https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105747
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Version|unknown |13.0
--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed with a cross from x86_64:
alias stmt walking : 2.47 ( 27%)
tree DSE : 1.52 ( 17%)
scheduling : 1.36 ( 15%)
TOTAL : 9.15
so it's not only scheduling. perf (release checking) shows:
Samples: 37K of event 'cycles', Event count (approx.): 41984911906
Overhead Samples Command Shared Object Symbol
26.20% 9765 cc1 cc1 [.] get_ref_base_and_extent
4.37% 1635 cc1 cc1 [.] find_base_term
3.75% 1392 cc1 cc1 [.] rank_for_schedule
3.13% 1168 cc1 cc1 [.] wi::lshift_large
2.52% 938 cc1 cc1 [.] canonize
2.36% 876 cc1 cc1 [.] sd_lists_size
2.27% 844 cc1 cc1 [.]
sd_find_dep_between_no_cache
1.89% 701 cc1 cc1 [.]
operand_compare::operand_equal_p
1.80% 671 cc1 cc1 [.] wi::mul_internal
1.71% 637 cc1 cc1 [.]
simplify_context::simplify_plus_minus
1.63% 613 cc1 cc1 [.] memrefs_conflict_p
note with a non-optimized compiler and checking enabled (but -fno-checking)
things are much slower.
