Lost in Translation: Challenges in Automating CUDA-to-OpenCL Translation