Heterogeneous Accelerator Toolkit (HAT) Update

Gary Frost - Architect (Java Platform Group - Oracle) and Paul Sandoz - Architect (Java Platform Group - Oracle), recorded during the JVM Language Summit (August 2024 - Santa Clara, CA).
Last year at JVMLS, we outlined plans for a Heterogeneous Accelerator Toolkit Project (HAT) as part of the Babylon project. HAT would focus on offering Java access to GPU's for acceleration of appropriate workloads. We described how HAT would leverage Babylon's Code reflection capability to model the compute and kernel code callgraphs, how it would use Panama's Foreign Function and Memory (FFM) to it allow to interact with vendor specific native code coded as HAT 'backends'. This year we will provide an update on HAT, focusing on the programming model and how Babylon, Panama FFM and the new ClassFile API were used to streamline GPU acceleration from Java. We will also show some demos, talk about community code contributions for upcoming backend implementations and highlight challenges and opportunities for the project going forward.
Make sure to check the • JVM Language Summit 2024 playlist.
Resources
• HAT ➤ github.com/ope...
• Project Babylon ➤ github.com/ope...
• JVLMS 2023 ➤ • Java and GPU … are we ...
• JVLMS 2024 ➤ • Project Babylon - Code...
• SIMD〈 SIMT〈 SMT: Parallelism in NVIDIA GPUs ➤ yosefk.com/blo...
• More on Inside Java ➤ inside.java/ta...
Tags #Java #ProjectBabylon #GPU #OpenJDK

[@MarioJunior258]

This talk made me want to get involved. Very very good stuff

[@shadeblackwolf1508]

This reminds me of the graph they showed for valhalla. this project seems to be at the slope where it works, and complexity goes up to match more usecases. Looking forward to see the complexity coming down on the other end once the usecases are all met, and elegance becomes a main goal.