Performance Evaluation of OpenMP's Target Construct on GPUs: Exploring Compiler Optimizations