Module 1 a: Overview and special talks
Module 1b: Performance Analysis Tools
Module 1c: Vectorization of C++ codes using SIMD types
Details below:
Module 1 a: Overview of performance analysis and improvement
Topics: The module is focused on the introduction to the fundamentals of serial and parallel performance, performance analysis, and tools for analyzing. Additionally, an overview about some concepts for performance improvements will be given and a talk about algorithmic differentiation (AD) will show possible improvements for differentiatons.
A special talk is addressed to researchers and administrators with the tilte "Quality-of-Service for the Lustre file-system".
Method: Lectures and talks
Target group: User with source code access
Organizer of module 1a: Christian Iwainsky
Trainers/Speakers: Brinkmann, Iwainsky, Sagebaum, Terboven, Wolf
Date: Monday September 10, Lectures: 10:00-16:00; Talks: 16:00-18:00
Location: TU Darmstadt, Lichtwiese, Franziska-Brauns-Straße 7, Hörsaal- und Medienzentrum (HZM) L4|02 301
Attendance fee: 15.00 EUR
Module 1b: Performance Analysis Tools
Topics: Experts will present tools for performance analysis. In hands on sessions, these tools will be applied to the codes of the participants.
Tools: ScoreP, ExtraP, Scalsca, HPC Toolkit, Vampir.
Participants must prepare test cases in advance of the workshop. Guidelines will be available two weeks before the workshop
Methods: Lectures and hands-on session
Target group: Users with source code access (FORTRAN, C, C++)
Date: Tuesday-Wednesday, September 11-13, 09:00-18:00, and Thursday, September 13, 09:00 - 15:00
Location: TU Darmstadt, Lichtwiese, Jovanka-Bontschits-Straße 7, Gerhard Pahl-Zentrum (Lehrzentrum Maschinenbau), L1|10 PC Pool
Trainers: Alexandru Calotoiu (TU Darmstadt), Michael Knobloch (FZ-Juelich), Christian Iwainsky (HKHLR), Ronny Tschueter (TU Dresden), Matthias Weber (TU Dresden)
Attendance fee: 45.00 EUR
For module 1b the participation at 1a is mandatory
Module 1c: Vectorization of C++ codes using SIMD types
Topics:
- SIMD & data parallelism introduction
- Algorithm vectorization
- Vectorizing data structures
- ISO C++ direction wrt. SIMD
Method: Lectures and hands on sessions
Target group: Developers and architects of HPC codes
Date: Friday, September 14, 09:00 - 16:00
Trainer: Matthias Kretz (GSI)
Location: TU Darmstadt, Lichtwiese, Jovanka-Bontschits-Straße 7, Gerhard Pahl-Zentrum (Lehrzentrum Maschinenbau), L1|10 PC Pool
Attendance fee: 15.00 EUR
