wlei-llvm wrote:
> > To double-check: did you test with or without dwarf-correlation? I tested > > once with it, expectedly pseudo probe parsing wasn't engaged, so there was > > no effect. > > Yes, I double-checked it's a pseudo-probed binary not dwarf-correlation. Note > that for end to end test, the pseudo-probe decoding is not the > bottleneck(only account for < 1% run time), so neutral for run time is kind > of expected. I was to test if the Address2ProbesMap `find` could regress the > run time. 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. In https://github.com/llvm/llvm-project/pull/102789, I tested for pseudo-probe decoding for all functions. 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
