Arm Architecture – Armv8.2-A evolution and delivery

One thing is certain in this world – everything is changing, whether evolution or revolution and Arm Architecture is no exception. In this case the Arm v8 has evolved initially to v8.1-A and now to v8.2-A, both of which are now available in the Arm Architecture Reference Manual, known as Arm Arm.

While many partners are happy with the existing v8.0, new features and enhancements to enable new applications and usability are often requested and have been introduced in the new versions, although these are fully backwards compatible with the original architecture.

New features in v8.1-A are aimed at market segments with very large systems, a blog on The Armv8 architecture and its ongoing development can be found here.

v8.2-A improvements

Version 8.2-A includes enhanced memory model, half-precision floating point data processing and introduces both RAS (reliability availability serviceability) support and statistical profiling extension (SPE). Read the new v8.2-A documentation on Arm Developer. These updates have been described more fully in Armv8-A architecture evolution.

Reliability, Availability and Serviceability

This extension includes essential features (such as RAS and SPE) to enable enterprise computing and enhance performance in continuous sampling applications.The Spec documentation for Reliability, Availability and Serviceability - RAS - can be downloaded here.

Statistical Profiling Extension (SPE)

An analogy for SPE is using slow motion video to monitor and improve athletic performance. One beneficiary is the Server market where sampling and program optimising can produce big performance improvements and savings in both cost and processing time. Download the supplement on SPE here. More information on SPE can be found in this blog Statistical Profiling Extension for Armv8-A.

For the mobile industry, we have added further support including half-precision floating point and integer dot products. An area that benefits from these features is inference within machine learning applications.

Scalable Vector Extensions (SVE)

A further enhancement is Scalable Vector Extensions (SVE). As demand for more and more data collection and analysis grows, so does the demand for better and faster data crunching. This latest improvement is a key enabler for the High Performance Compute (HPC) and Scientific Computing sectors. This supports the Arm ecosystem partners’ development of HPC and paves the way for efficient Exascale computing. You can download the new SVE documentation here.

See our SVE blog and white paper ‘A Sneak Peek into SVE and VLA Programming’ for further depth.

Architecture Exploration Tools

Last but not least, we are pleased to offer a brand new delivery method – Arm is releasing the Instruction Set Architecture (ISA) in downloadable XML format (in addition to PDF). This will make it machine readable and more manageable for partners to understand the Arm architecture and write tooling based around it. There will be further iterations as we develop the tooling.Try it out for SVE by clicking on the button below.

Try for SVE