https://hgpu.org/?p=14251
Evaluating the capabilities of the Xeon Phi platform in the context of software-only, thread-level speculation