hgcd1/2

Torbjörn Granlund tg at gmplib.org
Tue Sep 3 08:49:52 UTC 2019


nisse at lysator.liu.se (Niels Möller) writes:

  And this on a laptop with an Intel U4100 (5 years old?), so I'd assume
  it doesn't have a particularly fast div instruction. Should we just
  delete div1 ? On which architectures can we expect it to be beneficial?
  It should be fairly easy to find out, if we define a HGCD_DIV1_METHOD
  known to tuneup, to select between plain division and the div1 function.

Interesting but not too surprising results.

Intel ark doesn't seem to know any processor called "U4100" so I cannot
figure out what generation it belongs to.

IIRC, Intel has not improved plain 64b/64b division since Haswell, which
is older than 5 years.

Again, if IIRC, small quotients may result in 16 cycle latency.  That's
the lowest possible timing.



-- 
Torbjörn
Please encrypt, key id 0xC8601622


More information about the gmp-devel mailing list