1cfg_if! {
2 // Use the SSE2 implementation if possible: it allows us to scan 16 buckets
3 // at once instead of 8. We don't bother with AVX since it would require
4 // runtime dispatch and wouldn't gain us much anyways: the probability of
5 // finding a match drops off drastically after the first few buckets.
6 //
7 // I attempted an implementation on ARM using NEON instructions, but it
8 // turns out that most NEON instructions have multi-cycle latency, which in
9 // the end outweighs any gains over the generic implementation.
10 if #[cfg(all(
11 target_feature = "sse2",
12 any(target_arch = "x86", target_arch = "x86_64"),
13 not(miri),
14 ))] {
15 mod sse2;
16 use sse2 as imp;
17 } else if #[cfg(all(
18 target_arch = "aarch64",
19 target_feature = "neon",
20 // NEON intrinsics are currently broken on big-endian targets.
21 // See https://github.com/rust-lang/stdarch/issues/1484.
22 target_endian = "little",
23 not(miri),
24 ))] {
25 mod neon;
26 use neon as imp;
27 } else {
28 mod generic;
29 use generic as imp;
30 }
31}
32pub(crate) use self::imp::Group;
33pub(super) use self::imp::{
34 BitMaskWord, NonZeroBitMaskWord, BITMASK_ITER_MASK, BITMASK_MASK, BITMASK_STRIDE,
35};
36