llvm dev - Sep 2013 - [LLVMdev] Semantics of noalias with vector of pointers?

Hi,

What are the intended semantics of noalias when applied to a vector of 
pointers? signext/zeroext parameter attributes are incorrectly rejected 
for vectors (bug 15391), and I'm not sure if the same is true for 
noalias when applied to pointer vectors. There's no mention of it in the 
langref, and the verifier currently rejects it. Is it supposed to work? 
Are pointers within the vector allowed to alias other elements in the 
vector?

Do you actually have a use case for such attributes, or are you just trying
all possible attributes on all possible types to see what breaks? ;-)

It would be theoretically possible to define semantics for signext/zerext
or even noalias for vectors. However it's not clear whether those are
really desirable without motivating use cases.

Dan


On Wed, Sep 4, 2013 at 2:31 PM, Matt Arsenault <Matthew.Arsenault at
amd.com>wrote:
> Hi,
>
> What are the intended semantics of noalias when applied to a vector of
> pointers? signext/zeroext parameter attributes are incorrectly rejected for
> vectors (bug 15391), and I'm not sure if the same is true for noalias
when
> applied to pointer vectors. There's no mention of it in the langref,
and
> the verifier currently rejects it. Is it supposed to work? Are pointers
> within the vector allowed to alias other elements in the vector?
>
> ______________________________**_________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>
http://lists.cs.uiuc.edu/**mailman/listinfo/llvmdev<http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130904/67bd76f5/attachment.html>

On 09/04/2013 03:16 PM, Dan Gohman wrote:
> Do you actually have a use case for such attributes, or are you just 
> trying all possible attributes on all possible types to see what 
> breaks? ;-)
I'm mostly just breaking things. I randomly ended up on the bug about 
the zext vector parameters and thought it should work.
>
> It would be theoretically possible to define semantics for 
> signext/zerext or even noalias for vectors. However it's not clear 
> whether those are really desirable without motivating use cases.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130904/c91991f0/attachment.html>