Course material

Lectures are available as reveal.js presentations. In case you prefer pdf-files, you can print the presentation to pdf-file in your browser (works with Chromium and Falkon, but does not seem to function correctly with Firefox). Click one of the links to acces them. The material will be continuously updated as the lecture advances. Slides of the previous course in 2017 are also available.

Here is a rough overview of how I have schedules various topics and how they align with the assignments’ deadlines. Notes that actual dates might be shifted compared to the planned ones.

Introduction

These lectures

• give an overview of the course’s content,
• motivate high performance computation in general,
• hightly the aspect of choice in hard- and software that comes with designing a high performance application.

To get an impression of HPC news you can consult, for example, insideHPC or HPCwire. Both are focused on the USA. The HPC policy of the European Commission is based on the EuroHPC declaration. An overview by countries is presented on Wikipedia.

Development Environment

These lectures

• given an introduction to working with the terminal multiplexer,
• building applications using make,
• the compiler gcc and the linker ln.

Hardware Architecture

These lectures

• give a simple hardware model to work with,
• contain a discussion fo latency of cache, memory, and storage,
• contain a brief discussion of common CPU features,
• introduce assembly code.

I recommend to have a look at conference talks by Scott Meyer and Mike Acton on hardware and programm performance.

Optimization

These lectures

• connect the lecture on hardware to actual implementations,
• summarize typical means to improve program performance,
• survey some tools available to inspect performance.

Parallel Computing

These lectures

• contain coarse treatment of network and memory layout in parallel systems,
• give bounds for efficiency of parallelization,
• introduce the concept of load balancing,
• give an overview of POSIX threads.

OpenMP

These lectures introduce the parallization framework OpenMP, handeling

• parallelization of loops and tasks,
• synchronization of concurrent operations,
• off-loading to accelerators.

OpenCL

These lectures introduce the framework OpenCL for heterogeneous computing. They

• given overview of various accelerators with focus on GPUs,
• illustrate how to access GPUs for computating,
• suggest a reduction implemention for GPUs.

MPI

These lectures introduce the MPI framework for multinode parallel processing. They

• introduce message based communication,
• deadlocks arising from concurrency,
• group communication patterns.