coregx · kolkov · Mar 25, 2026 · Mar 25, 2026
@@ -12,6 +12,38 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - ARM NEON SIMD support (Go 1.26 `simd/archsimd` intrinsics — [#120](https://github.com/coregx/coregex/issues/120))
 - SIMD prefilter for CompositeSequenceDFA (#83)
 
+## [0.12.20] - 2026-03-25
+
+### Performance
+- **Premultiplied State IDs** — StateID stores byte offset into flat transition table,
+  eliminating multiply from DFA hot loop. Single `flatTrans[sid+classIdx]` lookup.
+  Inspired by Rust `LazyStateID` (hybrid/id.rs).
+
+- **Tagged State IDs** — match/dead/invalid/start flags encoded in StateID high bits.
+  Single `IsTagged()` branch replaces 3 separate comparisons in DFA hot loop.
+  4x loop unrolling breaks to slow path only on tagged states.
+
+- **1-byte match delay** (Rust determinize approach) — match reporting delayed by 1 byte,
+  enabling correct look-around assertion resolution (^, $, \b) at match boundaries.
+  Reference: Rust `determinize` mod.rs:254-286.
+
+- **Rust-aligned DFA determinize: break-at-match** — replaced `filterStatesAfterMatch`
+  with Rust's `determinize::next` break-at-match semantics (mod.rs:284). Epsilon closure
+  uses add-on-pop DFS with reverse Split push, matching Rust sparse set insertion order.
+  Incremental per-target epsilon closure preserves correct state ordering for leftmost-first.
+  **Eliminates Phase 3** anchored re-scan: bidirectional DFA reduced from 3-pass to 2-pass.
+  Verified against Rust regex-automata `find_fwd` — identical results on all test patterns.
+
+- **Memmem: Memchr(rareByte) + verify** (Rust approach) — replaced `MemchrPair`-based
+  paired search in `simd.Memmem` with single rare byte Memchr scan + `bytes.Equal`
+  verify, matching Rust `memchr::memmem` architecture.
+
+### Benchmarks (LangArena LogParser, 7.2 MB, 13 patterns)
+
+| vs stdlib | vs Rust | Wins vs Rust |
+|-----------|---------|-------------|
+| **30x faster** total | 2-5x gap (local i7) | ip 18.5x, multiline_php 2.0x, char_class 1.3x |
+
 ## [0.12.19] - 2026-03-24
 
 ### Performance

@@ -64,19 +64,19 @@ Cross-language benchmarks on 6MB input, AMD EPYC ([source](https://github.com/ko
 
 | Pattern | Go stdlib | coregex | Rust regex | vs stdlib | vs Rust |
 |---------|-----------|---------|------------|-----------|---------|
-| Literal alternation | 475 ms | 4.4 ms | 0.7 ms | **109x** | 6.3x slower |
-| Multi-literal | 1391 ms | 12.6 ms | 4.7 ms | **110x** | 2.6x slower |
-| Inner `.*keyword.*` | 231 ms | 0.29 ms | 0.29 ms | **797x** | **~parity** |
-| Suffix `.*\.txt` | 234 ms | 1.83 ms | 1.07 ms | **128x** | 1.7x slower |
-| Multiline `(?m)^/.*\.php` | 103 ms | 0.66 ms | 0.66 ms | **156x** | **~parity** |
-| Email validation | 261 ms | 0.54 ms | 0.31 ms | **482x** | 1.7x slower |
-| URL extraction | 262 ms | 0.84 ms | 0.35 ms | **311x** | 2.4x slower |
-| IP address | 498 ms | 2.1 ms | 12.0 ms | **237x** | **5.6x faster** |
-| Char class `[\w]+` | 554 ms | 48.0 ms | 50.1 ms | **11x** | **1.0x faster** |
-| Word repeat `(\w{2,8})+` | 641 ms | 185 ms | 48.7 ms | **3x** | 3.7x slower |
+| Literal alternation | 466 ms | 4.2 ms | 0.65 ms | **110x** | 6.4x slower |
+| Multi-literal | 1391 ms | 12.4 ms | 5.3 ms | **112x** | 2.3x slower |
+| Inner `.*keyword.*` | 227 ms | 0.34 ms | 0.32 ms | **668x** | **~parity** |
+| Suffix `.*\.txt` | 228 ms | 2.9 ms | 1.3 ms | **78x** | 2.3x slower |
+| Multiline `(?m)^/.*\.php` | 101 ms | 0.35 ms | 0.72 ms | **288x** | **2.0x faster** |
+| Email validation | 258 ms | 0.51 ms | 0.27 ms | **506x** | 1.8x slower |
+| URL extraction | 259 ms | 0.71 ms | 0.35 ms | **364x** | 2.0x slower |
+| IP address | 493 ms | 0.73 ms | 13.5 ms | **675x** | **18.5x faster** |
+| Char class `[\w]+` | 483 ms | 40.9 ms | 56.0 ms | **11x** | **1.3x faster** |
+| Word repeat `(\w{2,8})+` | 628 ms | 167 ms | 54.8 ms | **3x** | 3.0x slower |
 
