Fork me on GitHub
#cljs-dev
<
2015-12-10
>
spinningtopsofdoom15:12:28

@dnolen & @alexmiller I've improved the Lean HAMT perf for seqs to ~30%- 40% of the current HAMT implementation for all the JS engines (except FF). For small maps (<= 100) the perf differs narrows. It seems the perf improvements for the paper are born out in practice. I'll be working on improving hash and = next

dnolen15:12:15

@spinningtopsofdoom: very cool! what changes did you have to make for the perf bumps?

spinningtopsofdoom15:12:06

Changing the iterating over the sub nodes from inc to the number of sub nodes to dec to 0. That allowed me to have a a custom array copy function of an array of size 7. Those were the main perf bumps

bronsa15:12:31

@spinningtopsofdoom: do you have a repo with those impls?

spinningtopsofdoom15:12:30

The stack iteration that was used by the paper was designed for mutable oop so there was some allocation overhead that needed to be trimed for a seq based immutable version