[Bug tree-optimization/125602] [17 Regression] ICE at -O2 -fsanitize=undefined during GIMPLE pass: evrp

[rguenth at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125602

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2026-06-05
             Status|UNCONFIRMED                 |ASSIGNED
     Ever confirmed|0                           |1
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot 
gnu.org

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Looks like a latent issue though:

  /* {base, -C} < n.  */
  else if (tree_int_cst_sign_bit (iv0->step) && integer_zerop (iv1->step))
    {
      step = fold_build1 (NEGATE_EXPR, TREE_TYPE (iv0->step), iv0->step);
      /* MAX - C + 1 >= n.  */ 
      tree last = wide_int_to_tree (type, max - wi::to_wide (step) + 1);
      assumptions = fold_build2 (GE_EXPR, boolean_type_node, last, iv1->base);
      if (integer_zerop (assumptions))
        return false;

here step == -128 signed char.  I recently fixed some other places like this,
but here we have to question correctness as well.

I'll note the "opposite" case does

  /* n < {base, C}. */
  if (integer_zerop (iv0->step) && !tree_int_cst_sign_bit (iv1->step))
    {     
      step = iv1->step;
      /* MIN + C - 1 <= n.  */
      tree last = wide_int_to_tree (type, min + wi::to_wide (step) - 1);
      assumptions = fold_build2 (LE_EXPR, boolean_type_node, last, iv0->base);

so completely side-steps GENERIC folding with respect to overflow.  Iff
that's OK there it should be OK in the other case as well ...(?)

Also, we're doing max - wi::to_wide (step) + 1, it's tempting to just
drop the negation and use max + wi::to_wide (step) + 1 ...

I'm testing that.