 **Where coregex excels:**
-- Multiline patterns (`(?m)^/.*\.php`) — near Rust parity, 100x+ vs stdlib
+- Multiline patterns (`(?m)^/.*\.php`) — **2x faster than Rust**, 288x vs stdlib
 - IP/phone patterns (`\d+\.\d+\.\d+\.\d+`) — SIMD digit prefilter skips non-digit regions
 - Suffix patterns (`.*\.log`, `.*\.txt`) — reverse search optimization (1000x+)
 - Inner literals (`.*error.*`, `.*@example\.com`) — bidirectional DFA (900x+)

@@ -2,7 +2,7 @@
 
 > **Strategic Focus**: Production-grade regex engine with RE2/rust-regex level optimizations
 
-**Last Updated**: 2026-03-24 | **Current Version**: v0.12.18 | **Target**: v1.0.0 stable
+**Last Updated**: 2026-03-25 | **Current Version**: v0.12.19 | **Target**: v1.0.0 stable
 
 ---
 
@@ -12,7 +12,7 @@ Build a **production-ready, high-performance regex engine** for Go that matches
 
 ### Current State vs Target
 
-| Metric | Current (v0.12.15) | Target (v1.0.0) |
+| Metric | Current (v0.12.19) | Target (v1.0.0) |
 |--------|-------------------|-----------------|
 | Inner literal speedup | **280-3154x** | ✅ Achieved |
 | Case-insensitive speedup | **263x** | ✅ Achieved |
@@ -93,7 +93,12 @@ v0.12.16 ✅ → WrapLineAnchor for (?m)^ patterns
          ↓
 v0.12.17 ✅ → Fix LogParser ARM64 regression, restore DFA/Teddy for (?m)^
          ↓
-v0.12.18 (Current) ✅ → Flat DFA transition table, integrated prefilter, PikeVM skip-ahead
+v0.12.18 ✅ → Flat DFA transition table, integrated prefilter, PikeVM skip-ahead
+         ↓
+v0.12.19 ✅ → Zero-alloc FindSubmatch, byte-based DFA cache, Rust-aligned visited limits
+         ↓
+v0.12.20 (Current) → Premultiplied/tagged StateIDs, break-at-match DFA determinize,
+                      Phase 3 elimination (2-pass bidirectional DFA)
          ↓
 v1.0.0-rc → Feature freeze, API locked
          ↓

@@ -98,18 +98,20 @@ func TestDetectAccelerationFromCached(t *testing.T) {
 
 func TestDetectAccelerationFromFlat(t *testing.T) {
 	// Test acceleration detection via flat transition table
+	// Using premultiplied state IDs: sid = stateIndex * stride
 	stride := 256
-	sid := StateID(1)
-	flatTrans := make([]StateID, 2*stride) // 2 states
+	sid := StateID(1 * stride) // premultiplied: state 1 at offset 256
+	state2 := StateID(2 * stride)
+	flatTrans := make([]StateID, 3*stride) // 3 states (0, 1, 2)
 
 	// State 1: 250 self-loops, 3 exits to state 2, 3 dead
-	base := int(sid) * stride
+	base := sid.Offset()
 	for i := 0; i < 250; i++ {
 		flatTrans[base+i] = sid // Self-loop
 	}
-	flatTrans[base+250] = StateID(2)
-	flatTrans[base+251] = StateID(2)
-	flatTrans[base+252] = StateID(2)
+	flatTrans[base+250] = state2
+	flatTrans[base+251] = state2
+	flatTrans[base+252] = state2
 	flatTrans[base+253] = DeadState
 	flatTrans[base+254] = DeadState
 	flatTrans[base+255] = DeadState

@@ -525,7 +525,7 @@ func TestFindWithPrefilterAtWordBoundary(t *testing.T) {
 // TestFindWithPrefilterAtCacheClear tests the cache-clear recovery path
 // in findWithPrefilterAt using a very small cache.
 func TestFindWithPrefilterAtCacheClear(t *testing.T) {
-	config := DefaultConfig().WithMaxStates(3).WithMaxCacheClears(10)
+	config := DefaultConfig().WithMaxStates(6).WithMaxCacheClears(20)
 	compiler := nfa.NewDefaultCompiler()
 	nfaObj, err := compiler.Compile("[a-zA-Z]+[0-9]+")
 	if err != nil {