Small operands gcd improvements

Torbjörn Granlund tg at
Wed Aug 7 09:28:13 UTC 2019

nisse at (Niels Möller) writes:

  Just not sure what order to do things. That patch just adds the gcd_11
  entrypoint for that arch, with nothing but speed using it. (So I realize
  it's not as tested as I thought).

At least it is fast!

  If it works out to replace one foo/gcd_1.asm with foo/gcd_11.asm, one by
  one, that's faster progress (and the HAVE_NATIVE_mpn_gcd_11 test in
  gcd_11.c is unncecessary if we go that way).

I have made the trivial conversion of all gcd_1.asm code but x86-32.  I
might check it in soon.  Removing gcd_1.asm can be as a separate check

We will see a slight slowdown with all calls which go through gcd_1, but
we should regain some of that when callers avoid gcd_1 for things which
actualy need gcd_11.  (There will be some slowdown for things like
3-limb with 1-limb gcd, which will need another call.)

Please encrypt, key id 0xC8601622

More information about the gmp-devel mailing list