https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87609
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Priority|P3 |P2
Status|UNCONFIRMED |ASSIGNED
Known to work| |4.8.5
Keywords| |wrong-code
Last reconfirmed| |2018-10-15
CC| |amker at gcc dot gnu.org,
| |rguenth at gcc dot gnu.org
Assignee|unassigned at gcc dot gnu.org |rguenth at gcc dot
gnu.org
Ever confirmed|0 |1
Summary|miscompilation with |[6/7/8/9 Regression]
|restrict and loop |miscompilation with
| |restrict and loop
Target Milestone|--- |6.5
Known to fail| |5.5.0
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed. GCC 5 and 6 output
7 7
even. GCC 4.8 works.
This is is GIMPLE level unrolling exposing non-dependence between iterations
where that doesn't actually hold. We've talked about this issue when
trying to expose to later passes the result of runtime dependence checks
but it of course also exists when the non-dependence is recorded via
restrict.
We have to treat unrolling similar to inlining and re-map the base/clique
sets accordingly. Versioning should be fine(?).
I'll see if I can find time to address this in our generic infrastructure.
