C++ on GPUs done right?
General-purpose computing on graphics processing units (GPGPU) has become a widely adopted way of leveraging highly demanding workloads in e.g. scientific simulations, computer aided engineering, visualization and data analysis in recent years. The number of open-source libraries that use GPUs to mitigate algorithmic bottlenecks exposing a high degree of parallelism is increasing almost weekly. After a brief overview of state-of-the-art hardware, I would like to introduce the current programming paradigms to execute code for GPGPU from different vendors (nVIDIA, AMD, etc) through a "Hello World" style example. I will then discuss advantages and disadvantages of these approaches based on personal experiences made with an open-source C++ library. Finally, I conclude by discussing the efforts of the ISO committee to establish a parallel STL in C++17.
Speaker: Peter Steinbach
Slides: C++ on GPUs done right?