[llvm-dev] Optimizing diamond pattern in DAGCombine

[Nirav Davé via llvm-dev]

This is a little hard to diagnose in the abstract, but it sounds like
you're having a loop along the lines of visit A -> visit B/C -> visit D ->
visit A and that at each step you're making a real reasonable change to the
DAG and returning to the original DAG.

One possible solution is to to rework the optimization to occur when
visiting A not D. so the last edge in the visit loop no longer happens.

If that's not feasible, I don't think there's much you can do that is not
effectively preventing one of those transitions from occurring when it
would cause the infinite visit loop.

HTH,

-Nirav


On Mon, May 22, 2017 at 2:21 AM, Amaury SECHET via llvm-dev <
llvm-dev at lists.llvm.org> wrote:

> I'm trying to optimize a pattern that goes roughly as:
>     A
>    / \
>   B   C
>    \ /
>     D
>
> Problem is, when A gets modified, B and C get added back to the worklist,
> but D doesn't. Readding D to the worklist just create an infinite loop
> where one process D again and again.
>
> Is there a proper way to make this work ?
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20170522/c5bcccb8/attachment.html>