[llvm] [ctxprof] don't inline weak symbols after instrumentation (PR #128811)

Mircea Trofin via llvm-commits llvm-commits at lists.llvm.org
Wed Feb 26 07:18:43 PST 2025


mtrofin wrote:

It doesn't appear to take a hit. Not completely surprising: we already do some inlining before instrumentation.

If we ever want to avoid taking any hit, we can subsequently add some metadata to each such function capturing the original (if any) inline attribute; go through thinlto; then restore the original attribute from metadata after function import and let the backend inliner do its job (which, at this point, would be only working with prevailing copies)

https://github.com/llvm/llvm-project/pull/128811


More information about the llvm-commits mailing list