[Bug tree-optimization/127966] [15/16/17 Regression] wrong code at -O1 and above with a loop over string literal bytes >= 0x80

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127966

--- Comment #2 from Drea Pinski <pinskia at gcc dot gnu.org> ---
I think the problem is here: tree-ssa-loop-niter.cc:3738 :
                      /* iv.base can be &"Foo" but also (char *)&"Foo" + 1.  */
                      split_constant_offset (iv.base, &str, &off);
                      STRIP_NOPS (str);
                      if (TREE_CODE (str) == ADDR_EXPR
                          && TREE_CODE (TREE_OPERAND (str, 0)) == STRING_CST
                          && tree_fits_uhwi_p (off))
                        {
                          str = TREE_OPERAND (str, 0);
                          unsigned i = 0;
                          for (unsigned HOST_WIDE_INT idx
                               = (tree_to_uhwi (TREE_OPERAND (mem, 1))
                                  + tree_to_uhwi (off));
                               idx < (unsigned)TREE_STRING_LENGTH (str)
                               && i < MAX_ITERATIONS_TO_TRACK;
                               idx += tree_to_uhwi (iv.step), ++i)
                            {
                              int res = compare_tree_int
                                (op[1], TREE_STRING_POINTER (str)[idx]);
                              if ((cmp == NE_EXPR && res == 0)
                                  || (cmp == EQ_EXPR && res != 0))
                                return build_int_cst (unsigned_type_node, i);
                            }
                        }


Which does not take into account the different types for the MEM_REF vs
STRING_CST.