[PATCH] D49388: [DAG] Fix Memory ordering check in ReduceLoadOpStore.

[Eli Friedman via Phabricator via llvm-commits]

efriedma added inline comments.


================
Comment at: llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp:13064
+    // expensive so we bound the search. To mitigate failures due to size of
+    // DAG, we also add LD's predecessors to the pruning list.
+
----------------
Instead of doing this predecessor check, could we just check that the load's chain has one use?  I guess that misses some cases that could be transformed, but I'd like to avoid these expensive walks as much as possible.


Repository:
  rL LLVM

https://reviews.llvm.org/D49388