https://hgpu.org/?p=26677
Improving performance of SYCL applications on CPU architectures using LLVM-directed compilation flow