[patch,wip] warn on noncontiguous pointers

[Thomas Koenig]

Hi Cesar,

> How's this patch look? I bootstrapped and regression tested it for
> x86_64 Linux with nvptx offloading (not that it exercised any offloading
> functionality).
> 
> Is it OK for trunk?

OK.

Maybe we should one day revisit this one day and separate out the cases
where we can prove that the array cannot be contiguous vs. the ones
where we cannot prove either way.

For now, this fixes a rejects-valid.

Thanks!

Regards

	Thomas