GPU-CoDaSHep-Princeton

This repository provides tutorials, demos, and analysis scripts for GPU programming, tailored for the Princeton CoDAS-HEP course. Materials are organized into modules covering CUDA basics, unified memory, streaming, profiling, HEP physics usecases and neural network demos.

Presentations:

• Session 1 - https://1drv.ms/p/c/5a70ac10b7f66de0/EbfhsTCX3nZJjHrZLxds4tEB-xDZK00qdVukcO2Nc0b7Lg?e=3aYGxA

• Session 2 - https://1drv.ms/p/c/5a70ac10b7f66de0/ERfEQVlc9P1Avl1ZHmfzGu8ByfAcnfC_iya-4LsfrViYXw?e=89gSlM

• Session 3 - https://1drv.ms/p/c/5a70ac10b7f66de0/EaGidsT09EtFlgUtTcgQb4gBoWJmIyFjskZbDWFtlI0oUQ?e=78C8aH

Repository Structure

• 01/ – Introductory CUDA C notebooks and exercises

  • Presentations (AC_CUDA_C_*.pptx)
  • Main tutorial notebook: Session1.ipynb
  • Subfolders:
    • 01-hello/ to 09-heat/: Step-by-step CUDA examples (hello world, parallelism, indices, loops, memory allocation, error handling, vector addition, matrix multiplication, heat conduction)
    • edit/: Editable files for exercises

• 02/ – Unified Memory tutorials

  • Main notebook: Session2_advanced.ipynb
  • Subfolders: Vector addition, device properties, page faults, prefetching, SAXPY example, and editable files

• 03/ – Streaming and Visual Profiling

  • Main notebook: Streaming and Visual Profiling.ipynb
  • Subfolders: Vector addition, kernel initialization, prefetch checks, stream introduction, manual memory allocation, overlap transfer, n-body simulation, and editable files

• 04/ – Additional CUDA modules

  • Notebooks: lesson-5-project.ipynb, lesson-5-workbook.ipynb, Session3_Python-GPU_HEP.ipynb

• 05/ – Generator tutorials

  • Notebooks: generator_tutorial_gpu.ipynb, generator_tutorial_gpu_annotated.ipynb

• 06/ – HEP analysis and demos

  • Notebooks: gpu_dd4hep_tilecal.ipynb
  • Documentation: gpu_geant4_dd4hep_cuda_notebook.md
  • Tutorials: particle_physics_generators_tutorial.ipynb

• 07/ – Neural Network Demos and Analysis Scripts

  • Notebooks: cuda_neural_network_demo_complete.ipynb
  • Scripts: verify_notebook.py
  • Demos: gpu-demo/

• 08/ – Miscellaneous challenges and verification

  • Code: 01-nbody.cu
  • Notebook: GPU challenge.ipynb

Notebooks

Each module contains Jupyter notebooks (.ipynb) with explanations, code samples, and exercises. Topics include:

• CUDA programming basics
• Memory management (device, unified memory)
• Parallel algorithms (vector/matrix operations, heat conduction)
• Streaming and concurrency
• Profiling and performance analysis
• HEP event generation and analysis
• Neural network demos

Scripts

• 07/verify_notebook.py:
  Analyze notebook structure, extract code cells to files, and report statistics on content and coverage.

Credits

• The heat conduction CPU source code in 01/AC_CUDA_C.ipynb is credited to An OpenACC Example Code for a C-based heat conduction code from the University of Houston.

Getting Started

1. Clone the repository.
2. Open the notebooks in Jupyter or VS Code.
3. Follow the step-by-step exercises in each module.
4. Use the verification script to analyze or extract code from notebooks:

   python 08/verify_notebook.py <notebook_path>

License

Refer to individual files and notebooks for licensing and attribution information.

---

For more details, explore the subfolders and notebooks in each module.