Hello, On Fri, Jan 21 2022, Aldy Hernandez via Gcc-patches wrote: > As discussed in PR103721, the problem here is that we are crossing a > backedge and causing us to use relations from a previous iteration of a > loop. > > This handles the testcases in both PR103721 and PR104067 which are variants > of the same thing. > > Tested on x86-64 Linux with the usual regstrap as well as verifying the > thread count before and after the patch. The number of threads is > reduced by a miniscule amount. > > I assume we need release manager approval at this point? OK for trunk? > > gcc/ChangeLog: > > PR 103721/tree-optimization > * gimple-range-path.cc > (path_range_query::relations_may_be_invalidated): New. > (path_range_query::compute_ranges_in_block): Reset relations if > they may be invalidated. > (path_range_query::maybe_register_phi_relation): Exit if relations > may be invalidated on incoming edge. > (path_range_query::compute_phi_relations): Pass incoming PHI edge > to maybe_register_phi_relation. > * gimple-range-path.h (relations_may_be_invalidated): New. > (maybe_register_phi_relation): Pass edge instead of tree. > * tree-ssa-threadbackward.cc (back_threader::back_threader): > * value-relation.cc (path_oracle::path_oracle): Call > mark_dfs_back_edges. > (path_oracle::register_relation): Add SSA names to m_registered > bitmap. > (path_oracle::reset_path): Clear m_registered bitmap. > * value-relation.h (path_oracle::set_root_oracle): New.
this has caused around 5% regression of 429.mcf when built with -O2 -flto (generic march) on x86_64 (I tried and confirmed on AMD Zen3, Zen2 and Intel Cascadelake), the two former cases can be seen here: https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=469.60.0 https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=413.60.0&plot.1=292.60.0&; This does not seem to be a regression against gcc11 and I am not sure whether it is worth a bug-report, but perhaps it is worth looking at as it may indicate where we can improve further? Martin
