Skip to main content

Cray XC40 Programming Seminar

publication date :
Event date

We will carry out a lecture on the Cray XC40 programming environment.

※ Supercomputer is under the scope of the Export Administration Regulations. Therefore, these seminars are also restricted under the Foreign Exchange Laws and related laws. Please see "Rules for Use of Supercomputers for Foreign Nationals and Japanese Nationals Residing Overseas " for details.

Rules for Use of Supercomputers for Foreign Nationals and Japanese Nationals Residing Overseas
http://www.iimc.kyoto-u.ac.jp/en/services/comp/apply/other/gaikokujin.html


Date December 1 (Thu), 2016 9:00~16:00
Venue 404 Conference Room on the fourth floor of the Academic Center for Computing and Media Studies, Kyoto University (North bldg)
Eligibility for participation XC40 users (users of one of the supercomputer system A or those who are planning to use)
Application Required
Capacity 50
Sponsor Research support section
Cooperation Unit of Computational Science, Kyoto Univ
Inquiry Research Information Section, IT Promotion Division, Information Management Department, Kyoto University
TEL:075-753-7407
E-mail:comp-info
Image
icon_atmark
kudpc.kyoto-u.ac.jp
Inquiry Form
Program

Abstract: This is a full-day tutorial on programming for the Cray XC system with Intel Knights Landing (KNL), the second generation Intel Xeon Phi. Attendees will learn how to use the Cray Programming Environment to port and optimize applications for KNL systems. We will describe the KNL architecture and emphasize some of the main aspects related to KNL features, such as the high-bandwidth memory, and how to optimize for them with the memory directives and the Cray memory analysis assistant tool. We will also focus on experiences and best-practice for optimization of scientific applications on KNL systems.


Agenda:
09:00 - 09:15 Introduction
09:15 - 09:45 Overview of the Cray XC system with KNL
09:45 - 10:30 Cray Compiling Environment (CCE) & Scientific Libraries for KNL
Default optimizations
Cray compiler feedback
Using CSML or MKL with CCE
Using MCDRAM (aprun NUMA control, memory directives)
10:30 - 11:00 Break
11:00 - 12:00 Cray MPI for KNL
MPI optimizations for KNL
Threads vs ranks
Thread hot MPI
Using huge pages with MCDRAM
Core specialization
12:00 - 13:30 Lunch
13:30 - 15:00 Cray performance tools & Reveal
CrayPAT-lite
Bottleneck detection
Grid order / Rank order
Loops Statistics
Reveal
Memory analysis assistance
15:00 - 15:30 Break
15:30 - 15:45 Tips for KNL
15:45 - 16:00 Summary & questions