Refactor/remove runtime dispatch - #18
Conversation
Select SIMD backends entirely through compile-time target features, remove the CPUID cache and dispatch feature, and document native compilation. Preserve leading-zero performance with short-string and AVX2 all-zero scan optimizations.
Match the AVX2 forward scanner with a four-vector reverse scan, add two-word fast paths for owned and aligned borrowed strings, and group leading/trailing benchmark references by comparable cases.
Route BitString matches-at and suffix stripping through BitStr alignment-aware entry points. Add regression coverage for multi-word patterns at unaligned offsets and oversized suffixes.
Define true alignment consts as caller guarantees, enforce them with debug assertions, and cover matching, equality, ordering, and hashing across word-boundary lengths and all alignment combinations.
|
Codecov Report❌ Patch coverage is
📢 Thoughts on this report? Let us know! |
Merging this PR will regress 5 benchmarks
|
No description provided.