Throughout the project, the Vector API balanced providing a cross-platform API with staying close to the metal, performance-wise.

While the project succeeded in that goal, the Vector API had to give up some important functionality when it didn’t fit the design. As a result, the API became unsuitable for implementing an important class of vectorized algorithms that depend on access to particular hardware functionality. With recent progress in Project Panama (on the Foreign Function & Memory API and jextract), new opportunities have emerged to bring Java even closer to hardware.

The talk covers how the Vector API evolved (from machine code snippets to JVM intrinsics), and then focus on a novel approach to providing access to individual machine code instructions from Java code in a performant manner. Such a “hardware intrinsics” API complements the Vector API by providing access to low-level platform-specific primitives. It opens up new opportunities for the Java Platform and enables simplifications in the Vector API, JDK, and JVM implementations.

Make sure to check the JVMLS 2025 playlist.