IAS Seminar "Same old codes, new performance portability" (hybrid)

Speaker:

Dr. Alexandru Calotoiu, Staff of Professorship for Scalable Parallel Computing, ETH Zürich

Abstract:

Most scientific applications are written in Fortran or C. Fortran codes are highly tuned towards CPU performance but struggle to achieve the same performance on GPUs despite their increasing adoption in supercomputers all over the world. Modifying C or Fortran programs to obtain good performance on GPUs is challenging and time consuming. We discuss how to combine data and control-centric optimizations to better support scientific applications in HPC. We automatically translate Fortran, C, or Python to a flexible, data-centric representation, preserving and exposing semantic information. We leverage this information, along with data-centric transformations to express both functional and data parallelism and allow efficient code to be generated for both CPUs and GPUs. Using a representative code from an operational weather forecasting model as a case study,  we create a perfectly data-parallel program representation and generate CPU and GPU implementations with runtimes outperforming both serial and OpenMP parallel CPU implementations as well as being on par with the best manually written versions on GPU, without changing the original Fortran code.

Join Zoom Meeting
https://fz-juelich-de.zoom.us/j/61651944750?pwd=JBbQ3bhoXqHWAaS6bUVVBHhzl4H0kJ.1
Meeting ID: 616 5194 4750 | Passcode: 415816