LLVM Social Berlin

In May, Bruno Caballero will show us how to target hardware accelerators from Java with MLIR based on his fork of the OpenJDK Project Babylon.

Bruno leads the Java Virtual Machines Engineering team at MicroDoc, which develops JVMs and maintains ports across various processors and operating systems.

In this talk, we’ll explore a toy project that translates Java code into MLIR for the TOSA dialect, which is used for AI and accelerator workloads. We’ll briefly introduce current approaches (e.g., TornadoVM and Babylon/HAT), before looking at how this project reuses existing compiler infrastructure instead of building new backends. Finally, we’ll touch on whether DSLs should be embedded in general-purpose languages like Java or designed as standalone languages.

Small snacks and drinks will be available. There is always space for outside topics and discussions. If you have specific questions, please bring them. Looking forward to meeting everyone again!