Parallel Numerics - Winter 11

Term

Winter 11

Lecturer

Univ.-Prof. Dr. Thomas Huckle

Time and Place

Lecture: Monday 10:15 - 11:45 Uhr (MI 02.07.023), Tutorial Friday 10:15 - 11:45 Uhr (MI 02.07.023)

Audience

CSE (compulsory course, 3rd semester), Mathematics (Master), Informatics (Master) (Modul IN2012)

Tutorials

Matous Sedlacek

Exam

t.b.a.

Semesterwochenstunden / ECTS Credits

SWS (2V + 2Ü) / 5 Credits

TUMonline

{{{tumonline}}}

---

IN2012

This course will be given in every winter term. The lectures and tutorials are conducted in English, and the course substitutes the German lecture "Numerik auf Parallelrechnern".

News

1. First lecture: 24.10.2011
2. First tutorial: 28.10.2011

Contents

1. High-Performance Computing
2. Performance: Analysis, Modeling, and Measurements
3. Basic Linear Algebra Subprograms
4. Direct Solution of Sparse Linear Systems
5. Iterative Methods for Linear Systems
6. Linear Eigenvalue Problems
7. Programming in MPI

Course Material

Lecture Notes

• [Lecture notes]

Slides

• [Lecture 1, Introduction]
• [Lecture 2, OpenMP, Graphs]
• [Lecture 3, Graphs, Colouring, BLAS ]
• [Lecture 4, LE with dense matrices, (parallel) Gaussian Elimination ]
• [Lecture 5, QR decomposition, Sparse storage formats ]

Tutorials

Corrections with new version are marked in red.

Tutorial	Topics	Worksheet	Proposal for solution / code	Slides
1	Flynn's Taxonomy & MPI Basics	worksheet1.pdf	worksheet1_solution.pdf ws1_task6.cpp
2	Numerical Integration & P2P Communication I	worksheet2.pdf	worksheet2_solution.pdf ws2_task1.cpp	Flynn_SPMD.pdf
3	Vector-Vector Operations & P2P Communication II	worksheet3.pdf	worksheet3_solution.pdf ws3_task5.cpp	Comm_Modes.pdf
4	Matrix-Matrix-Operations & P2P Communication III	worksheet4.pdf	worksheet4_solution.pdf ws4_task5.cpp
5	Parallel Gaussian elimination & Collective Operations	worksheet5.pdf	worksheet5_solution.pdf ws5_task3.cpp	Collective.pdf
6	Tridiagonal Matrices, Hockney/Golub method & Message Tags	worksheet6.pdf	worksheet6_solution.pdf ws6_task1_serial.cpp ws6_task1_parallel.cpp ws6_task3.cpp	Hockney_Golub_Notes.pdf
7	Sparse Matrix-Vector Multiplication & MPI Communicators	worksheet7.pdf	worksheet7_solution.pdf ws7_task3.cpp
8	Stationary Methods	worksheet8.pdf	worksheet8_solution.pdf ws8_task3_serial.cpp ws8_task3_parallel.cpp	Jacobi_Relaxation_Notes.pdf
9	Domain Decomposition	worksheet9.pdf	worksheet9_solution.pdf

Literature & External Links

1. Introduction to High Performance Scientific Computing (Eijkhout, Chow, van de Geijn) [free download]
2. Numerical Linear Algebra for High-Performance Computers (Dongarra, Duff, Sorensen, van der Vorst)
3. Parallel Algorithms for Matrix Computations (Gallivan, Heath, Ng, Ortega,...)
4. A User's Guide to MPI (Pacheco)
5. Iterative Methods for Sparse Linear Systems (Saad)
6. Loesung linearer Gleichungssysteme auf Parallelrechnern (Frommer)

1. [An Introduction To Quantum Computing for Non-Physicists]

Exam

• Date: t.b.a.
• Room: t.b.a.
• Registration: t.b.a.

Allowed Material for the exam

• 1 two-sided hand-written sheet of paper
• "Message Passing Interface Quick Reference in C" [MPI Quick Reference]

Regulations

The exam takes place on t.b.a.. Make sure you are in place on time. As the lecture hall is reserved from t.b.a., and as we have to leave at t.b.a., there's no chance to wait for late arrivals. A handwritten sheet of paper (size A4, front and back page) may be used during the exam as mnemonic as well as the MPI operation reference distributed during the tutorials. No other material besides dictionaries is allowed.

Please study your examination regulations. Students in most disciplines have to register for the exams throughout the semester via TumOnline.

Students that did not pass the exam or did not participate and still need the credits should contact Prof. Thomas Huckle for an appointment as soon as possible. In this case, the appointment always is the second try, i.e. if you do not take part in the exam, it is a failed try.

Old Exams

• [Winter 10], [solution]
• [Winter 09], [solution]
• [Winter 08], [partial solution]
• [Winter 07]
• [Winter 06]
• [Winter 05]