https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92862
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2019-12-09
CC| |hubicka at gcc dot gnu.org
Ever confirmed|0 |1
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Feng Xue from comment #0)
> Function loop_only_exit_p() in tree-ssa-loop-niter.c:
>
> for (i = 0; i < loop->num_nodes; i++)
> {
> for (bsi = gsi_start_bb (body[i]); !gsi_end_p (bsi); gsi_next (&bsi))
> if (stmt_can_terminate_bb_p (gsi_stmt (bsi)))
> {
> return true;
> }
> }
>
> we should return false, not true?
Indeed... A patch fixing this is pre-approved (and obvious).
>
> Function predict_paths_leading_to_edge() in predict.c:
>
> FOR_EACH_EDGE (e2, ei, bb->succs)
> if (e2->dest != e->src && e2->dest != e->dest
> && !unlikely_executed_edge_p (e)
> && !dominated_by_p (CDI_POST_DOMINATORS, e->src, e2->dest))
> {
> has_nonloop_edge = true;
> break;
> }
>
> "e" is loop invariant, I guess that unlikely_executed_edge_p (e) might be
> unlikely_executed_edge_p (e2), which is more reasonable. And we can find
> similar code in predict_paths_for_bb () that uses the latter.
CCing Honza for the above.
