Branch Divergence in GPU Programming

Memory coalescing and branch divergence are, among others, two important aspects of SIMD programming to take care of.

We will present and discuss concrete ways to avoid or reduce branch divergence in GPU kernel development. We will provide examples and profiling statistics to assess impact on performance.

This will also be a good opportunity to meet and share our experience on GPU programming, share tools and best practices.

Slides and sample source code in English, talk in French. English speakers OK of course.

You are all welcome, newcomers included :-)

!! the meetup location is the O'Sullivans pub this time !!