squaring vs multiply
Paul.Zimmermann at inria.fr
Fri Nov 8 23:30:06 CET 2013
> A significant fraction of the point-wise products' time will be spent in
> mpn_mul_basecase/mpn_sqr_basecase. The time ration for those is close
> to 1/2. This ratio seems to be reflected in toom functions.
this is a very good point. As a consequence, we should have S/M = 2/3
asymptotically, whatever the kind of FFT algorithm used (i.e., with
negligible or non-negligible pointwise products).
More information about the gmp-devel