Xeon Phi Tools Workshop
EPCC is please to announce a one-day workshop on tools for programming and optimising performance on Intel Xeon and Xeon Phi systems. Specifically, this one day workshop will provide half a day's training on using Allinea's debugger (DDT) and profiler (MAP) for Xeon and Xeon Phi systems, and half a day's training on Intel's Vector Advisor tool.
The workshop will take place in Edinburgh, after the EASC2015 conference, on April 24th 2015. The event is free to attend but places are limited so attendees must register through the following link: register Catering is also provided at the workshop, please specify if you have any dietary constraints when you register.
The outline agenda for the workshop is as follows:
- 09:30 Introduction to Allinea tools
- 09:45 Getting started with Allinea tools on small examples
- 10:30 Choosing an application for Intel Xeon Phi with Allinea Performance Reports
- 11:00 Coffee break
- 11:30 Preparing an application for Intel Xeon Phi with Allinea MAP
- 12:15 Resolving bugs with Allinea DDT
- 13.00 Lunch
- 14.00 Why is Vectorisation so important for performance?
- 14.30 So just how vectorised is my code? - Using the Intel Vector Advisor to profile vectorisation
- 15.30 Coffee
- 16.00 How can I vectorise my code? - A look at different methods available
- 16.30 Advanced vectorisation tuning - A deeper dive into vectorisation issues.
- 17.30 Finish
This event will include hands-on training on the tools, so please bring along laptops/computers to take part in these sessions.
The workshop will be held in the John McIntyre Centre, Pollock Halls, Edinburgh:
Detailed workshop descriptions
Title: From Xeon to Intel Xeon Phi with Allinea tools
In this instructor led workshop, we will first see how to determine with Allinea Performance Reports whether an application will benefit from Intel Xeon Phi or if it should be executed on Intel Xeon. Based on this information, we then focus on one application that we will migrate. Using the Allinea Forge development toolkit, we will cover how to profile an existing application and identify the most compute intensive code regions. We will then replace these regions with targetted Intel Xeon Phi implementations and review the results - before turning to the task of debugging a code to fix an error introduced during the workshop.
This workshop will cover the Allinea Forge toolkit, using its debugging tool Allinea DDT and profiling tool Allinea MAP, for the Intel Xeon Phi systems. You will learn key techniques to:
- Discover the critical performance bottlenecks
- Identify missing vectorization or poor cache performance
- Analyse application memory usage
- Measure the scaling properties of codes across threads and processes
- Debug threads and processes simultaneously
Title: Intel's Vector Advisor
The Vectorization Advisor features of Intel(r) Advisor XE radically improves the ease at which you can analyse the nature of the vectorization in the hot loops of a program. With the tool you can:
- Discover how your code is vectorised - or what is stopping the code being vectorised
- View the vectorization efficiency of each loop
- Check for loop dependencies
- Observe the memory access patterns
- Get advice on what to do to improve the vectorization efficiency of a loop
- Read other optimisation information generated by the compiler
Although the current version of the tool is only available for Intel(r) Xeon processors, the results can be directly applied to code that will be rebuilt for running on Intel(r) Xeon Phi coprocessors. If you are writing programs for Xeon Phi you should plan to attend this training. The examples we will use will clearly show what can be achieved using the Vector advisor. The tool is a real pearl!