https://hgpu.org/?p=11965
Address Selection for Efficient Barriers on the Intel Xeon Phi