Perf/optimize leading trailing - #17

Merged

jcfangc merged 75 commits into

mainfrom

perf/optimize-leading-trailing

Jul 4, 2026

Commits on Jun 28, 2026

perf: const-generic FILL_ONES, split leading/trailing modules, baseline benchmarks

committed
0df3b52
refactor: extract funcs_for_value_words_core, const-generic FILL: u64

committed
cba8ab5
perf: replace lane-scan with ptest skip-while in leading/trailing SIMD

committed
f56d9f3
perf: eliminate word-level SIMD layer — inline ptest directly into bit scans

committed
e7fa73f
refactor: extract funcs_for_value_bits_core with chunk_eq, add trait methods

committed
1ac4969
refactor: route BitStr leading/trailing through trait inner methods

committed
c879eb5
test: add backend equivalence tests for leading/trailing

committed
bfb0d8d
refactor: flatten leading/trailing core into separate files

committed
96471f0
feat: add HS_WORD_ALIGNED/ND_WORD_ALIGNED to bits_equal_at

committed
6e56dc2
perf: add dedicated starts_with_inner/ends_with_inner with alignment signals

committed
d52fabf
perf: add contains_inner with HS_WORD_ALIGNED/ND_WORD_ALIGNED

committed
26809d5
refactor: rename public APIs with _str/_string suffix convention

committed
47cc405
perf: add count_ones_inner::<WORD_ALIGNED> to BitStr

committed
acb909c
perf: add cmp_inner::<HS_WORD_ALIGNED, ND_WORD_ALIGNED>

committed
f5c3f39
perf: add find_inner/rfind_inner::<WORD_ALIGNED>

committed
51a8ee5
perf: add alignment dispatch to matches_at_str, add matches_at_string

committed
07de745
perf: add hash_inner::<WORD_ALIGNED, H> to BitStr

committed
71e0b82
perf: add alignment dispatch to PartialEq::eq

committed
bd2be90
perf: add strip_prefix/suffix inner methods and _string variants

committed
ff947c3
fix: thread alignment signals through find/rfind verification callbacks

committed
faa1a20
refactor: remove lossy bits_equal_at wrapper

committed
2d1dab9
fix: HS_WORD_ALIGNED must account for offset in ends_with

committed
1e77df7
perf: skip haystack_shift modulo when HS_WORD_ALIGNED is true

committed
6a8cac2
refactor: merge eq_words aligned/unaligned cores into one file

committed
0517a7e
refactor: add HS_WORD_ALIGNED to BitsOrd::cmp_words

committed
b4f3778
perf: add WORD_ALIGNED to read_word_at and write_word_at

committed
f38a745
chore: fix warnings, add ours_str/ours_string bench entries

committed
781cb9f
chore: update codspeed filter to include matching/ord/hash benches

committed
c20fb75
bench: rename bit_string→ours_str/ours_string, codspeed filter→ours_

committed
c6c1bee
refactor: extract count_ones_inner to mod inner

committed
5cac48e
refactor: extract hash_inner and cmp_inner to mod inner

committed
1063d8a
refactor: extract all pub(crate) inner methods to mod inner

committed
c0251d8
bench: 4-mode starts_with/ends_with, fix ours underscore

committed
238d1df
bench: rename all bench entries to ours_string_str etc.

committed
b714724
fix: /ours" → /ours_string" in bench names

committed
be9aded
bench: expand starts_with/ends_with to helper function pattern

committed
eaeef45
refactor: descriptive bench function names across all files

committed
88bf87c
refactor: extract WordsScan trait for counting/scanning ops

committed
2960b81

Commits on Jun 29, 2026

perf: inline SIMD in leading_zeros/ones, add SSE2 chunk_eq backend, countdown loop

committed
34e9301

Commits on Jun 30, 2026

perf: aggressive leading_zeros SIMD with tiny-path unroll and thresholded alignment

committed
9281af9
perf: unroll unaligned SIMD loop 2× to close 4096-bit gap

committed
91bfd70
perf: 2×-unroll both aligned and unaligned AVX2 paths

committed
e80148e
refactor: add 2×-unrolled and aligned chunk_eq SIMD primitives

committed
48a5b22
refactor: extract leading_zeros SIMD into WordsScan trait layer

committed
1b8737c
fix: use total (not total-1) for SIMD iteration count in leading()

committed
303c265
perf: use unsafe pointer access in BitString tiny path

committed
4de7fdb
bench: expand trailing_zeros/ones benchmarks with bitvec_simd baselines

committed
a73cc62
bench: add BitStr leading_zeros/ones as trailing reference baselines

committed
270c353
perf: 2×-unrolled SIMD + raw intrinsics + rightmost fast path for trailing

committed
7806e91

Commits on Jul 1, 2026

refactor: group chunk_eq/leading/trailing under funcs_for_ends

committed
60e8048
docs: add SAFETY comments to leading.rs SIMD and pointer ops

committed
567683b
docs: add SAFETY comments to trailing.rs reverse SIMD scans

committed
4d9c845
docs: add SAFETY comments to BitString leading/trailing raw pointer ops

committed
1b56dd1
feat: runtime AVX2 dispatch for leading_zeros/ones

committed
55b555b
feat: runtime AVX2 dispatch for trailing_zeros/ones

committed
6f6fe41
perf: add first-word fast path to BitStr leading/trailing inner

committed
b2b26bd
feat: runtime AVX2 dispatch for words_arith binary ops (and/or/xor)

committed
8c9b7ea
feat: runtime AVX2 dispatch for not, shl, shr ops

committed
6333af7

Commits on Jul 2, 2026

feat: extend runtime SIMD dispatch to all backend files

committed
d786784

Commits on Jul 3, 2026

perf: inline count_leading_within in trailing scan, eliminate double shifts

committed
36a5984
test: add adversarial leading/trailing attack tests for BitString and BitStr

committed
e6814bd
docs: add missing SAFETY comments to all unsafe blocks

committed
bc12154
refactor: deduplicate ALIGN_THRESHOLD in leading.rs

committed
9addd88

Commits on Jul 4, 2026

refactor: clean up leading.rs SIMD dispatch, extract sse2/neon modules

committed
91cffea
refactor: clean up trailing.rs SIMD dispatch, extract sse2/neon modules

committed
80ac0ef
refactor: reorganize funcs_for_ends, inline SIMD backends, fix alignment

committed
25e9d0a
refactor: extract CPUID detection to crate-level cpuid module

committed
87d03fa
refactor: simplify cpuid to raw-pointer OnceCell, drop AtomicU8

committed
43fce00
refactor: use once_cell::sync::OnceCell for CPUID cache

committed
904636b
perf: relax trait impl inline for leading_value_bits

committed
945e010
perf: add BMI1 scalar path for BitString::leading_zeros

committed
8c668d2
build: add .cargo/config.toml.example for local max-performance benches

committed
f295020
chore: fix missing newline in .gitignore

committed
0ddc58d
test: add adversarial leading/trailing tests for long inputs, deep views, ones, invariants

committed
463be64
docs: rewrite README with quick start and per-category examples

committed
578a926