Links

• Time Table for alpaka and OpenPMD Workshop on 23-25 October 2024

  • TimeTable

• Presentations

  • Introduction to Parallel Programming using alpaka
  • alpaka Features by 2D Heat Equation Solution
  • OpenPMD: Open Standard for Particle Mesh Data
  • PIConGPU:Introduction, Concepts and use of Libraries
  • PIConGPU:Designing an application

• Repository for hands-on exercises and presentations

  • All documents of Workshop Alpaka and OpenPMD Section, October 2024
  • Preparation and Alpaka Hands-On1: Connecting to LUMI and Installing Alpaka (Day1 Hands-On 1)
  • OpenMPD First Hands-On (Day2 Hands-On 9)

• Repositories:

  • https://github.com/alpaka-group/alpaka
  • https://github.com/openPMD
  • https://github.com/openPMD/openPMD-api/
  • https://github.com/ComputationalRadiationPhysics/picongpu

• alpaka Documentation:

  • Main Page: https://alpaka.readthedocs.io/en/latest/index.html
  • Installation Guide
  • Cheat Sheet
  • CMake Variables
  • API Docs

• OpenPMD Documentation

  • OpenPMD Landing Page
  • Main Documentation
  • First Write
  • First Read
  • OpenPMD Projects
  • API Documentation

• Others

  • Webinar May 2024: alpaka concepts and usage (pdf)

The outline of the alpaka and openPMD hands-ons

Section I: Introduction to Alpaka

1. Introduction: What is alpaka, where it is used?

• Support for heterogeneous architectures (e.g., CPUs, GPUs, FPGAs).
• Write once, run anywhere—alpaka abstracts hardware specifics for parallel computing.

2. Hands-on 1: Installing alpaka and running an example (LUMI)

• Install alpaka with correct cmake backend options
• Compile and run an alpaka example from the repository.
• Verify it runs on the available hardware (CPU, GPU, etc.).

3. Parallel programming concepts and portable parallelization by alpaka

• Grid Structure and WorkDivision
• Data Parallelism
• Indexing

4. Hands-on 2: HelloIndex kernel which prints indexes

Section II: Alpaka Features in Action - 2D Heat Equation

1. Memory management for 1D and 2D data

• Memory Allocation, Padding and Pitch

2. Filling buffers in parallel

• Use indexing to match thread to data

3. Hands-on 3: Kernel to fill initial conditions of heat equation

4. Heat Equation

• Double accelerator-buffers method to solve Heat Equation

5. Preparing stencil kernel

• Difference Equation as a stencil operation

6. Hands-on 4: Heat Equation stencil kernel

7. alpaka programming features and data-structures

• Accelerator, Device, Queue, Task
• Buffers and Views: Managing memory across devices
• alpaka mdspan
• Setting work division manually

8. Usability and optimization

• Hands-on 5: Using alpaka mdspan for easier indexing
• Hands-on 6: Domain Decomposition, chunking or tiling
• Hands-on 7: Using multiple Queues to increase performance. Explore overlap between computation and data transfer
• Hands-on 8: Using Shared Memory for chunksExplore overlap between computation and data transfer

Timing Runs

• Measure the performance of your kernels and analyze the timing with and without shared memory

Section III: OpenPMD

This session will introduce the participants to the scientific metadata format openPMD. The practical sessions and exercises will include the basic modeling of scientific data via the openPMD-api, possibilities for visualizing openPMD data, streaming I/O workflows, data compression, parallel I/O and more.

9. Hands-On 9: OpenPMD Basic Object Model

10. Hands-On 10: Metadata

11. Hands-On 11: OpenPMD Visualisation and Json-Toml

12. Hands-On 12: Streaming IO

13. Hands-On 13: OpenPMD Streaming IO Solution

14. Hands-On 14: OpenPMD Viewer Compression

15. Hands-On 15: Span API

16. Hands-On 16: Span API Solution

17. Hands-On 17: Constant Components Python Example

18. Hands-On 18: Parallel Python Example

19. Hands-On 19: Parallel Python Example Solution