simCUDA: A C++ based CUDA simulation framework

The primary objective of this thesis is to develop a CUDA simulation framework (simCUDA) that effectively maps the existing application written in CUDA to be executed on top of standard multi-core CPU architectures. This is done by specifically annotating the application at the source level itself, and making the relevant changes required for the application to run in a similar and functionally equivalent manner on a multi-core CPU as it would run in a CUDA-supported GPU. The simulation framework has been developed using C++11 threads, which provides an abstraction for a thread of execution, as well as several classes and class templates for mutexes, condition variables, and locks, to be used for their management. As an extension to the simulation framework, the basic block sequence of execution on a per thread basis is also computed for analysis. This information can in turn be used to derive the basic block sequence of execution on a per warp basis, and thus emulate and replicate real-world behavior of a GPU.