> On Mon, 2020-01-06 at 15:08 +0100, Martin Liška wrote: > > Hi. > > > > As Honza noticed in the PR, we are quite strict about TOP N > > counter invalidation due to multiple values that can't > > fit in a counter. We due it in order to have a reproducible > > builds. I guess we should do a compromise in between reproducibility > > and possible speed up. That's why I'm suggesting to invalidate > > a TOP N counter only if param_profile_topn_invalid_threshold percent > > of profile are missing. > > > > Patch can bootstrap on x86_64-linux-gnu and survives regression tests. > > > > Ready to be installed? > > Thanks, > > Martin > > > > gcc/ChangeLog: > > > > 2020-01-06 Martin Liska <mli...@suse.cz> > > > > PR tree-optimization/92924 > > * params.opt (param_profile_topn_invalid_threshold): > > New parameter. > > * profile.c (sort_hist_values): Mark TOP N counter > > invalid only if significant amount of samples > > is missing. > > > > gcc/testsuite/ChangeLog: > > > > 2020-01-06 Martin Liska <mli...@suse.cz> > > > > PR tree-optimization/92924 > > * gcc.dg/tree-prof/pr92924-2.c: New test. > > * gcc.dg/tree-prof/pr92924.c: New test. > > > > libgcc/ChangeLog: > > > > 2020-01-06 Martin Liska <mli...@suse.cz> > > > > PR tree-optimization/92924 > > * libgcov-merge.c (merge_topn_values_set): Replace > > value with lowest count. > OK Actually I am not so sure about this patch - how do we ensure reproducibility in this case?
Honza
