[PATCH] D59129: [SROA] Don't lower expensive allocas with minsize

[Sjoerd Meijer via Phabricator via llvm-commits]

SjoerdMeijer added a comment.

> I'm not sure those benchmark results are good enough to justify this: there are only 4 changes, and the ups and downs almost balance each other out.

I see what you mean. But overall we save 198 bytes, and I would be happy to take a 198 bytes size reduction. Probably what I wanted to say in my previous message is that it's not a one trick pony; it not only helps with my motivating example, but other benchmarks too. But I will indeed now look into fine tuning it and see if we can avoid the regressions. As I said, I will look into doing it on a per-slice basis.

> Have you investigated the regressions to see what this makes worse, and if there's a way we could avoid them?

Yep, only 1, but some time ago. I thought we were "unlucky", but can't remember what it was. Will looking into this, and the other cases.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D59129/new/

https://reviews.llvm.org/D59129