Intel’s oneAPI project has been unveiled to deliver unified programming model that simplify application development across multi-architectural computing systems.
The oneAPI will help facilitate programming of CPU and accelerator using the modern C++ features which expresses parallelism through new programming language known as Data Parallel C++ (DPC++).
It supports both direct and API programming, delivering a unified language and libraries to offer native code experience across a range of different hardware, like GPUs, CPUs, FPGAs and even AI accelerators.
About Data Parallel C++ (DPC++)
Data Parallel C++ (DPC++) is a new open and direct programming language that is cross-industry compliant, based on C++ and incorporates SYCL.
It's most distinguishing feature is support for open/cross-vendor applications, and been based on SYCL/C++, with Intel offering new analysis/debug tools and optimized libraries for various target domains.
DPC++ will allow developers to code across different hardware like GPUs, CPUs, FPGAs, and other accelerators with just a single source code document.
What One API Initiative means for developers?
The world’s diverse data-centric workloads means that developers have a lot of tasks to scale, as these architectures often require an efficient software programming model to deliver performance, oneAPI will help facilitate programming of CPU and accelerator using the modern C++ features.
It addresses the issue of diverse workloads with ease-of-use and performance based programming model, while also eliminating the need to maintain different code bases, multi-language programming, and diverse workflows.
The oneAPI’s libraries span several domains that benefit from workload acceleration, with library functions that are custom-coded for each of the target architecture.
Intel has promised to deliver enhanced versions of analytics and debug tools for DPC++ and other range of SVMS architectures, with the release of a developer beta and more details on the project within this fourth quarter.