ni...@lysator.liu.se (Niels Möller) writes:
I have started to look a little into elliptic curve cryptography, and
there the sizes are pretty small. E.g., Using the standard curve over a
256-bit prime means that numbers are just four limbs on a 64-bit
machine. So in this case, I'd expect a
Niels,
I think there's some potential for speed up of the linear term, which is
mostly relevant for small sizes. The addmul_1 calls can run at 3 cycles
per limb or so. But then the computing the quotient involves dependent
multiplications with longer latency, so one may be able to
Torbjorn Granlund t...@gmplib.org writes:
I am afraid Marco posted both a long time ago (2009?). They live in
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