Heterogeneous Computing
HSA Standard seeks to pair CPU and GPU more efficiently

The HSA foundation has published the version 1.0 of a specification for heterogeneous data processing. Amongst other things, it is meant to facilitate the integration of graphic processors for computational tasks. It also offers functions for the efficient implementation of programming languages such as C++, Java or Python.

One of the essential parts of the specification controls access by the CPU and graphic processor (GPU) to a shared address field and working memory. This should make it easier for applications to use GPU, for example for floating decimal tasks, while the CPU core carries out single thread integer tasks.

The specification consists of three parts: the HSA System Architecture Specification, which describes the operation of the hardware, the HSA Programmers’ Reference Manual, which provides a description of the software, and the HSA Runtime Specification, which deals with the interplay between the HSA platforms and applications. The companies AMD, ARM, Imagination, MediaTek, Samsung and Texas Instruments have worked together on the documents. (Source: HSA Foundation/rf)