Small operands gcd improvements

[Torbjörn Granlund]

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

  Great! What additional testing do you think we need? mpn_gcd_11 unit
  tests and or devel/try support.

The former is useful, in particular to avoid end-user miscompiles.
The latter is good for asm development.

You already added tune/speed code, didn't you?  How is it used?

I recently added gcd_1 data for 1 x 1, see e.g:
https://gmplib.org/devel/tm/gmp/tuneup/success/pirigentoo64.gmplib.org-stat:64.txt

It uses -s16-64 -t48 for 64-bit limbs and -s8-32 -t24 for 32-bit limbs
to generate these numbers.

It would save some time to add analogous gcd_11 measurments to the
nightly builds to make sure the gcd_11 loops do not lose some
performance wrt the old gcd_1 loops.  (No, not all my edits were
trivial.  I respun many of the loops to save some insns and exit asap
after a zero result.)

-- 
Torbjörn
Please encrypt, key id 0xC8601622