Heterogenous Accelerator Toolkit (HAT) is a novel parallel programming framework designed to accelerate Java parallel workloads on heterogeneous hardware, including Graphics Processing Units (GPUs). HAT leverages the code-reflection APIs introduced in the OpenJDK project Babylon to tackle code transformation that helps developers translate sections of the Java programs to foreign programming languages, such as CUDA and OpenCL. HAT also provides a runtime system to automatically manage the generated code, data handling, and data migration automatically.
In this talk, we'll discuss the core components of the HAT API and show using examples how Java developers can offload data-parallel workloads to GPUs to achieve significant performance improvements.
Type: Learning Session (50 min)
Track: Machine Learning and Artificial Intelligence
Audience Level: Expert
Speaker: Juan Fumero Alfonso