mulmid

[bodrato at mail.dm.unipi.it]

Dear Niels,

Il Lun, 3 Ottobre 2011 2:23 pm, Niels Möller ha scritto:
> I've merged David's mulmid implementation (based on the June 9 revision

Great!
We should probably decide a coherent interface for mullo and mulmid (to be
extended to mulhi) so that we can consider documenting it. I argue (from
the presentation by Zimmermann at ARITH-20) that MPFR would use them.

>   * mpn_toom42_mulmid_itch: Should this be a macro, real function, or
>     inline function? Currently it's a macro, but with a lower-case name

Some *_itch are macros, some are (inline) functions... they all are
undocumented, for internal-use-only. I'd keep a coherent name (lower-case)
in any case.

>   * There's currently no code using mulmid. Did you have an
>     implementation of invert.c or binvert.c using mulmid? If so, what's

There was some code by Harvey IIRC, for mpn_invertappr, but I do not
remember if it was intended for GMP.
I did also try to integrate it in the current code, to test the speed, but
there was some small problem with error estimate I did not work on. The
middle_product version of invertappr have a somehow larger error on
approximation with respect to the one managed by the Newton iteration
algorithm.
We should ask Harvey for details.

Regards,
Marco

-- 
http://bodrato.it/papers/