I have a short code snippet (purely as an example of an issue I think I'm running into a lot with icx). We are trying to switch from icl to icx, but we''re experiencing a performance degradtion on our code base of around 30%.
Godbolt compiler explorer link: https://godbolt.org/z/GWEaGqj3o
As you can see, I have a very simple loop that fills an array. icl vectorizes it, icx does not, unless I change the pragma vector aligned to pragma vector always. The (measured) performance difference is a factor 4. But, vec-report says that vectorizing it does not help, so icx does not vectorize it without adding this pragma.
Am I missing something obvious here?
(I tried to add code snippets in this post, but unfortunately the forum keeps flagging it as spam).
