[llvm] r309662 - [PM] Add a comment clarifying what a particular predicate is doing.

[Chandler Carruth via llvm-commits]

Author: chandlerc
Date: Mon Jul 31 23:40:11 2017
New Revision: 309662

URL: http://llvm.org/viewvc/llvm-project?rev=309662&view=rev
Log:
[PM] Add a comment clarifying what a particular predicate is doing.

This came up as a point of confusion while working on a fundamental
problem with the combination of CGSCC iteration and the inliner.

Modified:
    llvm/trunk/lib/Analysis/CGSCCPassManager.cpp

Modified: llvm/trunk/lib/Analysis/CGSCCPassManager.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/CGSCCPassManager.cpp?rev=309662&r1=309661&r2=309662&view=diff
==============================================================================
--- llvm/trunk/lib/Analysis/CGSCCPassManager.cpp (original)
+++ llvm/trunk/lib/Analysis/CGSCCPassManager.cpp Mon Jul 31 23:40:11 2017
@@ -619,6 +619,14 @@ LazyCallGraph::SCC &llvm::updateCGAndAna
       AM.invalidate(*C, PA);
     }
     auto NewSCCIndex = RC->find(*C) - RC->begin();
+    // If we have actually moved an SCC to be topologically "below" the current
+    // one due to merging, we will need to revisit the current SCC after
+    // visiting those moved SCCs.
+    //
+    // It is critical that we *do not* revisit the current SCC unless we
+    // actually move SCCs in the process of merging because otherwise we may
+    // form a cycle where an SCC is split apart, merged, split, merged and so
+    // on infinitely.
     if (InitialSCCIndex < NewSCCIndex) {
       // Put our current SCC back onto the worklist as we'll visit other SCCs
       // that are now definitively ordered prior to the current one in the