https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118862
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Yes. Also, looking at that commit
+ /* Anything ** 0 is one. */
+ if (tree_int_cst_sgn (rhs) == 0)
+ {
+ se->expr = build_int_cst (type, 1);
+ return 1;
+ }
+
+ if (!wi::fits_shwi_p (wrhs))
+ return 0;
+
+ n = wrhs.to_uhwi ();
The first check should be integer_zerop (rhs) rather than tree_int_cst_sgn
(rhs) == 0.
And, then it checks if wrhs fits into shwi and uses it as uhwi, that means it
will ICE if wrhs is negative. Either fits_uhwi_p + to_uhwi, or fits_shwi_p +
to_shwi, or just use wide_int.
