https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109983
--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> --- Unsurprisingly, for the full testcase on current trunk with release checking: ipa points-to :1080.80 (100%) 0.56 ( 47%)1081.52 (100%) 9739k ( 4%) TOTAL :1083.27 1.20 1084.62 261M 1083.27user 1.21system 18:04.64elapsed 99%CPU (0avgtext+0avgdata 2778064maxresident)k 0inputs+9168outputs (0major+712792minor)pagefaults 0swaps Note there are PRs showing PTA itself is slow for large functions, IPA PTA isn't very much special here. The bitmap data structure isn't the main issue, the main issue is the working set becomes too big and the iteration isn't optimized for memory locality. The biggest hits tends to be the bitmap_ior_into done in solve_add_graph_edge which is necessary to support the solver working on incremental solution changes only (we only have 'changed' on nodes, not on graph edges which are only represented implicitly).
