https://hgpu.org/?p=23282
Performance portability through machine learning guided kernel selection in SYCL libraries