[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 #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
Btw, should be reproducible with negative step cases on x86_64 as well, as we
do
there

  if (negative)
    offset = size_int (-TYPE_VECTOR_SUBPARTS (vectype) + 1);

?

I suppose interpreting offset in terms of scalar element size is what was
intended.