udiv_qr_3by2 vs divappr

Torbjörn Granlund tg at gmplib.org
Wed Sep 5 11:17:18 UTC 2018


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

  > I believe we could find CPUs (mainly low-end and obsolete hig-end ones)
  > where the old code will beat the new code because of the old code's
  > lower submul_1 size argument.

  Any particular machines you have in mind? Would be nice with some
  numbers.

Any Sparc system except for T4/T5.
Alpha except for ev6*.
Pentium4 (in particular nocona in 64-bit mode).
The original Intel Atom in particular in 64-bit mode.
The newer inte Atom/Silvermont (but not the even newer Goldmont)
in 64-bit mode.
Arm A53/A57 and perhaps other current Arm cores.

(I compiled this list mainly from https://gmplib.org/devel/asm.html

Are any of these CPUs relevant for GMP?  Many of them are very old, but
not all of them are,

-- 
Torbjörn
Please encrypt, key id 0xC8601622


More information about the gmp-devel mailing list