In this chapter, we have covered the CUDA programming methods using CUDA libraries, and other compatible languages. We have also covered the basic use of cuBLAS and its mixed-precision operation feature. Also, we explored the cuRAND, cuFFT, NPP, and OpenCV libraries. Thanks to these libraries, we could implement GPU applications with little effort, as discussed at the beginning of the chapter.

We have implemented some GPU applications using other languages that are compatible with CUDA. Firstly, we covered several Python packages, which enable Python and CUDA interops. They provide Pythonic programmabilities and compatibilities with other Python features. Then, we covered CUDA accelerations in other scientific computing languages, such as Octave, R, and MATLAB.

Now, we have one more GPU programming method to cover—OpenACC. With this we can covert the original C/C++ and Fortran host codes to work on GPUs using directives such as #pragma acc kernels. We will cover this in the next chapter.