Running mpn_add_n at less than 1 cycle/limb

[Torbjörn Granlund]

Albin Ahlbäck <ahlback at lix.polytechnique.fr> writes:

  Is `< 1' only applicable for throughput, or does it apply for latency
  as well?  I'm guessing that latency with less than 1 cycle per limb
  would require value prediction.

What I am saying that a mpn_add_n invocation for n limbs can take < n
cycles even with an add-with-carry instruction which have a latency of 1
cycle.  (Of course that assumes the the add-with-carry instruction has a
throughput of > 1/cycle, but that is generally the case.)

-- 
Torbjörn
Please encrypt, key id 0xC8601622