Recording of the course 2025-HY-HLRS (Jan 21-23, 2025 at HLRS)

The lectures (without the exercises) are recorded in the course 2025/HY-HLRS.

• Hunter's hardware architecture and its programming models
  Dr. Christian Simmendinger (HPE) and Igor Pasichnyk (AMD), and Johanna Potyka (AMD)
  HPE slides, AMD slides, recording
• Hybrid Programming in HPC - MPI+X
  Claudia Blaas-Schenner, Georg Hager, Rolf Rabenseifner
  slides / slides with animations
  • Introduction:
    slides 1-13 recording_{(the first 100 seconds are organizational stuff, so please skip them, I apologize for that)}
  • Programming Models and Optimizations:
    • MPI + OpenMP
      • General considerations /
        How to compile, link, and run
        slides 14-29 recording
      • Hands-on: MPI+OpenMP: he-hy - Hello Hybrid! - compiling, starting: see Exercises below
      • System topology, ccNUMA , and memory bandwidth /
      • Memory placement on ccNUMA systems
        slides 32-52 recording
      • Topology and affinity on multicore
        slides 53-67 recording_{(the first 57 seconds are quiet, so please skip them, I apologize for that)}
      • Hands-on: MPI+OpenMP: he-hy - Hello Hybrid! - pinning: see Exercises below
      • Case study: Simple 2D stencil smoother
        slides 69-77 recording
      • Overlapping communication and computation / 
        Communication overlap with OpenMP taskloops
        slides 88-100 recording
      • Hands-on: MPI+OpenMP: jacobi - hybrid through OpenMP parallelization: see Exercises below
      • Main advantages, disadvantages, conclusions
        slides 106-109 recording
    • MPI + Accelerator:
      slides 110-154 recording
    • MPI + MPI 3 shared memory
      • General considerations & uses cases /
        Re-cap: MPI_Comm_split & one sided communication /
        How to
        slides 156-183 recording
      • Hands-on: see lecture slides 184-195, MPI: data-rep - how to avoid replicated data: see Exercises below
      • MPI memory models & synchronization /
        Shared memory problems /
        Advantages & disadvantages, conclusions
        slides 197-214 recording
    • Optimized node to node communication
      slides 216-274 recording
  • Conclusions
    slides 275-282 recording