GPU Programming

E. Carlinet

J. Chazalon

April 2022

Moodle page

Access the Moodle page here to take quizzes, select your project group and submit assignments.

Lectures

Architectures des GPUs
Programmation des GPUs avec CUDA
- Part 0 - Forewords: Slides Handout
- Part 1 - CUDA overview: Slides Handout
- Part 2 - Host view of GPU computations: Slides Handout
- Part 3 - Kernel programming: Slides Handout
- Part 4 - Compilation and runtimes: Slides Handout
Patterns de programmations efficaces: considération mémoire
- Part 1 - The global memory Slides Handout
- Part 2 - The shared memory Slides Handout
Patterns de programmations efficaces: Histogrammes & Réductions
- Part 1 - Reductions Slides
- Part 2 - Histograms Slides
- Part 3 - Scans Slides

TP

Project

Spring-2022: Slides

How to use OpenStack GPUs

Make sure you have your SSH key registered on https://cri.epita.fr/users/{EPITA_LOGIN}/ssh-keys/.

Connect to any available machine (pick one at random, your ~/afs/ directory will be shared):

ssh -X -l $USER -p 22000 gpgpu.image.lrde.iaas.epita.fr
ssh -X -l $USER -p 22001 gpgpu.image.lrde.iaas.epita.fr
ssh -X -l $USER -p 22002 gpgpu.image.lrde.iaas.epita.fr
ssh -X -l $USER -p 22003 gpgpu.image.lrde.iaas.epita.fr

Activate your AFS:

kinit  # Then enter you CRI password
aklog
$AFS_DIR/.confs/install.sh

Activate the development environment featuring all CUDA tools:

wget https://www.lrde.epita.fr/~carlinet/cours/GPGPU/shell.nix
nix-shell shell.nix

Then use NVidia tools directly to compile, debug and profile your programs:
```
nvcc -o device_query -g -G device_query.cu
```