[Bug tree-optimization/126404] [16/17 Regression] ICE at -O2/3 on aarch64 during GIMPLE pass: vect

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126404

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
OK, so the issue is that we have 

 # _14 = PHI <_18, ..>
 _16 = __builtin_ctz (_14);
 _18 = (unsigned int) _16;

so we skip the conversion, but that has also been consumed by the pattern
which now has the reduction, and the next stmt is the PHI so we skip the
actual reduction stmt.

      if (CONVERT_EXPR_CODE_P (op.code)
          && tree_nop_conversion_p (op.type, TREE_TYPE (op.ops[0]))
          && (first
              || is_a <gphi *> (STMT_VINFO_STMT (next_stmt))))
        ;

Hmm.

I have a fix I think.