#[cfg(target_arch = "x86")]
use core::arch::x86::*;
#[cfg(target_arch = "x86_64")]
use core::arch::x86_64::*;
#[target_feature(enable = "sse4.1")]
pub(crate) fn polish_count_nonzero_sse41(values: &[i32]) -> usize {
let mut count = 0;
let mut i = 0;
while i + 4 <= values.len() {
let v = unsafe { _mm_loadu_si128(values.as_ptr().add(i).cast()) };
let eq = _mm_cmpeq_epi32(v, _mm_setzero_si128());
count += _mm_movemask_ps(_mm_castsi128_ps(eq)).count_ones() as usize;
i += 4;
}
values.len() - count - values[i..].iter().filter(|&&v| v == 0).count()
}
#[target_feature(enable = "sse4.1")]
pub(crate) fn polish_add_scaled_sse41(row: &mut [f32], delta: &[f32], sign: f32) {
assert_eq!(row.len(), delta.len());
let factor = _mm_set1_ps(sign);
let mut i = 0;
while i + 4 <= row.len() {
let v = unsafe { _mm_loadu_ps(row.as_ptr().add(i)) };
let d = unsafe { _mm_loadu_ps(delta.as_ptr().add(i)) };
let scaled = _mm_mul_ps(d, factor);
let sum = _mm_add_ps(v, scaled);
unsafe { _mm_storeu_ps(row.as_mut_ptr().add(i), sum) };
i += 4;
}
for (v, &d) in row[i..].iter_mut().zip(&delta[i..]) {
*v += sign * d;
}
}