Masterpraktikum Scientific Computing - High Performance Computing - Winter 15
- Term
- WS 15
- Lecturer
- Michael Bader
Sebastian Rettenberger - Time and Place
- Preliminary meeting: Wednesday 8.7.15 16:00, room 02.07.023
Monday 16:00, room 02.07.023 - Audience
- Masterpraktikum (Modul IN2106), Wahlpflichtpraktikum im Bereich theoretische Informatik (Modul IN8904)
Studierende im Masterstudiengang Informatik, Studierende der Informatik im Hauptstudium, Studierende der Techno-, Diplom- und Finanz-Mathematik
Language English
- Tutorials
- -
- Exam
- -
- Semesterwochenstunden / ECTS Credits
- 10 Credits
- TUMonline
- https://campus.tum.de/tumonline/wblv.wbShowLvDetail?pStpSpNr=950199269
Contents
News
- Presentation Project: Monday 25.1.16
- Report: Project Phase: Monday 11.1.16
- Presentation "assignment 4": Monday 14.12.15 16:00
- Presentation "assignment 3": Monday 23.11.15 16:00
- Presentation "assignment 2": Monday 9.11.15 16:00
- Presentation "assignment 1": Monday 26.10.15 16:00
- Kickoff: Monday 19.10.15 16:00, room 02.07.023
- Preliminary meeting: Wednesday 8.7.15 16:00, room 02.07.023
- if you have questions in advance -> Sebastian Rettenberger
Content
High Performance Computing has become a critical success factor in research and industry. Even the computational power of commodity parts like desktop systems increased rapidly due to vectorization, EPIC, multi-core. This lab course uncovers the principles behind these buzzwords and highlights programming concepts, software design and coding styles which are mandatory to unleash the power of modern chips.
We focus on developing datastructers and algorithms with respect to their performance, for both single and parallel executions. We deep-dive into different platforms like CPUs, coprocessors and multi-node clusters. Furthermore, we cover several programming concepts like auto-vectorization, explicit vectorization, OpenMP and MPI. The platform differences are evaluated through detailed runtime experiments.
Topics:
- Performance-Optimization of sequential applications
- Pipelining
- Loop unrolling
- Data dependencies
- Caches
- Cache blocking
- Performance-Measurement
- Pipelining
- State-of-the-Art Hardware
- CPUs
- Memory Subsystems
- Supercomputer
- Xeon Phi coprocessor
- Parallele Programming
- OpenMP
- MPI
The course is held in two parts. The first one introduces basics, concepts and programming models through theoretical and practical exercises. You work on these exercise in small teams. The second part is a real-life task: each team has to optimize a given scientific application with respect to the underlying hardware architecture.
C-Programming and Linux Know-How are helpful!
Slides
- 12/14/15: Repast HPC
- 11/23/15: Profiler
- 11/09/15: MPI, CG
- 10/26/15: OpenMP
- 10/19/15: Organization, Vectorization
Exercise Sheets
- 11/23/15: Exercise Sheet 4
- 11/09/15: Exercise Sheet 3 - Update | Code
- 10/26/15: Exercise Sheet 2 | Code
- 10/19/15: Exercise Sheet 1 | Code | hint: DGEMM Optimization 1, DGEMM Optimization 2
Material
- MPI 2.2 Standard: http://www.mpi-forum.org/docs/mpi-2.2/mpi22-report.pdf
- OpenMP Tutorial: https://computing.llnl.gov/tutorials/openMP/
- Intel C++ Compiler Reference Manual: https://software.intel.com/en-us/compiler_15.0_ug_c
- Intel Architecture Manual: http://www.intel.com/products/processor/manuals/index.htm