suprepupre/wow-optimize v2.1.0

v2.1.0 — Lock Optimization + File I/O Redirect + GlobalAlloc Fast Path

on GitHub

⚡ Three System-Level Optimizations

1. CriticalSection — Spin-First Lock Acquisition

WoW uses many short-held locks for internal synchronization (Lua state, sound, network buffers). When a lock is contended, EnterCriticalSection immediately enters a kernel wait — an expensive ~10-15 microsecond context switch.

Fix: New EnterCriticalSection hook:

1. TryEnterCriticalSection() — instant, no kernel transition
2. If fails: spin 32 iterations with _mm_pause (yield to SMT sibling)
3. If still fails: real EnterCriticalSection kernel wait

Most WoW locks are held for microseconds — the spin-first approach avoids kernel transitions for the vast majority of lock acquisitions.

2. SetFilePointer → SetFilePointerEx

WoW uses the legacy 32-bit SetFilePointer API with awkward dual-return error handling. Internally, Windows redirects this to SetFilePointerEx anyway.

Fix: Direct redirect to SetFilePointerEx — cleaner code path, one fewer API layer.

3. GlobalAlloc/GlobalFree — mimalloc for Fixed Allocations

WoW uses GlobalAlloc for clipboard operations and some legacy data paths. GMEM_FIXED allocations are plain memory blocks.

Fix: Route GMEM_FIXED allocations through mimalloc. GMEM_MOVEABLE stays on the original heap (required for GlobalLock/GlobalUnlock semantics).

Summary

All optimizations are pure Win32/runtime — zero addon/Lua impact.

Full Changelog

v2.0.9...v2.1.0