Is it more correct to write to a second native array using read-only on the original?

[wide sedge]

For a parallel job I want to edit every vertex position and i need to do it in a performant way.

Is it faster to use a read-only native array of the vertices and write to a blank native array or just edit the vertex native array in question? What's considered the right way to do it?

[potent shard]

if you can avoid allocation - that's probably better

so it's heavily algorythm you use dependent