Organization of multiplication functions

Kevin Ryde user42 at zip.com.au
Fri Sep 19 12:50:56 CEST 2003


Torbjorn Granlund <tege at swox.com> writes:
>
> I'll reorganize this once the ongoing toom3 rewrite has settled.

Of course there's probably not much chance of anyone writing a
realistic program which uses mul but not sqr.

The other reorganisation was to split the toom3 out, so that by
setting the thresholds one could get only the karatsuba code, if
trying to save code space, esp if knowing only moderate sized operands
are going to arise.

> Since you're the main configure/make maintainer,
> I will wait for you to enhance the build system to accomodate
> vector machines with that suggested improvement.  :-)

:-) Somehow I think it'd be more trouble than it's worth.  Better if
Cray made their system more like traditional C where a function call
has relatively low overhead, for most functions at least.

Maybe if someone's got a cray they're actively using then they could
think about which of the low level routines should be inlined.  Or
partially inlined so the usual cases are inline but there's a call to
propagate multiple carries, or something like that.


More information about the gmp-devel mailing list