1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
//! Donnelly-family stem strategies.
//!
//! The Donnelly family shares the same broad layout idea: stems are arranged in
//! fixed-height minor triangles parameterized by a `const BH: usize` block
//! height, so traversal stays cache-friendly while still supporting the normal
//! `StemStrategy` query surface.
//!
//! The variants differ by how much traversal work is specialized:
//!
//! - [`Donnelly`] is the default scalar variant. It traverses one level at a
//! time, advances dimensions once per level, and includes the current
//! software-prefetch behavior.
//! - [`DonnellyNoPf`] is the same scalar traversal shape without the “default”
//! naming. It exists so the public API can distinguish the non-prefetched
//! scalar baseline from [`Donnelly`].
//! - [`DonnellyUnrolled`] keeps the same Donnelly ordering and per-level
//! dimension cadence, but unrolls traversal within each minor triangle.
//! - [`DonnellyUnrolledBlockDim`] uses the same unrolled structure, but changes
//! dimensions once per block rather than once per level. It is primarily the
//! scalar reference variant for the more specialized block-at-once traversal
//! strategies.
//! - [`DonnellySimdDescent`] performs block-at-once SIMD child selection during
//! descent, but still uses scalar backtracking and pruning.
//! - [`DonnellySimdFull`] takes the same block-at-once descent idea and also
//! uses SIMD-aware backtracking and pruning.
//!
//! Internally, these variants share `core` for scalar Donnelly indexing/state
//! and `simd_full` for the reusable SIMD comparison and backtrack machinery.
pub use DonnellyNoPf;
pub use Donnelly;
pub use DonnellySimdDescent;
pub use DonnellySimdFull;
pub use DonnellyUnrolled;
pub use DonnellyUnrolledBlockDim;