https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81958
Jeffrey A. Law <law at redhat dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2017-09-30
CC| |law at redhat dot com
Blocks| |19794
Ever confirmed|0 |1
--- Comment #1 from Jeffrey A. Law <law at redhat dot com> ---
So I see a couple things here.
In DOM we do catch the fact that eqidx will have the value zero if the loop
does not iterate. But the equivalence we record is actually eqidx >= maxidx.
That's because we don't cprop into the conditionals in
record_temporary_equivalences. We *should* do that, but we have to be careful
about structure sharing and I'd like to avoid creating a lot of unnecessary
tree nodes if it can be avoided.
If we were to fix that, we'd end recording maxidx <= 0. Given that maxidx is
unsigned that simplifies into maxidx == 0. Of course we don't those kind of
simplifications when recording the temporary equivalences either and the same
concerns WRT structure sharing come into play.
When things simplify into something like maxidx == 0, instead of recording a
conditional into the equivalency tables, we should record that maxidx = 0 in
the const/copies table.
If we were to fix those 3 issues, then I think we have all the key pieces in
place to see the jump thread. But we still have to be concerned that jump
threading may be suppressed because of its conservatism WRT not clobbering the
loop structure. It should be safe in this case, but I don't offhand know if
the suppression code would kick in or not.
So, anyway, there's several TODO items in here.
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=19794
[Bug 19794] [meta-bug] Jump threading related bugs
