Introduction to oneAPI, SYCL2020 and OpenMP offloading

Intel's oneAPI logo showing a 1 above the text "oneAPI"

Most current HPC systems are heterogenous and use accelerators. oneAPI is a standardized and portable programming model adapted to heterougenuous computing. In this course we will provide an introduction to Intel's oneAPI implementation, which supports two portable methods of heterougenuous computing: Data Parallel C++ (DPC) with SYCL and OpenMP for C, C++, and Fortran. Both are portable on any Intel CPU and Intel based accelerator. The course will give an introduction in these two programming methods, Intel's libraries like oneMKL and tools for performance analysis, profiling, and debugging. Further an introduction to Intel's DPC compatibility Tool to facilitate code migration from CUDA to SYCL and to Intel's MPI implementation support with GPU awereness completes the program. Insights into migrating an application to oneAPI/SYCL will be provided by GROMACS developers.

Veranstaltungsort

Online course
Organizer: HLRS, University of Stuttgart, Germany

Veranstaltungsbeginn

26. Okt 2022
08:30

Verstaltungsende

28. Okt 2022
13:15

Sprache

Englisch

Einstiegslevel

Basis

Themenbereiche

Programmiersprachen für wissenschaftliches Rechnen

Themen

Code-Optimierung

GPU-Programmierung

OpenMP

Zurück zur Liste

Prerequisites and content levels

Good knowlegde any of C/C++/Fortran and familiarity with usual OpenMP programming is sufficient for the OpenMP part. For Data Parallel C++/SYCL knowlegde of C++11 or later is recommended (C++17 very much faciliates SYCL2020 programming).

Content levels
  • Basic: 6:35 hours
  • Intermediate: 3:30 hours

Learn more about course curricula and content levels.

Instructors

Intel staff and GROMACS developers.

Learning outcomes

After this course, participants will:

  • be familiar with the oneAPI programming model,
  • have an overview over DPC++/SYCL programming,
  • have gained knowledge about fundamental OpenMP offloading,
  • have an overview over oneAPI libraries (oneMLK, ...),
  • basic knowledge about profiling and performance analysis,
  • basic knowlegde on (dynamical) debugging of programms using the oneAPI programming model,
  • be aware how Intel's Compatibility tool can help to migrate CUDA to SYCL code.

Agenda

The preliminary agenda is as follows. All times are CEST.

Day 1
Start End  
8:45 9:00 Drop in to Zoom

09:00

09:10

Welcome and Introduction to Day 1

09:10

09:30

oneAPI – Introduction to a new Development Environment
- Concept and oneAPI Standardization initiative
- Intel’s Tools Implementation – Intel oneAPI Toolkits and libs
- Transition from Intel Parallel Studio XE to Intel oneAPI toolkits

09:30

09:50

Introduction to the DevCloud
- Purpose: Demoing, testing and porting applications
- Hardware and Software offerings
- How to onboard & how to get an DevCloud account

09:50

10:00

Break

10:00

11:00

Direct programming with oneAPI Compilers (Part 1) – with Demos

- Intro to heterogenous programming model with SYCL 2020
- SYCL features and examples
   o  “Hello World” Example
   o  Device Selection
   o  Execution Model

11:00

11:15

Break

11:15

12:30

Direct programming with oneAPI Compilers (Part 2) – with Demos

   o  Compilation and Execution Flow 
   o  Memory Model; Buffers, Unified Shared Memory (USM)
   o  Performance optimizations with SYCL features

Day 2

Start End  
8:45 9:00 Drop in to Zoom

09:00

10:00

Intel OpenMP for Offloading – with Demos
- Parallelizing heterogenous applications with OpenMP 5.1
- Mixing of OpenMP and SYCL

10:00

10:35

Intel oneAPI libraries (oneMKL) for HPC  - with demos
- Performance optimized libraries for numerical simulations and other purposes

10:35

10:50

Break

10:50

11:20

Intel Debugging Tools for heterogenous programming (  CPU, GPU ) - with demos

11:20

12:00

Open Source Compatibility tool for porting purposes (SYCLomatic) - with demo
- Migration Cuda based GPU Applications to SYCL

12:00

12:30

Dynamic Debugging with Intel Inspector - with demos
- Identifying Memory and Threading Errors (Data Races and Deadlocks)

Day 3

Start End  
8:45 9:00 Drop in to Zoom

09:00

10:10

Application profiling for heterogenous hardware - Demos
- Profiling Tools Interfaces for GPU - Open Source lightweight Tools
- Profile heterogenous SYCL/OpenMP Workloads with Intel VTune Profiler
- Share experiences/key findings  with Gromacs related porting and optimization efforts

10:10

10:20

Break

10:20

11:30

Application profiling for heterogenous hardware - Demos
- Estimate performance potential gains with Offload Advisor (CPU -> HW Accelerator)
- Analyse heterogenous SYCL/OpenMP Workloads with Intel Advisor and Roofline analysis

11:30

11:40

Break

11:40

12:40

A 3rd Party oneAPI Case Study: GROMACS - A Molecular Dynamics Engine
- Heterogenous Design consideration,  alternatives and comparisons
- Real Scheduling
- SYCL - oneAPI and other Implementations
- SYCL in GROAMCS 2022

12:40

12:45

Break

12:45

13:10

Programming for Distributed HPC Systems using Intel MPI

13:10

13:15

- Questions and Answers - Wrap up

Exercises

In the course only demonstrations will be shown. However, we will also show how to access Intel's DevCloud where participants can explore and work on the examples given themself in the afternoon.

Registration-information

Register via the button at the top of this page.
We encourage you to register to the waiting list if the course is full. Places might become available.

Registration closes on October 12, 2022 (extended registration phase).

Late registrations after that date are still possible according to the course capacity.

Fees

This course is free of charge.

Our course fee includes coffee breaks (in classroom courses only).

Contact

Tobias Haas phone 0711 685 87223, tobias.haas(at)hlrs.de

PRACE PATC and bwHPC

HLRS is part of the Gauss Centre for Supercomputing (GCS), which is one of the six PRACE Advanced Training Centres (PATCs) that started in Feb. 2012.

HLRS is also member of the Baden-Württemberg initiative bwHPC.

This course is not part of the PATC curriculum and is not sponsored by the PATC program.

Further courses

See the training overview and the Supercomputing Academy pages.

Ähnliche Trainingskurse

Alle Trainingskurse

November 04 - 08, 2024

Online


November 11 - 15, 2024

Hybrid Event - Stuttgart, Germany


Dezember 02 - 05, 2024

Online by JSC


Dezember 09 - 13, 2024

Online


Januar 21 - 23, 2025

Hybrid Event - Stuttgart, Germany