High Performance Computing I

This lecture provides an introduction to parallel computer architectures and standard programming interfaces for parallel numerical algorithms. We focus on few numerical algorithms like dense matrix multiplications and LU decompositions which allow us to approach step by step the theoretical peak performance.

Contents

  • Introduction to the programming language C++ with a special focus on numerical linear algebra
  • Memory hierarchies
  • Parallel computer architectures
    • shared memory with POSIX threads and OpenMP
    • distributed systems with MPI
    • GPUs

Schedule

  • Monday, 2 pm to 6 pm, Helmholtzstraße 18, room E.20, and in the labs Helmholtzstraße 18, room E.44, and O27/211.
  • Friday, 2 pm to 4 pm, Helmholtzstraße 22, room E.04, and in the labs Helmholtzstraße 18, room E.44, and O27/211. Please note that Friday sessions are open-ended, i.e. we provide support in the labs until 6 pm.
  • The first lecture will be on 15 October 2018, 2 pm.

Prerequisites

Linear algebra, calculus, numerical linear algebra, and programming are required. We expect some basic knowledge of C. Knowledge of C++ is not required as we provide an introduction to C++. Introduction to High Performance Computing is recommended but not strictly required. Consider, however, that probably considerable more time will be required for the assignments of the lab sessions without the associated introductory lecture of the bachelor program.

Exams

The exams will be held orally in March and April 2019. An oral exam takes about 45 minutes. You are free to chose English or German as exam language. An active and successful participation in the labs (at least half of the quizzes are passed) and an registration for the exam at HIS is required.

Lectures

Resources and examples of the lectures will be posted here:

  • 15 October 2018: First steps with vectors in C: lecture notes, session
  • 19 October 2018: First steps with matrices in C: lecture notes, session
  • 22 October 2018: Selected BLAS level 1 functions, benchmarks and gnuplot: session
  • 26 October 2018: Simple cache optimizations: session
  • 29 October 2018: Simple cache optimizations for GEMM: session
  • 2 November 2018: Cache optimizations for GEMV: session
  • 5 November 2018: First steps with C++: session
  • 9 November 2018: C++ tools for managing memory buffers: slides, session
  • 12 November 2018: Packing matrix blocks for an efficient GEMM implementation: slides, session
  • 16 November 2018: GEMM micro kernel, GEMM macro kernel, GEMM frame routine: session
  • 19 November 2018: Generic classes, template functions, and static polymorphism: session
  • 23 November 2018: Function objects and lambda expressions: session
  • 26 November 2018: Unblocked LU factorization: session
  • 30 November 2018: More on vector and matrix classes: session
  • 3 December 2018: First steps with threads in C++: session
  • 7 December 2018: Mutex and condition variables: session
  • 10 December 2018: Thread pools (part one): session
  • 14 December 2018: Thread pools (part two): session

Lab sessions

Lab sessions will be held on two times per week. Each session comes with an online guide that summarizes the preceding lecture and provides practical exercises.

A registration at SLC for HPC I is necessary to participate in the lab sessions.

Some of the lab sessions come with a quiz which are due a week later:

 

NumberIssuedDeadlinePDF
126 October 20182 November 2018PDF
25 November 201812 November 2018PDF
316 November 201823 November 2018PDF
47 December 201814 December 2018PDF

Contact

Dr. Andreas F. Borchert
Helmholtzstr. 20
Room 1.23

Dr. Michael Lehn
Helmholtzstr. 20
Room 1.09

M. Sc. Constantin Greif
Helmholtzstr. 20
Room 1.28