feat(math): binary exp

[milancermak]

Updating the pow impl to use binary exponentiation, which is cheaper.

Please check the type of change your PR introduces:

• Bugfix
• Feature
• Code style update (formatting, renaming)
• Refactoring (no functional changes, no API changes)
• Build-related changes
• Documentation content changes
• Other (please describe):

What is the current behavior?

pow is O(n)

What is the new behavior?

pow is O(log n)

Does this introduce a breaking change?

• Yes
• No

Other information

I did a simple benchmark. Stepwise performance is the same for a simple case, the greater the exponent the more efficient this new implementation is.

Added a testcase to check for the whole 2^N range.

FWIW, it's pretty much the same as BitShift::fpow 😁 but without the use of Bitwise ops.

[0xLucqs]

I think that we want to keep slow pow for small numbers

[milancermak]

slow pow for small numbers

What do you mean?

[0xLucqs]

@enitrat Were you the one saying that the slow power is more efficient than fast power ?

[enitrat]

yes but I didn't run the benchmarks yet. Basically the overhead cost brought by additional div/rem opérations outweigh the gains for small exponents. I think the difference is also bigger when ran on u256s

edit: Actually @milancermak ran the benchmarks and it looks similar for small exponents as well!

[milancermak]

Yeah, if the test runner can be trusted, the small exponents gas usage stays the same.

[0xLucqs]

ok you win let's merge it