TornadoVM Continues Adapting Java OpenJDK/GraalVM For Heterogeneous Hardware

A new release of TornadoVM is now available, the open-source plug-in to OpenJDK and GraalVM to allow for Java code to run on heterogeneous hardware with ease -- including various GPU models as well as FPGAs.

The TornadoVM plug-in allows for Java code to taeget OpenCL, NVIDIA PTX, and SPIR-V compatible devices for easily extending the scope beyond just traditional CPUs. Particularly with OpenCL and Khronos SPIR-V format that is used by Vulkan too, it opens up GraalVM/OpenJDK Java execution on many graphics processors. TornadoVM exposes task-level, data-level, and pipeline-level parallelism to developers via its own API.

TornadoVM sample code of its Loop Parallel API.

Building off January's release of TornadoVM 0.15 is now the v0.15.1 point release that brings a few notable improvements. TornadoVM 0.15.1 brings a device selection heuristic based on the computing capabilities of devices. TornadoVM can now automatically select the default device to use for execution based on the fastest device based on its computing capability.

The TornadoVM 0.15.1 release also has an optimization for eliminating redundant data copies for read-only and write-only buffers between the CPU and GPU/device based on the Tornado Data Flow Graph. There is also a new installer script with this TornadoVM release. Rounding out this release is improved support for Apple M1 and M2 hardware with the OpenCL back-end.

Downloads and more details on the TornadoVM 0.15.1 release via GitHub.