libCEED
libCEED copied to clipboard
Arm SVE intrinsics
Arm SVE intrinsics are vector-length agnostic, so represent a nontrivial difference in strategy from Intel intrinsics. An example shows how one might program with these intrinsics:
https://developer.arm.com/documentation/100891/0612/coding-considerations/using-sve-intrinsics-directly-in-your-c-code
Scatters and gathers are supported.
The new Neoverse V1 and N2 will support SVE, so we can expect to see it appearing in more server environments (i.e., not just A64FX). https://fuse.wikichip.org/news/4564/arm-updates-its-neoverse-roadmap-new-bfloat16-sve-support/