https://hgpu.org/?p=10733
Understanding and Modeling the Synchronization Cost in the GPU Architecture