As experienced programmers know, to even remotely utilize the power of today's computers it is necessary to parallelize one's code. The first methods devised require explicit creation and control of parallelism which proved to be too difficult and dangerous. Highlevel approaches to specify parallel actions exist for a long time but come with a significant performance penalty. In the last few years techniques matured which allow to specify parallelism in an abstract form without sacrificing much if any performance.
The talk will provide an overview over the many ways gcc allows to exploit the power of modern CPUs and accelerators without abandoning the C/C++ development environment.
Skills
To follow the talk medium to high levels of C++ experience will be needed. Some knowledge about compilers and the Unix development environment are useful but not essential.
Lernziele
Educate people about modern ways of C++ programming which transparently can bridge CPU technologies from embedded to high-end workstation CPUs to accelerators with the same code base. Most people's knowledge about this world is unfortunately stuck in the 90s which is completely insufficient.
// Referent
// Ulrich Drepper
worked on free software projects, mostly in the Linux environment, for 20+ years. He maintained the C/C++ runtime environment for Linux for 15 years and still contributes to all kinds of OS projects and programming language implementations.