Accelerating and Scaling
Python for HPC

Python is powering breakthrough exascale scientific discoveries – in this interactive tutorial attendees learned how to write, debug, profile, and optimize high-performance, multi-node GPU applications in Python: CuPy for drop-in GPU acceleration of NumPy workflows, Numba for writing custom kernels that match the performance of C++ and Fortran, and mpi4py for scaling across thousands of nodes. Along the way we profiled our code, debugged tricky kernels, and leveraged foundational and domain-specific accelerated libraries. Everything was hands-on: short and interactive lectures by expert instructors were paired with guided Jupyter notebooks that introduced each concept and had attendees immediately apply it in bite-sized exercises (shallow water equation, SVD image compression, Mandelbrot, etc).

Materials

Slides and notebooks from the session will be archived here.

If you are interested in our work you can join our compiler-research-announce google groups forum or follow us on LinkedIn.