Description

This course introduces the fundamentals of programming for graphics processors (GPUs). The architecture of GPUs and specificities of their programming (languages, constraints, optimization) will be illuminated under various axes of application with an emphasis on particular on deep learning. Only the pipeline compute (CUDA, OpenCL, intensive computing) will be studied in detail.

Course Objectives

• Assimilate the standard architecture of graphics processors.
• Assimilate the general principles of their programming.
• Practical introduction to parallel programming on the GPU for high-performance computing.
• Experience the fundamental operators of deep learning at the Through simple examples.

Course

Plan:

1. Details of the module.
2. Introduction: GPUs Architecture, GPUs Programming
3. GPU Memory Architecture
4. Optimized Matrix Multiplication
5. Introduction to CuBLAS

Practical Assignments

• page

Toolbox

Here are various tools that may be useful in this module :

• The CUDA development environment.
• The GNU GCC compiler.
• GNU Emacs and its CUDA mode.
• cuBLAS
• Thurst

Resources

• Ressources CUDA
• Ressources OpenCL