Add UVM Support to current GPGPU-Sim

[yechen3]

Basically merge the implementation of functional and timing simulation support for Unified Virtual Memory from DebashisGanguly's UVMSmart to current version of GPGPU-Sim. In order to improve the simulation performance, an option "skip_cycles_enable" is provided to skip simulation of shader cores when all warps are either waiting for barrier or far memory transaction. It has been tested on rodinia-2.0 benchmark suites, and the result shows that adding UVM support has very tiny impact on simulation time.