Intel workshop 19.09.2016

Introduction

We are pleased to announce a tutorial on Code Optimization with Advanced Development Tools on September 19 2016 at the ETH Zurich.

The tutorial, which includes lab exercises, will focus on using Intel Advisor for comprehensive vectorization analysis and reporting of codes. It reveals reasons for poor vectorization and advises on ways to improve it to run well on current on future processors, covering AVX/AVX2/AVX-512, OpenMP 4.x, and Xeon Phi.

The tutorial will be given by Kirill Rogozhin and Raul Akhmetshin from Intel.

Time and place

The tutorial takes place on September 19 2016 at LEE E 101, Leonhardstrasse 21, 8092 Zurich near the main building of the ETH Zurich.

Agenda

10:00–10:30 Intel Parallel Studio 2017: What's New

10:30–11:30 Unlock next-gen SIMD hardware performance secrets: AVX/AVX2 vectorization, OpenMP 4.x, Compiler vectorization challenges

11:30–12:30 Labs: Vectorization Advisor and Intel Compiler optimizing customer fluid dynamics code

12:30–13:30 Lunch break

13:30–14:30 Labs: Vectorization Advisor and Intel Compiler optimizing customer fluid dynamics code

14:30–15:30 Additional overview of new SIMD Xeon Phi/AVX-512-specific analysis including new precise FLOPs-and-Mask utilization profiler

15:30–16:30 Automated cache-aware Roofline preview feature

16:30–17:00 Q&A session

Requirements

To benefit from this tutorial you should: have some basic programming knowledge in C/C++ or Fortran. for the lab exercises: you should be able to use the Linux command line (shell) including common tools such as ssh, etc.

To participate in the lab exercises, you should bring your own laptop with a working X Server and SSH client. Participants will work on the Euler cluster of the ETH for the lab exercises. Temporary accounts for the tutorial will be provided to participants.

Linux

An X server should already be installed by default.

Mac

Install the XQuartz package.

Windows

Install PuTTy and configure to allow X11 forwarding.

Install Xming X Server (recommended), Cygwin-X, or another.

Participation

The tutorial is open to members of both academia (ETH/EPFL and ETH domain research institutes, cantonal universities and universities of applied sciences) and industry. The maximal number of participants is 60 on a first-come, first-served basis.

Registration

To register, fill out the Doodle poll where you will need to enter your email address.

Slides

Parallel Studio XE 2017: What's New

Vectorization Advisor Overview

Vectorization Intro

Feedback

You are kindly requested to give feedback to Intel about the tutorial.

Please contact use if you would like to see this or similar workshops in the future.