Introduce more faster and memory optimal BitSet as replacment of Set<i32>

[MaxGraey]

Benchmark results for 1M items (JS):

BitSet add: 7.995ms
BitSet has: 4.894ms

Set<i32> add: 15.898ms
Set<i32> has: 15.414ms

Benchmark results for 100M items (Wasm):

BitSet add: 39.5000
BitSet has: 29.3000

Set<i32> add: 194.6000
Set<i32> has: 207.3000

• I've read the contributing guidelines
• I've added my name and email to the NOTICE file

[dcodeIO]

What I'd find interesting in this context is a comparison of Set vs BitSet when compiled to Wasm. Mostly just curious, but perhaps the existing benchmark in JS can be extended? :)

[MaxGraey]

Here AS bench BitSet vs Set:

playground

My results for Chrome for items 100M:

BitSet add: 39.5000
BitSet has: 29.3000

Set<i32> add: 194.6000
Set<i32> has: 207.3000

1M is too small for WebAssembly =)

[HerrCai0907]

This algorithm use 1bit to store whether i32 number exists. What will happen if adding I32.MAX? It looks like will alloc 536870912(= 2 ** 32 / 8) bytes memory.

[MaxGraey]

This algorithm use 1bit to store whether i32 number exists. What will happen if adding I32.MAX? It looks like will alloc 536870912(= 2 ** 32 / 8) bytes memory.

Yes, BitSet should be used for low-sparse scenarios. Where it is actually our case. The maximum number of locales per-function is 65536 for most Wasm engines. In fact, in real apps locale count does not exceed 50 per function (even after inlines). This will often be within backed buffer size