Performance Evaluation of an easily retargeted C compiler using the LLVM framework

When considering processor architectures (either existing ones or when developing new ones), native code for functional testing and performance evaluation will generally be required. In theory, the work load involved in developing such code can be alleviated by compiling existing test cases written in a higher level language. This thesis focuses on evaluating the feasibility of this approach by developing a basic C compiler using the LLVM framework and porting it to a number of architectures, finishing by comparing the performance of the compiled code with existing results obtained using the CoreMark benchmark. The resulting comparison can serve as a guideline when deciding which approach to choose when taking on a new architecture. The developed compiler and its back end ports can also serve as reference implementations. While not conclusive, the final results indicate that the approach is highly feasible for certain applications on certain architectures