[LLVMdev] Expressing ambiguous points-to info in AliasAnalysis::alias(...) results?

[Daniel Berlin]

On Mon, Jun 15, 2015 at 1:46 PM, Christian Convey
<christian.convey at gmail.com> wrote:
> On Mon, Jun 15, 2015 at 3:29 PM, Daniel Berlin <dberlin at dberlin.org> wrote:
>>
>> On Mon, Jun 15, 2015 at 10:33 AM, Christian Convey
>> <christian.convey at gmail.com> wrote:
>> > On Mon, Jun 15, 2015 at 11:02 AM, Daniel Berlin <dberlin at dberlin.org>
>> > wrote:
>> Which paper are you using?
>
>
> I'm mostly going from Robert Wilson's 1997 phd thesis, although I'm pretty
> sure I've seen a lot of the same ideas elsewhere as well.

Yes, using summary/transfer functions has been tried a lot.
Note: the numbers in most of these phd thesis do *not* get born out in practice.

See, e.g, http://www.cse.unsw.edu.au/~ysui/papers/spe14.pdf

vs

http://impact.crhc.illinois.edu/shared/report/phd-thesis-erik-nystrom.pdf

vs
the wilson thesis.


In particular, i expect the wilson algorithm is going to fall over
very badly on anything of even medium size.

Nowadays, the thing of the day seems to be using cfl reachability
based formulations and doing context-sensitivity on demand.