This PR makes bncurve less slow by reusing stint integer primtivies and
unrolling a few loops and arrays to avoid array length checks and the
like.
To give an idea, it brings down processing 8k nimbus-eth1 blocks around
the 18M block height mark from 24 to 16 minutes - this is quite
significant given that a lot of time in eth1 is spent reading the
database - this is at least an order of magnitude of bncurve improvement
but probably quite a lot more - how much doesn't greatly matter but now
there's at least a decent baseline for any future performance work ;)
Of course, reusing private primitives from `stint` is not pretty - the
plan is to extract them to a separate library, work started in
https://github.com/status-im/nim-stew/pull/187.
Jacek Sieka
jangko
Ștefan Talpalaru
cheatfate