[Bug tree-optimization/63341] [4.8/4.9/5 Regression] Vectorization miscompilation with -mcpu=power7

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63341

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
Note that vect_create_addr_base_for_vector_ref _does_ multiply offset by
the scalar element size (even if it calls it 'step'):

  tree step = TYPE_SIZE_UNIT (TREE_TYPE (DR_REF (dr)));
...
  if (offset)
    {
      offset = fold_build2 (MULT_EXPR, sizetype,
                            fold_convert (sizetype, offset), step);

so that would be consistent with the interpretation and all uses I can see.