[PATCH] D53106: [SelectionDAG] Fix behavior of glued nodes in hasPredecessorHelper.

[Tim Neumann via Phabricator via llvm-commits]

TimNN added a comment.

The approach in general looks good to me, however I don't think this interacts correctly with the `TopologicalPrune` optimization.

(At this time, I'm not familiar enough with the code to intuitively know in which direction the topological sorting is used, but one of the two scenarios below //will// apply).

Assume that the optimization limits the nodes to be consider to the range [`a`, `b`].

Scenario 1) A non-glue edge goes past `a` to a node `n`. Since `n` is outside the range considered, it won't be processed even though it may have reverse glue edges back into the range.

Scenario 2) A reverse glue edge goes past `b` to a node `m`. Since `m` is outside the range considered, it won't be processed even though it may have normal edges back into the range.


Repository:
  rL LLVM

https://reviews.llvm.org/D53106