[LLVMdev] LICM/store-aliasing of global loads

[Dan Gohman]

On Jul 22, 2008, at 1:44 PM, Duncan Sands wrote:

> Hi,
>
>> One way to fix this would be to have AliasSetTracker pretend that
>> pointers to constant memory never alias anything. That's a little
>> sneaky though, ...
>
> on the other hand it is simple and (presumably) effective.
> Do you think it really could cause trouble?

Right now, LICM is the only user of the AliasSetTracker abstraction.
For the immediate problem, the fix in LICM looks simpler :-).

For the future, it is theoretically possible to write an AliasSetTracker
client that would break if given pretended results. Unless there's a
need, I'd prefer to avoid it.

Dan