SIMD Vectors in the HotSpot JVM - Auto Vectorization and the Vector API

Explore the latest work on the HotSpot JVM, with a focus on auto-vectorization and the Vector API.

This session looks at recent improvements, including updates introduced in JDK 26, and explains what they mean for Java developers who care about performance. You’ll learn how modern JVM features can help unlock the full potential of today’s CPU vector instructions, and how to apply these capabilities in real-world Java code.

Auto Vectorization:

• High level explanation of SuperWord algorithm, how scalar operations are packed into vectors.

• Loop shapes: fill, copy, map and reduce.

• Extended topics: aliasing and alignment and their performance impact.

• Benchmarks on arrays and MemorySegments (project Panama).

Vector API:

• Why use the Vector API? When auto vectorization is not good enough yet. Quick intro to the API.

• Various examples, demonstrated with algorithms: compress/filter, shuffle, gather/scatter, simulating branches, etc. and benchmarks.

• Performance pitfalls: unavailable instructions, alignment, branch misprediction vs latency etc.

Recorded at Voxxed Days Zurich 2026.