Delivering Heterogeneous Computing in C++

Wednesday 7th September 2016, 6:30 pm.

Speaker: Duncan McBain, Codeplay.

Venue: Room 4.31, University of Edinburgh Informatics Forum, 10 Crichton Street, Edinburgh, EH8 9AB.

This event is free of charge and open to all. No registration required - just turn up.

Refreshments and networking from 6:00 pm.

This meeting is supported by NCR.

Synopsis

This talk will roughly cover:

Why heterogeneous compute is important
Current methods for targeting heterogeneous hardware
- APIs, languages etc.
- What options exist for C++ today
How C++ and other standards might converge
- Parallel STL, C++ 19 plans
What the future might bring

About the speaker

Duncan is a software engineer at Codeplay. He started out on the EU-funded LPGPU project, looking at ways of improving performance on and reducing power usage of GPU devices. Since then he has worked on client projects and recently was working on Codeplay's implementation of SYCL, ComputeCpp. Most recently, he has been providing support to external developers who use ComputeCpp.

Duncan was first exposed to heterogeneous programming in the High Performance Computing course at the University of Edinburgh. For his dissertation he used the CUDA API to speed up FFT calculations in an open-source primality-testing code. This experience was enough to instigate a fascination with heterogeneous devices and the performance possibilities that they bring.