[PATCH] D23432: [AliasSetTracker] Degrade AliasSetTracker results when may-alias sets get too large.

[Philip Reames via llvm-commits]

On 08/20/2016 12:59 PM, Daniel Berlin wrote:
> Even if you kept the ASTracker concept for some reason, moving the 
> ASTracker to MemorySSA also would make it scale significantly better 
> (let's ignore the updating through IR changes part for a second, as 
> it's orthogonal)
>
> For MemorySSA, ASTracker stores only clobbering MemoryAccesses as a 
> SmallPtrSet.
>
> For adding a load, you add MSSA->getClobberingMemoryAccess(load) (this 
> is only necessary if memoryssa gave up optimizing uses, we could fix 
> this. Otherwise it's just the defining access of the load)
> For adding a store, you add the store.
>
> For merging, you merge the sets.
> For querying for loads, you see if 
> MSSA->getClobberingMemoryAccess(load) is in the set (ditto the above)
> for querying for stores, you see if the store, or 
>  MSSA->getClobberingMemoryAccess(store) is in the set.
>
> This will scale at least a factor of N better.
You seem to have interpreted my comment as arguing against MemorySSA.  
That was not my intent at all.  If we have a realistic chance of porting 
to MemorySSA in the near term (and from your other response, it sounds 
like we do), then that's definitely the right direction to move in.
>
> On Sat, Aug 20, 2016 at 12:38 PM, Hal Finkel <hfinkel at anl.gov 
> <mailto:hfinkel at anl.gov>> wrote:
>
>     hfinkel added a comment.
>
>     In https://reviews.llvm.org/D23432#521624
>     <https://reviews.llvm.org/D23432#521624>, @reames wrote:
>
>     > I am not actively objecting to this patch, but I really don't
>     like the overall direction here.  Having a threshold where our
>     ability to optimize falls off a cliff just seems really
>     undesirable.  As Hal pointed out, there are likely options for
>     summarizing alias sets to allow quicker AA queries.  How much have
>     we explored that design space?
>
>
>     My understanding from the discussion is that all uses of the
>     ASTracker are going to be replaced with MemorySSA-based
>     algorithms; that is why I was okay with this (for now). If we
>     still need an AST concept, then we'll want to do something more
>     sophisticated.
>
>
>     Repository:
>       rL LLVM
>
>     https://reviews.llvm.org/D23432 <https://reviews.llvm.org/D23432>
>
>
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160820/09e60628/attachment.html>