This course gives an overview of the most relevant GPGPU computing techniques to accelerate computationally demanding tasks on HPC heterogeneous architectures based on GPUs.
The course will start with an architectural overview of modern GPU based heterogeneous architectures, focusing on its computing power versus data movement needs. The course will cover both a high level (pragma-based) programming approach with OpenACC for a fast porting startup, and lower level approaches based on nVIDIA CUDA and OpenCL programming languages for finer grained computational intensive tasks. A particular attention will be given on performance tuning and techniques to overcome common data movement bottlenecks and patterns.
The students will learn the strengths and weaknesses of GPUs as accelerators, the problem of data movement between host and device memories and how to overcome bottlenecks through concurrent computing/data-movement operations. The students will learn the basics of GPGPU programming using both higher and lower level programming approaches.
Overview of architectural trends of GPUs in HPC. GPGPU parallel programming in heterogeneous architectures. Basis of OpenACC, CUDA and OpenCL programming.
Researchers and programmers interested in porting scientific applications or use efficient post-process and data-analysis techniques in modern heterogeneous HPC architectures.
A basic knowledge of C or Fortran is mandatory. Programming and Linux or Unix. A basic knowledge of any parallel programming technique/paradigm is recommended.