Hi, I'm re-posting patches which I have posted at the end of stage 1 but which have not passed review yet.
8<-------------------------------------------------------------------- Looking at some benchmarks I have noticed many cases when IPA-CP cloned a function for all contexts just because it knew that some parameters were not used at all. Then IPA-SRA looked at the function and cloned it again to split another parameter or two. The latter pass is better equipped to detect when parameters can be altogether removed and so the IPA-CP cloning was for no good reason. This patch simply alters the IPA-CP not to do that in the situations where IPA-SRA can (for nodes which can be made local). Bootstrapped and tested individually when I originally posted it and now bootstrapped and LTO-bootstrapped and tested as part of the whole series. OK for master? gcc/ChangeLog: 2022-11-11 Martin Jambor <mjam...@suse.cz> * ipa-cp.cc (estimate_local_effects): Do not clone potentionally local nodes for all contexts just to remove unused parameters. --- gcc/ipa-cp.cc | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/gcc/ipa-cp.cc b/gcc/ipa-cp.cc index cc031ebed0f..172ea353c49 100644 --- a/gcc/ipa-cp.cc +++ b/gcc/ipa-cp.cc @@ -3722,7 +3722,10 @@ estimate_local_effects (struct cgraph_node *node) &removable_params_cost); int devirt_bonus = devirtualization_time_bonus (node, &avals); if (always_const || devirt_bonus - || (removable_params_cost && node->can_change_signature)) + || (removable_params_cost + && node->can_change_signature + /* If IPA-SRA can do it, it can do it better. */ + && !node->can_be_local_p ())) { struct caller_statistics stats; ipa_call_estimates estimates; -- 2.38.1