Implicit radix sorting on GPUs

In this chapter, we present a high performance sorting function on GPUs that is able to exploit the parallel processing power and memory bandwidth of modern GPUs to sort large quantities of data at a very high speed. We revisit the traditional radix sorting framework, analyze the weaknesses, and then propose a solution based on the implicit counting data presentation and its associated operations. We also improve the bandwidth utilization with our hybrid data structure and redefine the concept of arithmetic intensity as a guidance for GPU optimization process.