[PATCH] Fix always creating GEP with i32 indices

[Matt Arsenault]

On 08/12/2013 06:12 PM, Eli Friedman wrote:
> I'm not sure your test is actually
> testing anything useful given the current architecture of instcombine;
> even if we build a GEP with wrong-sized indices, the GEP visitor will
> fix it.
>
I wasn't thinking of that. It will just be an added bonus then, since 
that test doesn't already have a test for non-zero arrays