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