Efficient implementation of GPGPU synchronization primitives on CPUs