|
|
tCSC2013 Parallelism Theme
|
Parallelism Theme
|
|
Session
|
Description
|
Lecturer
|
|
Lecture 1
|
Introduction
This lecture motivates multicore parallelism
with focus on energy consumption, and
connects to the various sources of
parallelism in today's hardware. The
audience will learn how to calculate the
peak performance of common hardware such as
a multi-socket CPU system as well GPGPU
hardware. With the sources of hardware
parallelism in mind, the lecture introduces
how performance scales to a larger level.
The software aspect starts from an operating
system's point of view and continues with
parallel patterns that are used to extract
and express parallelism. This guides into
parallel programming and actual programming
languages. This continues into parallelism
issues and tools that help mastering these
problems. In the last part of the lecture,
the Pthreads library as a foundation of
multicore programming is introduced which
includes a discussion of task scheduling and
synchronization. The lecture closes with a
motivation of higher level programming
models.
|
Hans Pabst
|
|
Lecture 2
|
OpenMP and Cilk Plus
This lecture thoroughly introduces OpenMP as
a parallel programming standard. The lecture
motivates its use in scientific computing,
but also covers task parallelism and even
recent additions in OpenMP 4.0. Towards the
end of the lecture, Cilk Plus is introduced
as another compiler-based programming model.
Cilk Plus is given as a reference that
focuses on task scheduling and lock-free
programming. Some more advanced topics such
as serial elision and hyper-objects are
touched as time permits. The lecture closes
with a discussion of each models strengths
and connects back to the Pthreads
foundation.
|
Hans Pabst
|
|
Lecture 3
|
C++11 and TBB
This
lecture thoroughly introduces Intel TBB as a
library-based programming model. Due to its
focus on C++, new parallelism aspects of
C++11 such as the memory coherence model are
introduced. This part also connects back to
the foundations of parallelism where
parallel patterns, tasks, and scheduling
have been introduced. The lecture continues
with performance tuning aspects, and
reiterates parallelism issues with practical
examples on concurrent data structures as
well as general synchronization primitives.
The lecture continues with a pattern that
helps to extract parallelism on a function
level, and closes with an outlook on
message-driven programming.
|
Hans Pabst
|
|
Exercise 1
|
TBW
TBW.
|
Hans Pabst
|
|
Exercise 2
|
TBW
TBW.
|
Hans Pabst
|
|
Exercise 3
|
TBW
TBW.
|
Hans Pabst
|
|
Prerequisite
and
References |
Desirable Prerequisite
|
|
|
 |
