uxlfoundation/oneDAL

What it solves

It provides high-performance, accelerated machine learning routines for tabular data, reducing the computation time required for common data science tasks like linear regression, K-means clustering, and random forests.

How it works

Written in C++ and DPC++, the library leverages hardware-specific optimizations. It uses SIMD instructions and cache structure exploitation for CPUs, and the SYCL framework and oneMKL library for GPU acceleration. It also supports multi-node distributed setups for scaling performance across multiple machines.

Who it’s for

Data scientists and software engineers who need to accelerate their ML workflows, specifically those using C++ or Python (via the scikit-learn extension).

Highlights

• Accelerated routines for tabular data (e.g., linear regression, K-means, random forests).
• Cross-platform hardware acceleration for CPUs and GPUs.
• Support for multi-node distributed computation for strong and weak scaling.
• Integration with scikit-learn through the scikit-learn-intelex extension.
• Performance boosts for Apache Spark MLlib via the OAP MLlib project.