https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122635

--- Comment #14 from Greg McGary <gkm at rivosinc dot com> ---
(In reply to Jeffrey A. Law from comment #13)
> I would kind of expect that DCE would clean this up if you've neutered the
> main operation.  I'd look at the next DCE dump and see if everything you
> want magically disappeared.

Neutering MASK_LEN_LOAD in 189t.cunroll leaves dangling references which cause
get_rank() in 204t.reassoc to SEGV, long before the next DCE at 217t.dce7.

Should I ...
* take greater care in CUNROLL to avoid generating the zero-length load/stores
in the first place?
* insert another DCE pass soon after cunroll?
* add checks to REASSOC (and others) to tolerate dangling references until the
next DCE can toss them?

Reply via email to