aaupov wrote: > > For RSS, it's probably because in llvm-profgen, we decoded the pseudo-probe > > only for profiled function(https://reviews.llvm.org/D121643), so the > > baseline memory is decreased. > > > > @aaupov actually do we need to decode profile for functions without profile? > that is no profile to be matched in this case, and there is no > optimization/reorder to be done either, hence no need to rewrite probes?
Good point, I think we can selectively parse function records in perf2bolt - we don't need unused function records for profile matching. For partial rewrite though that would be quite a bit of work. https://github.com/llvm/llvm-project/pull/102904 _______________________________________________ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
