[PATCH] D149768: [ArgumentPromotion] Bail if any callers are minsize

Fri May 5 10:35:07 PDT 2023

smeenai added a comment.

In D149768#4320982 <https://reviews.llvm.org/D149768#4320982>, @aeubanks wrote:

> but another thing to consider is that we only recently started running argpromo for -O1/2/s/z. I can see two cases where this patch would regress -Oz code size. one is running an -O3 post-link with an -Oz pre-link, which wouldn't make sense in general. the second is that https://reviews.llvm.org/D148269 actually helped with -Oz code size in your case, then this patch regressed it back to where it was before

I was thinking about this too. I verified that this patch is what caused the regression, and then the reland of D148269 <https://reviews.llvm.org/D148269> (which took place after this patch) had no effect for us.

How does that pipeline change work for FullLTO though? Does the `Level == OptimizationLevel::O3` check only apply to the pre-link compilations, or does it apply to the actual FullLTO phase based on the `--lto-Ox` value as well? We should just be using the default for LTO, which is `--lto-O2` IIRC. We do build some TUs with `-O3` and some TUs with `-Oz` that are FullLTO'd together.

I'm trying to think of a way to provide a reduced example demonstrating the size increase, but it's proving to be pretty tricky. This is a pretty large FullLTO build from a mix of `-Oz` and `-O3` resources, and there seem to be interactions with other optimizations (in particular outlining) that are significant, so it'll take me a while to narrow things down.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D149768/new/

https://reviews.llvm.org/D149768