[SCEV] Don't perform implication checks with many predicates (#158652)

When adding a new predicate to a union, we currently do a bidirectional
implication for all the contained predicates. This means that the number
of implication checks is quadratic in the number of total predicates (if
they don't end up being eliminated).

Fix this by not checking for implication if the number of predicates
grows too large. The expectation is that if there is a large number of
predicates, we should be discarding them later anyway, as expanding them
would be too expensive.

Fixes https://github.com/llvm/llvm-project/issues/156114.

1 files changed, 7 insertions, 2 deletions

diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index a1703a2..9698ed8 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -15176,15 +15176,20 @@ void SCEVUnionPredicate::add(const SCEVPredicate *N, ScalarEvolution &SE) {
     return;
   }
 
+  // Implication checks are quadratic in the number of predicates. Stop doing
+  // them if there are many predicates, as they should be too expensive to use
+  // anyway at that point.
+  bool CheckImplies = Preds.size() < 16;
+
   // Only add predicate if it is not already implied by this union predicate.
-  if (implies(N, SE))
+  if (CheckImplies && implies(N, SE))
     return;
 
   // Build a new vector containing the current predicates, except the ones that
   // are implied by the new predicate N.
   SmallVector<const SCEVPredicate *> PrunedPreds;
   for (auto *P : Preds) {
-    if (N->implies(P, SE))
+    if (CheckImplies && N->implies(P, SE))
       continue;
     PrunedPreds.push_back(P);
   }