use crate::entropy::{CompactToken, Token};
use crate::lz_match::{match_len_compact_scalar, match_len_scalar};
use std::arch::x86_64::*;
#[target_feature(enable = "avx2")]
pub(crate) fn match_len_avx2(a: &[Token], b: &[Token]) -> usize {
let n = a.len().min(b.len());
let mut i = 0;
unsafe {
while i + 4 <= n {
let x = _mm256_loadu_si256(a.as_ptr().add(i).cast());
let y = _mm256_loadu_si256(b.as_ptr().add(i).cast());
const VALUE_LANES: u32 = 0xF0F0_F0F0;
let mask = _mm256_movemask_epi8(_mm256_cmpeq_epi32(x, y)) as u32 | !VALUE_LANES;
if mask != u32::MAX {
return i + ((!mask).trailing_zeros() as usize / 8);
}
i += 4;
}
}
i + match_len_scalar(&a[i..n], &b[i..n])
}
#[target_feature(enable = "avx2")]
pub(crate) fn match_len_compact_avx2(a: &[CompactToken], b: &[CompactToken]) -> usize {
let n = a.len().min(b.len());
let mut i = 0;
unsafe {
while i + 8 <= n {
let x = _mm256_loadu_si256(a.as_ptr().add(i).cast());
let y = _mm256_loadu_si256(b.as_ptr().add(i).cast());
let value_mask = _mm256_set1_epi32(CompactToken::MAX_VALUE as i32);
let diff = _mm256_and_si256(_mm256_xor_si256(x, y), value_mask);
let mask =
_mm256_movemask_epi8(_mm256_cmpeq_epi32(diff, _mm256_setzero_si256())) as u32;
if mask != u32::MAX {
return i + ((!mask).trailing_zeros() as usize / 4);
}
i += 8;
}
}
i + match_len_compact_scalar(&a[i..n], &b[i..n])
}