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

[Christian Convey]

On Mon, Jun 15, 2015 at 6:03 PM, Daniel Berlin <dberlin at dberlin.org> wrote:

> > From a personal perspective, I'm particularly interested in the maximum
> > analytic precision each AA approach can take, almost without regard to
> how
> > much time or memory the computation takes to run.
>
> I'm wildly curious why.
>

One reason is that I'm simply curious about the trade-off space between AA
precision and AA running-time.  Since I don't make compilers for a living,
I have the luxury of staring at corners of the design space which would be
idiotic to actually include in LLVM :)

Another reason is that in past work, I've sometimes worked on code where
I'd gladly accept a 10-day compilation time, if it bought me an extra 10%
in performance.  So I sometimes wonder what's involved in making such
optimization passes available, even though people wouldn't want to use them
on a day-to-day basis.

Constraints contain stride info, and intersection is then used to
> compute what sub-variables a constraint touches during solving.
> It is a variant of this:
> http://homepages.mcs.vuw.ac.nz/~djp/files/paste04.ps


Thanks, I'll try to give it a read.


> Some programs have so many sub-variables you will run out of memory.
> This is true in the location set approach as well.
>

I'm surprised you ran into that kind of trouble with memory.  Was that back
in 32-bit days?


> GCC only created variables for possibly-used field ranges too.
> I can point you at GCC bugs offline if you are interested
>

Yeah, if you don't mind I'd be grateful for the links.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20150615/a3168783/attachment.html>