This workshop introduces SYCL*, an open-standard, cross-architecture programming language that supports the oneAPI programming model for heterogeneous computing across targets: CPU, GPU, FPGA, and more.

This session covers:

• oneAPI and how it can solve the challenges of programming in a heterogeneous world
• The SYCL language for high-performance computing (HPC) applications, a free download available as part of the Intel® oneAPI Base Toolkit
• SYCL buffers and accessors for data and memory management between host and device
• The basics of graphs and dependencies
• Intel® DevCloud for testing Intel® oneAPI tools and performance libraries
• Hands-on training through simple SYCL code with a Jupyter* notebook on Intel® DevCloud

Highlights

0:00 Speaker introductions

2:00 Objectives

3:10 Agenda

4:20 Programming challenges

5:05 What is oneAPI?

6:45 Overview of Intel® oneAPI toolkits

10:00 How the Intel® DPC++ Compatibility Tool minimizes code migration

11:03 A look at Intel® VTune™ Profiler

12:10 Intel® Advisor—a design assistant

13:43 Get set up on Intel DevCloud and Jupyter environments

19:44 Intel oneAPI implementation of SYCL

22:03 A peek at a complete SYCL program

23:30 Buffer memory model

24:54 What happens when you submit to a device?

26:19 Important classes in SYCL

28:05 Accessor modes

29:20 How to write basic parallel kernels

32:58 A SYCL application example

36:35 An implementation of sub buffers

37:34 Asynchronous running

40:00 Host accessors

41:10 Buffer destruction

41:49 Custom device selector

43:30 Run graph scheduling

44:15 Read after write (RAW)

45:28 Write after read and write after write

46:28 Linear-dependency chain graphs and Y-pattern graphs

48:00 Hands-on lab

54:45 Module 1: Introduction to oneAPI and SYCL

1:02:45 Module 2: SYCL program structure

1:12:52 Another look at Intel Advisor

1:23:20 Q&A