How to program AMD GPU devices with HIP: from writing new applications to porting existing CUDA code
TimeTuesday, 10 November 20202:30pm - 6:30pm
DescriptionThe goals of this tutorial are to: 1) cover essential AMD GPU programming principles; and 2) demonstrate how to write portable GPU programs, enabling compatibility with other GPU devices. Existing GPU programming experience is not necessary but familiarity with C/C++ is helpful. Attendees will learn about AMD's Compute DNA (CDNA) microarchitecture; how to transfer data to and from the GPU; and how to write HIP kernels that target CDNA. Attendees will then learn how to launch those kernels over multiple blocks of GPU threads. Hands-on exercises will be provided. During the tutorial attendees will have the opportunity to run codes on AMD GPU cloud instances. The tutorial will also cover available open-source tools, compilers and libraries, so attendees can utilize what they have learned and practice developing their own codes. Finally, we address how existing GPU codes written in CUDA can be ported to enable support for AMD's graphics cards.