1. 5
    Porting x86 vector bitmask optimizations to Arm NEON assembly community.arm.com
    olegkovalov avatar via olegkovalov 1 year ago |
    Archive.org Archive.today Ghostarchive
    | 1 comment
    1. 2
      valpackett avatar valpackett edited 1 year ago | link

      Ooh, interesting! How does this approach compare to this bulk one?

      upd: oh I see

      ZSTD uses “LD4 u8 interleaved” but SIMDJSON uses the “Pairwise” approach whereas there was no consistent winner across a set of benchmarks