https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127673
avieira at gcc dot gnu.org changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |ASSIGNED
CC| |avieira at gcc dot gnu.org
Ever confirmed|0 |1
Assignee|unassigned at gcc dot gnu.org |avieira at gcc dot
gnu.org
Last reconfirmed| |2026-10-03
--- Comment #7 from avieira at gcc dot gnu.org ---
Hi Hongtao,
I had a look at this and I'm 99% sure this is a backend issue with the code
that checks whether it's OK to use the tail predicated loops. Specifically in
that we should be checking that the number of original iterations and the input
to the tail predication instruction that counts the elements computed should be
equivalent. They aren't for this loop, and it should not be transformed, but it
doesn't seem to be realizing that.
I'll take this.
FYI for anyone being hit by this issue, you can work around it by using
-mno-dlstp
