Just-In-Time Compilation for Java Performance: Recent and Ongoing Improvements
summary
JIT work between JDK 21 and JDK 26 delivers measurable wins: auto-vectorization fixes make reduction loops up to 4x faster, ARM64 sees platform gains in the 5–20% range, and C2 compile overhead drops ~5–15% by lighter GC interaction. The compiler now speculates on non-overlapping arrays to safely vectorize, adds intrinsics for crypto (including post-quantum) and array/string hot paths, and advances scalarization/value-class support (Valhalla) to remove heap allocations; generalized scalar replacement remains in progress.
AI-written summary · about
chapters (10)
- 0:00Introduction to JIT
- 1:41Tiered compilation overview
- 5:57Continuous optimization goals
- 10:41Hardware-level vectorization
- 13:37Vectorization and intrinsics
- 21:55Language evolution support
- 26:27Low-level efficiency and Valhalla
- 31:53Scalarization advancements
- 41:06Modern deployment models
- 44:24Conclusion and takeaways
related talks
51mJava Performance Update: From JDK 21 to JDK 25
41mUp and Away: JDK Optimizations
51mJava at Speed: Getting the Most Out of Modern Hardware
48mJava 9 Performance
35mJava at Speed: Building a Better JVM
53mJava Performance Update 2024
41mJava 24, Faster Than Ever
51mAccelerating performance of Java applications on Arm64