https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113373
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Assignee|unassigned at gcc dot gnu.org |rguenth at gcc dot
gnu.org
Status|NEW |ASSIGNED
--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
All good after vect_do_peeling. It's wrecked by
vect_create_epilog_for_reduction here:
scalar_result = scalar_results[k];
FOR_EACH_IMM_USE_STMT (use_stmt, imm_iter, orig_name)
{
FOR_EACH_IMM_USE_ON_STMT (use_p, imm_iter)
SET_USE (use_p, scalar_result);
update_stmt (use_stmt);
we have two reduction epilogues, one for the early and one for the late
exit but we mismatch the scalar use to replace.
We're also picking up a reduction that's only live on one edge on the other
(but that doesn't seem to result in a problem here).
It also seems we're doing epilogue code generation before creating some
exit PHIs required for that - vect_do_peeling fails to do that.
We can try to hack around that a bit, but there's more issues with this
testcase, like we generate useless epilogues - it seems we should
identify the scalar exit PHI to transform when calling epilogue generation
for a specific edge and skip if there isn't one.
