Characterizing the Challenges and Evaluating the Efficacy of a CUDA-to-OpenCL Translator