By: Gabriele Svelto (email@example.com), January 15, 2020 7:12 am
Room: Moderated Discussions
Rather than tune for individual microarchitecture variations, we would prefer to leverage on fast string operations provided by the ISA (for example “rep movsb” on x86). This allows us to leverage wider data paths over time, without having to build custom dispatch logic which carries its own overheads.And also
Small Code Footprint: Our implementations consist of concise patterns for working with chunks of data. While further specialization can produce better results on microbenchmarks, we did not see these wins materialize on macrobenchmarks measuring application productivity.