llvm dev - May 2015 - [LLVMdev] API for flow/context -sensitive alias analysis?

I'm starting to play around with flow- and context- sensitive AA
implementations.  I thought it would be interesting to try fitting one into
LLVM's AA framework.

Should I assume that to support flow-/context-sensitive AA queries, I'll
need to define my own AA API, rather than supporting
"llvm::AliasAnalysis"?

The AA doc ( http://llvm.org/docs/AliasAnalysis.html ) seems to indicate
that the API (or its implementations?) don't support flow-sensitive
analysis.  I also don't see any obvious way for the API to let clients ask
context-sensitive questions.

I also looked at the poolalloc docs (Section 4 of those doc's Latex DSA
manual), and my impression is that they too ended up providing their own
API rather than implementing "llvm::AliasAnalysis".

Thanks,
Christian
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20150519/7b45178b/attachment.html>

On Tue, May 19, 2015 at 9:42 AM, Christian Convey
<christian.convey at gmail.com> wrote:
> I'm starting to play around with flow- and context- sensitive AA
> implementations.  I thought it would be interesting to try fitting one into
> LLVM's AA framework.
George, cc'd, is starting to explore this stuff for CFL.
>
> Should I assume that to support flow-/context-sensitive AA queries,
I'll
> need to define my own AA API, rather than supporting
"llvm::AliasAnalysis"?
Yes.  It would be interesting to see what you come up with.
>
> The AA doc ( http://llvm.org/docs/AliasAnalysis.html ) seems to indicate
> that the API (or its implementations?) don't support flow-sensitive
> analysis.  I also don't see any obvious way for the API to let clients
ask
> context-sensitive questions.
You cannot :)
>
> I also looked at the poolalloc docs (Section 4 of those doc's Latex DSA
> manual), and my impression is that they too ended up providing their own
API
> rather than implementing "llvm::AliasAnalysis".

It's hard to design a generally useful compiler API without
clients/providers.   :)