In this video, Manuel Ujaldon from the University of Malaga presents: Programming Models for Heterogeneous Computing. Recorded at the HPC Advisory Council Spain Workshop 2012 in Malaga.
After a decade of being used as hardware accelerators, GPUs constitute nowadays a solid alternative for HPC at an affordable cost. Since then, and followed by the evolution of many-core architectures, programming models have evolved based on SIMD parallelsim to provide powerful and scalable methods for mapping algorithms and applications on those platforms. In this talk, we will review existing alternatives for programming codes targeted on many-core GPUs, like CUDA, OpenCL, and OpenACC. We will also describe third party wrappers like those available for Fortran, Java, and Matlab, and analyze future trends which look promising for next-gen heterogeneous computing.