[LLVMdev] possible addrspacecast problem

[David Chisnall]

On 18 Mar 2015, at 18:20, Sanjoy Das <sanjoy at playingwithpointers.com> wrote:
> 
>> but I'm happy to prepare a test for upstreaming if others agree with this assessment.
> 
> I think it is very valuable to have that upstreamed.  I think we
> should also add other general test cases that show what we definitely
> cannot do with addrspacecasts.

Here's the commit in our tree:

https://github.com/CTSRD-CHERI/llvm/commit/751a6c89afd47ee8e7f0c6c3ec3ef7bc1f057806

I'll rebase on the latest LLVM soon and upload a diff to phabricator, but I'd appreciate early feedback if anyone wants to take a look.  In particular, we seem to have tests that explicitly check for the old behaviour, but have no comments indicating why.  If anyone depends on GEPs being reordered across AS casts then now would be a good time to speak up...

David