Advanced Concepts in CUDA

In the last chapter, we looked at memory architecture in CUDA and saw how it can be used efficiently to accelerate applications. Up until now, we have not seen a method to measure the performance of CUDA programs. In this chapter, we will discuss how we can do that using CUDA events. The Nvidia Visual Profiler will also be discussed, as well as how to resolve errors in CUDA programs from within the CUDA code and using debugging tools. How we can improve the performance of CUDA programs will also be discussed. This chapter will describe how CUDA streams can be used for multitasking and how we can use them to accelerate applications. You will also learn how array-sorting algorithms can be accelerated using CUDA. Image processing is an application where we need to process a large amount of data in a very small amount of time, so CUDA can be an ideal choice...