https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114589
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |rguenth at gcc dot gnu.org
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
This is the "old" issue with the bogus
/* If BEST_BB is at the same nesting level, then require it to have
significantly lower execution frequency to avoid gratuitous movement. */
if (bb_loop_depth (best_bb) == bb_loop_depth (early_bb)
/* If result of comparsion is unknown, prefer EARLY_BB.
Thus use !(...>=..) rather than (...<...) */
&& !(best_bb->count * 100 >= early_bb->count * threshold))
return best_bb;
check. We have a BB2 count of 118111600 and BB5 we'd sink to has 105119324
which is higher than 75% of BB2.
This was supposed to prevent sinking across if (x) unlikely (); blocks
but it has odd side-effects like this one.
IIRC there's duplicates for this.
I'll note that we probably want to avoid sinking to a post-dominator
unless it sinks out of a loop (that's cross-BB scheduling). But we no
longer compute post-dominators. Without re-introducing those and
limiting sinking this way removing the odd code will cause regressions.
