Machine-Learning-Driven Runtime Optimization of BLAS Level 3 on Modern Multi-Core Systems
This work addresses performance optimization for scientific computing on HPC platforms, but it is incremental as it builds on the existing ADSALA library.
The paper tackles the challenge of optimizing the number of threads for BLAS Level 3 operations on modern multi-core systems by extending the ADSALA library with machine learning to predict the best thread count based on matrix dimensions and system architecture, achieving speedups of 1.5 to 3.0 compared to using the maximum number of threads.
BLAS Level 3 operations are essential for scientific computing, but finding the optimal number of threads for multi-threaded implementations on modern multi-core systems is challenging. We present an extension to the Architecture and Data-Structure Aware Linear Algebra (ADSALA) library that uses machine learning to optimize the runtime of all BLAS Level 3 operations. Our method predicts the best number of threads for each operation based on the matrix dimensions and the system architecture. We test our method on two HPC platforms with Intel and AMD processors, using MKL and BLIS as baseline BLAS implementations. We achieve speedups of 1.5 to 3.0 for all operations, compared to using the maximum number of threads. We also analyze the runtime patterns of different BLAS operations and explain the sources of speedup. Our work shows the effectiveness and generality of the ADSALA approach for optimizing BLAS routines on modern multi-core systems.