Exploiting More Parallelism from Applications Having Generalized Reductions on GPU Architectures