Commits on Feb 4, 2022

1. internal/lz4block: Short literal copying in arm64 decoder …

   This borrows a trick from the amd64 decoder: copy short literals by
   always copying 16 bytes, using the space beyond the literal's length as
   scratch space (but only when we're >=16 bytes from the end of src, dst).
   
   Benchmark results on RPi4B:
   
   name                old speed      new speed      delta
   UncompressPg1661-4   101MB/s ± 1%   138MB/s ± 0%  +36.46%  (p=0.000 n=10+9)
   UncompressDigits-4   453MB/s ± 0%   525MB/s ± 0%  +15.81%  (p=0.000 n=10+10)
   UncompressTwain-4    110MB/s ± 0%   156MB/s ± 0%  +41.42%  (p=0.000 n=10+10)
   UncompressRand-4    1.14GB/s ± 1%  1.13GB/s ± 1%     ~     (p=0.075 n=10+10)
   
   Also moved a SUBS to just before the associated branch to allow
   instruction fusion on ARMs that do that.

   

   greatroar committed Feb 4, 2022
   Copy the full SHA

   e99166d View commit details

   Browse the repository at this point in the history