Skip to content

Extract packed float implementation from SYCL only build

Paul Bauer requested to merge 4954-move-amd-fast-float-impl-to-gerneral into main

Both the HIP and SYCL build can benefit from the AMD specific optimized Float3 datastructure that ensures optimal register use and avoids additional register spilling.

To make sure we only have one such datastructure, extract the current one from the SYCL specific location and make it possible to use it for both HIP and SYCL builds.

Refs #4954 (closed)

Merge request reports