[PATCH] Allow global structures to be vectorized
Renato Golin
renato.golin at linaro.org
Wed Feb 6 10:11:52 PST 2013
On 6 February 2013 17:42, Hal Finkel <hfinkel at anl.gov> wrote:
> If I'm reading this correctly, this will also handle cases like
> for (i = 0; i < n; ++i)
> x[i+n] = x[i];
> If not, why not?
>
It should, though not at the current stage the patch is.
> alias only if x ==z and y ==a.
>
> This was the point I had tried to make previously; You might be able to
> use SE to prove disjointness for affine accesses, but in general you'll
> need runtime overlap checks.
>
Yes, I'm aware of that. Again, not yet thinking about it.
I want first to catch the static cases (when the offsets are all
compile-time constants) and only then, when that's proven to be correct,
I'll extend to run-time checks.
cheers,
--renato
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20130206/53db350b/attachment.html>
More information about the llvm-commits
mailing list