Commits on Dec 2, 2022

1. zstd: Improve best compression's match selection …

   The best encoder selects matches based on the criterion
   
   	a.est+(a.s-b.s)*bitsPerByte>>10 < b.est+(b.s-a.s)*bitsPerByte>>10
   
   If this were computed on the reals, it would be equivalent to
   a.est < b.est, so the added terms only capture round-off error
   (this is also why CSE doesn't eliminate them).
   
   Changing the formula to
   
   	a.est-b.est+(a.s-b.s)*bitsPerByte>>10 < 0
   
   captures the intention better, I think, and improves compression:
   
   	enwik9           260989017 259699309 -0.4942%
   	silesia/dickens    3233958   3222189 -0.3639%
   	silesia/mozilla   16980973  16912341 -0.4042%
   	silesia/mr         3505223   3505553  0.0094%
   	silesia/nci        2313702   2289871 -1.0300%
   	silesia/ooffice    2915199   2896410 -0.6445%
   	silesia/osdb       3364752   3390871  0.7763%
   	silesia/reymont    1658404   1656006 -0.1446%
   	silesia/samba      4330660   4326783 -0.0895%
   	silesia/sao        5399736   5416932  0.3185%
   	silesia/webster    9987784   9966351 -0.2146%
   	silesia/xml         542081    538378 -0.6831%
   	silesia/x-ray      5756210   5733061 -0.4022%
   
   ... as well as throughput:
   
   	name                              old speed      new speed      delta
   	Encoder_EncodeAllSimple/best-8    12.1MB/s ± 1%  12.2MB/s ± 1%  +1.17%  (p=0.000 n=18+20)
   	Encoder_EncodeAllSimple4K/best-8  10.4MB/s ± 1%  10.5MB/s ± 1%  +0.82%  (p=0.000 n=20+20)

   

   greatroar committed Dec 2, 2022
   Copy the full SHA

   2b47341 View commit details

   Browse the repository at this point in the history