Re: sparse polynomial arithmetic

On Apr 9, 1:00 pm, bluescarni <bluesca...@xxxxxxxxx> wrote:
......

Thanks for pointing that out! I believed the C++ bindings for GMP
automatically enabled FMA when composing arithmetic operations, but
after a quick search on Google it seems I was wrong. Using mpz_addmul
I get the following time:

real 3m12.544s
user 3m9.239s
sys 0m1.198s

More than 50% off :)

In my tests, using a generic "C" version of GMP is much slower, on
some machines, than the specific versions of GMP on a per-architecture
basis. As I recall, using a Pentium-4 specific library (on a Pentium
4) made GMP about 5-10 times faster.

This might significantly affect your benchmarks, depending on your
architecture.

Do your times include the time to convert to and from your encoding,
and to compute the appropriate bounds needed for the representation of
the product?

RJF
.