GMP 4.3 multiplication performance

Torbjorn Granlund tg at
Mon Jun 1 01:05:54 CEST 2009

Torbjorn Granlund <tg at> writes:

  While working on multiplication improvements for GMP 4.4, I realized
  that we are doing something quite silly for unbalanced operands in GMP
  4.3.  For some machines, GMP 4.3 might actually be substantially slower
  than GMP 4.2!
I have checked in new code to the mainline gmp repository.

It solves the really bad cases, and makes better use of the toom code
that appeared in GMP 4.3.

The strategies for decomposing unbalanced multiplication needs
further work.  More unbalanced toomMN functions would help,
in particular toom43, and toom63.


More information about the gmp-devel mailing list