HHVM Performance Status
summary
Claims roughly 2x VM speedup year-over-year, achieved by a new JIT plus a region compiler prototype, memory-layout work (packed and mixed arrays, flatter strings) and whole-binary reordering driven by Linux perf. A small script implementing birs heuristic yields a rare 3–4% step improvement, and combined array/layout changes cut cache misses dramatically so the CPU becomes more often CPU‑bound. Ongoing work: region compilation, whole‑program optimization, better GC and cluster‑wide profiling.
AI-written summary · about
chapters (10)
- 0:00Start
- 3:21What do we do? Optimize! Build things measure
- 10:06PHP Workload
- 15:55Wall Clock Time
- 16:20Machine Sympathy
- 19:11Arrays
- 22:02Tracelet Compiler
- 23:15Region Compiler
- 25:052013 in Review Doubled CPU Efficiency New JIT Compiler
- 28:25Aspirations Shrink Data Structures
