be, 33);
  }
  assert_eq!(out_le, out_be, "x2bgr10→rgb SIMD BE/LE parity");

  let mut out_le = std::vec![0u8; 33 * 4];
  let mut out_be = std::vec![0u8; 33 * 4];
  unsafe {
    x2bgr10_to_rgba_row::<false>(&le, &mut out_le, 33);
    x2bgr10_to_rgba_row::<true>(&be, &mut out_be, 33);
  }
  assert_eq!(out_le, out_be, "x2bgr10→rgba SIMD BE/LE parity");

  let mut out_le = std::vec![0u16; 33 * 3];
  let mut out_be = std::vec![0u16; 33 * 3];
  unsafe {
    x2bgr10_to_rgb_u16_row::<false>(&le, &mut out_le, 33);
    x2bgr10_to_rgb_u16_row::<true>(&be, &mut out_be, 33);
  }
  assert_eq!(out_le, out_be, "x2bgr10→rgb_u16 SIMD BE/LE parity");
}

// BE-input SIMD-vs-scalar parity.
//
// Without a BE-aware load (`load_endian_u32x4::<BE>`) the NEON X2 10-bit
// kernels gate their SIMD body on `if !BE` and silently fall through to
// scalar for BE input.

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn neon_x2rgb10_be_simd_matches_scalar() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut s_rgb = std::vec![0u8; w * 3];
    let mut n_rgb = std::vec![0u8; w * 3];
    scalar::x2rgb10_to_rgb_row::<true>(&input, &mut s_rgb, w);
    unsafe {
      x2rgb10_to_rgb_row::<true>(&input, &mut n_rgb, w);
    }
    assert_eq!(s_rgb, n_rgb, "NEON x2rgb10_to_rgb<BE> diverges (width={w})");

    let mut s_rgba = std::vec![0u8; w * 4];
    let mut n_rgba = std::vec![0u8; w * 4];
    scalar::x2rgb10_to_rgba_row::<true>(&input, &mut s_rgba, w);
    unsafe {
      x2rgb10_to_rgba_row::<true>(&input, &mut n_rgba, w);
    }
    assert_eq!(
      s_rgba, n_rgba,
      "NEON x2rgb10_to_rgba<BE> diverges (width={w})"
    );

    let mut s_u16 = std::vec![0u16; w * 3];
    let mut n_u16 = std::vec![0u16; w * 3];
    scalar::x2rgb10_to_rgb_u16_row::<true>(&input, &mut s_u16, w);
    unsafe {
      x2rgb10_to_rgb_u16_row::<true>(&input, &mut n_u16, w);
    }
    assert_eq!(
      s_u16, n_u16,
      "NEON x2rgb10_to_rgb_u16<BE> diverges (width={w})"
    );
  }
}

#[test]
#[cfg_attr(miri, ignore = "NEON SIMD intrinsics unsupported by Miri")]
fn neon_x2bgr10_be_simd_matches_scalar() {
  for w in [1usize, 7, 15, 16, 17, 31, 32, 33, 1920, 1921] {
    let input = pseudo_random_rgba(w);
    let mut s_rgb = std::vec![0u8; w * 3];
    let mut n_rgb = std::vec![0u8; w * 3];
    scalar::x2bgr10_to_rgb_row::<true>(&input, &mut s_rgb, w);
    unsafe {
      x2bgr10_to_rgb_row::<true>(&input, &mut n_rgb, w);
    }
    assert_eq!(s_rgb, n_rgb, "NEON x2bgr10_to_rgb<BE> diverges (width={w})");

    let mut s_rgba = std::vec![0u8; w * 4];
    let mut n_rgba = std::vec![0u8; w * 4];
    scalar::x2bgr10_to_rgba_row::<true>(&input, &mut s_rgba, w);
    unsafe {
      x2bgr10_to_rgba_row::<true>(&input, &mut n_rgba, w);
    }
    assert_eq!(
      s_rgba, n_rgba,
      "NEON x2bgr10_to_rgba<BE> diverges (width={w})"
    );

    let mut s_u16 = std::vec![0u16; w * 3];
    let mut n_u16 = std::vec![0u16; w * 3];
    scalar::x2bgr10_to_rgb_u16_row::<true>(&input, &mut s_u16, w);
    unsafe {
      x2bgr10_to_rgb_u16_row::<true>(&input, &mut n_u16, w);
    }
    assert_eq!(
      s_u16, n_u16,
      "NEON x2bgr10_to_rgb_u16<BE> diverges (width={w})"
    );
  }
}