Unlocking Python Multithreading Capabilities using OpenMP-Based Programming with OMP4Py

Python exhibits inferior performance relative to traditional high performance computing (HPC) languages such as C, C++, and Fortran. This performance gap is largely due to Python’s interpreted nature and the Global Interpreter Lock (GIL), which restricts multithreading efficiency. However, the introduction of a GIL-free variant in the Python interpreter opens the door to more effective exploitation of multithreading parallelism in Python. Based on this important new feature, we introduce OMP4Py with the aim of bringing OpenMP’s familiar directive-based parallelization paradigm to Python. Its dual-runtime architecture design combines the benefits of a pure Python implementation with the performance and low-level capabilities required to maximize efficiency in compute-intensive tasks. In this way, OMP4Py offers both full Python support and the high performance required by HPC workloads.