The idea behind this unified programming model is fueled by the changing requirements of how hardware is being used for deep learning algorithms.
What is Intel’s OneAPI initiative?
OneAPI initiative is Intel’s unified programming approach to simplify application development across different computing architectures.
It will facilitate programming of CPUs and accelerators using modern C++ features that express parallelism through a new programming language called Data Parallel C++ (DPC++).
Overview of Data Parallel C++ programming language
(DPC++) is a programming language that will allow developers to code for CPUs, GPUs, FPGAs, and other hardware accelerators with one source code document.
It is an open, cross-industry programming language (based on C++ concepts and SYCL) that is touted to be an “alternative to single architecture proprietary language.”
The goal here is to simplify programming and enable code reuse across hardware targets while allowing for tuning to specific accelerators.
The Intel OneAPI beta release supports two programming styles: Direct Programming and API Programming. DPC++ falls under the direct programming wing.
These two programming styles will provide an efficient unified development model that facilitates a complete, native code performance across a range of heterogeneous processing hardware.
So with OneAPI, we can expect faster application performance, more developer productivity, and greater innovation.