Kirk and Hwu, Chapter 6: §6.1. Re-read, including discussion on summation, p.158-162.
“Benchmarking GPUs to tune dense linear algebra,” by V. Volkov and J. Demmel. Proc. 2008 ACM/IEEE Conf. on Supercomputing, Austin, TX, Nov. 15 - 21, 2008. PDF. Read through §4.
To explore latency and tradeffs involving occupancy, see Vasily Vokov’s GTC presentation Better performance at Lower Occupancy