Адаптація JAVA-застосунків під гібридні процесорні архітектури

Abstract

The paper addresses the problem of adapting Java applications to hybrid processor architectures that combine CPU, GPU and NPU. It describes the specific features of working with hybrid chips, the challenges developers face, and the prospects of using such architectures in modern software solutions. Significant attention is given to analyzing tools that help optimize Java code for interaction with GPUs and NPUs, particularly Aparapi, TensorFlow Java API, ONNX Runtime, Project Panama and GraalVM. The study emphasizes the efficient distribution of tasks among computational components, which enables substantial performance improvements and reduces the energy consumption of software.