toom54

[bodrato at mail.dm.unipi.it]

Ciao,

Il Lun, 13 Febbraio 2012 2:08 pm, Niels Möller ha scritto:
> Torbjorn Granlund <tg at gmplib.org> writes:

>> shell:~tege/gmp/mpn/generic/toom{54,72}_mul.c.
>
> Ah. That version is virtually identical (not surprising, given that both
> versions are intimately related to the same toom63_mul.c). Just some

Yes, Toom-4.5 inversion is structured with the Toom'n'half strategy. It
shouldn't be difficult to write a single function working both as 54 and
63 :-)
For those people with no access to shell, my (old) code is available on my
site: http://bodrato.it/software/toom.html#TC4.5 .

>> The diagrams at https://gmplib.org/devel/ include timing for Marco's
>> functions.  It seems toom54 is quite useful, toom72 less so.  (These
>> diagrams are from 2009, things will have changed.)

Yes, things have changed, the main such change comes from the new toom6h
and toom8h functions. It would be nice to regenerate the diagrams with the
new algorithms. I guess toom72 will not cover a wide region in a current
version.

> toom52 and toom62 are also unused. Which reminds me that I should
> correct the toom63 row in the diagram in mul.c.

And that diagram reminds me that the unbalancement capability of toom6h
and toom8h are still unused...

Regards,
m

-- 
http://bodrato.it/toom-cook/