hgcd1/2

Torbjörn Granlund tg at gmplib.org
Fri Sep 6 10:57:41 UTC 2019


We have now about half of hosts asking for HGCD_METHOD = 1 as for
HGCD_METHOD = 3.  Itanic wants HGCD_METHOD 2, and so does beagle (Arm
a8), but the latter with a preference of just 0.05% (over method 3).

_METHOD 1 benefits from the special handling of q = 1 before invoking
DIV1.  _METHOD 2 & 3 are optimised for small q, so implicitly repeats
the special handling of q = 1.

The speed difference between the preferred _METHOD and the 2nd best
_METHOD is not huge.  In most cases it is < 10%.  The greatest
difference is for the system mati at 19%.

-- 
Torbjörn
Please encrypt, key id 0xC8601622


More information about the gmp-devel mailing list