Performance and Energy Prediction of OpenMP and CUDA Applications using Machine Learning and Pre-execution Features

This study investigates how machine learning techniques can predict the performance of CUDA applications on GPUs and the energy consumption of OpenMP applications on multi-core machines. Models such as ridge regression and random forest are applied using pre-execution data. The analysis, based on the Rodinia (GPU) and PolyBench (CPU) benchmarks, aims to understand how these ML techniques can predict the performance and energy efficiency of these hardware systems. The results indicate that task optimization can reduce energy consumption without affecting performance.