San Francisco: CUDA is a Low-Level Language

In November, Justin Lebar of Google will talk about how to program your GPU using Clang. He recently talked about GPUs, CUDA, and Clang at CppCon 2016 (https://cppcon2016.sched.org/event/7nM5/bringing-clang-and-c-to-gpus-an-open-source-cuda-compatible-gpu-c-compiler).

In this talk, Justin will:

• Provide a model of how a modern GPGPU works.

• Write some CUDA code, informed by our understanding of the hardware we're targeting. Based on this I want to argue that CUDA is a low-level language, and moreover that you probably want to write your GPU programs in such a language.

• Talk about why I am excited that Clang/LLVM can now compile CUDA code.

This talk won't assume any familiarity with GPUs or CUDA.

Justin Lebar leads Google's GPU compiler team.

Level of Expertise: Intermediate to Advanced