- separate i4/r8 code (core functions optimized by v8)
- prefer faster methods when convenient
- node-based performance tests