Parallel Computing for Data Science

Chapter 7

Thrust and Rth

As discussed in Section 6.2, GPU programming is difficult to do at true effi-

ciency, so it was recommended to make use of GPU code libraries whenever

possible. One such library is Thrust, to be presented in this chapter.

In turn, Thrust programming itself requires good facility with C++, espe-

cially some of the more esoteric features. An alternative is the Rth package,

which gives R programmers access to certain algorithms written in Thrust,

without those programmers needing to know Thrust/C/C++/CUDA; this

package too will be presented here.

Even more valuable is the fact that Thrust, and thus Rth, can be used not

only on GPU-equipped ...