> On 2019-04-24, at 1:29 PM, Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com> wrote:
>
> Hi Tim,
> I've posted some enhancements with a 3-way Toom-Cook multiplication algorithm
> (it is a generalization of karatsuba, but we divide in n parts instead of 2, here 3 parts for Toom-3)
Wow. That's nearly twice as fast for the fib(4784969) - 8.4 sec. Amazing. So about 10x over the original.
Now *printing* the number is rather slower and if you have any magic code to improve that it would be interesting.
I would *strongly* support putting the algorithm improvements into trunk. Very little code for colossal speedup and a really interesting exemplar of advanced algorithms.
+1
changes are in inbox now
they should have associated tests
For the printing, cost is dominated by division.
division cost is dominated by multiplication.
We could opt for a divide and conquer algorithm too, or mixed Barett-Newton-Raphson which is the most efficient known algorithm