[llvm] [SimplifyCFG] Avoid threading for loop headers (PR #151142)

[Nikita Popov via llvm-commits]

================
@@ -3604,15 +3605,23 @@ foldCondBranchOnValueKnownInPredecessorImpl(BranchInst *BI, DomTreeUpdater *DTU,
   return false;
 }
 
-static bool foldCondBranchOnValueKnownInPredecessor(BranchInst *BI,
-                                                    DomTreeUpdater *DTU,
-                                                    const DataLayout &DL,
-                                                    AssumptionCache *AC) {
+#include <iostream>
+
+bool SimplifyCFGOpt::foldCondBranchOnValueKnownInPredecessor(BranchInst *BI) {
+  // Note: If BB is a loop header then there is a risk that threading introduces
+  // a non-canonical loop by moving a back edge. So we avoid this optimization
+  // for loop headers if NeedCanonicalLoop is set.
+  bool InHeader = is_contained(LoopHeaders, BI->getParent());
----------------
nikic wrote:

Avoid the variable and move this behind the `&&` for short-circuiting. No need to do this linear scan if !NeedCanonicalLoop.

https://github.com/llvm/llvm-project/pull/151142