Arbb

arbb, often styled ArBB or ArBB (Array Building Blocks), refers to a data-parallel programming framework originally developed by Intel. It was provided as a C++ library and embedded domain-specific language intended to simplify writing portable parallel code for multi-core CPUs and accelerators by exposing high-level array operations.

The core idea behind arbb was to let programmers express computations on large arrays in a way that resembled ordinary scalar code, while the library built an execution graph that could be optimized and mapped to available hardware. Programs written with arbb were designed to be portable across different backends, with the runtime responsible for translating array operations into optimized kernels for CPUs, and potentially other devices.

ArBB emerged in the late 2000s as part of Intel’s broader effort to promote data-parallel programming models. It featured a combination of a C++ API, deferred evaluation, and just-in-time-style compilation to generate efficient code tailored to the target architecture. While it attracted interest as a high-level alternative to manual threading and explicit SIMD programming, adoption was limited in practice.

Over time, Intel shifted its parallel programming strategy toward other technologies, such as Threading Building Blocks (TBB) and, later, heterogeneous programming frameworks associated with oneAPI. As a result, active development and public support for arbb diminished, and the project was effectively retired in favor of newer tools and standards.

Today, arbb is primarily of historical interest as an early attempt to provide a portable, high-level data-parallel abstraction in C++, illustrating the evolving landscape of parallel programming approaches.