Getting non-vectorized code to the speed of vectorized ones

Specific Domains Numerics
21

@DNF, assuming this is correct, perhaps your ideas on this subject would merit a place in the performance tips section?

1 Like
22

It’s not really an idea, as such. There are people on this board who know about this. I’m just reporting my understanding of things I’ve read.

I think this, anyway, is a micro-optimization, perhaps too marginal for the performance tips.

1 Like
23

Tested again with @benchmark and the median speedups were sytematically of the order of ~20%. This is not a micro saving, and the idea behind, if correct, should be highlighted.