Advanced SYCL* Concepts for Heterogeneous Computing

Learn and practice the advanced concepts and features of DPC++ with live sample code on Intel® Developer Cloud. This workshop shows you how to efficiently write SYCL* code for heterogeneous computing. Be one of the first programmers at the forefront of this compelling and important new development that elevates computing to a new level.

In this workshop, Rakshith Krishnappa (a developer evangelist from Intel) covers some advanced concepts in SYCL programming for heterogeneous computing:

• Learn pointer-based memory management for heterogeneous computing using Unified Shared Memory (USM).
• Understand implicit and explicit ways of moving memory using USM and handling the data dependency between running kernels.
• Understand the advantages of using sub-groups in SYCL programming.
• Take advantage of group algorithms for sub-groups and understand how sub-group shuffle operations can avoid explicit memory operations.
• Use SYCL reduction to simplify reduction with parallel kernels.
• Apply the advantages reduce function to do reduction at the sub_group and work_group levels.

0:00 Introductions

1:55 SYCL 2020 agenda

2:43 Learning objectives

3:08 What is the oneAPI implementation of SYCL?

3:36 Extend the SYCL standard

4:04 C++, SYCL, and extensions

4:30 Language simplification

6:08 USM and a developer view of USM

6:53 USM setup

8:10 USM example

9:34 SYCL buffers method

11:03 Why is USM important?

11:55 Three ways to create USM allocations

13:04 Example of USM explicit data transfer

15:05 Example of USM implicit data transfer

15:52 Get started with Intel® Developer Cloud and a Jupyter* environment

19:00 Hands-on lab: Data transfer

26:20 When to use USM

28:00 USM data dependency in tasks

28:30 Data dependency example

33:05 Hands-on lab: Data dependency

38:22 USM summary slide

38:52 Sub-groups

39:18 How sub-groups map to hardware

40:22 Why use sub-groups

41:06 Sub-groups visualization and syntax

42:34 Sub-group shuffles syntax example

44:12 Sub-group group algorithms syntax example

44:53 Specifying the sub-group size

46:13 Hands-on lab: Sub-groups

1:01:12 Sub-groups recap

1:01:27 Reductions

1:02:16 Simple reduction example

1:03:00 Parallelizing reductions visualization

1:03:32 Work-group reduction

1:05:00 Simplified reduction

1:06:00 Multiple reductions in one kernel

1:06:48 Hands-on lab: Reductions

1:18:58 Reductions summary slide

1:19:14 Summary slide of workshop

1:20:36 Q&A