[llvm-commits] [llvm] r101658 - /llvm/trunk/lib/Analysis/InlineCost.cpp

[Chris Lattner]

Author: lattner
Date: Sat Apr 17 12:57:56 2010
New Revision: 101658

URL: http://llvm.org/viewvc/llvm-project?rev=101658&view=rev
Log:
fix PR6858: a dangling pointer use bug which was caused
by switching CachedFunctionInfo from a std::map to a 
ValueMap (which is implemented in terms of a DenseMap).

DenseMap has different iterator invalidation semantics
than std::map.

This should hopefully fix the dragonegg builder.

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

Modified: llvm/trunk/lib/Analysis/InlineCost.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/InlineCost.cpp?rev=101658&r1=101657&r2=101658&view=diff
==============================================================================
--- llvm/trunk/lib/Analysis/InlineCost.cpp (original)
+++ llvm/trunk/lib/Analysis/InlineCost.cpp Sat Apr 17 12:57:56 2010
@@ -319,8 +319,13 @@
     FunctionInfo &CallerFI = CachedFunctionInfo[Caller];
 
     // If we haven't calculated this information yet, do so now.
-    if (CallerFI.Metrics.NumBlocks == 0)
+    if (CallerFI.Metrics.NumBlocks == 0) {
       CallerFI.analyzeFunction(Caller);
+     
+      // Recompute the CalleeFI pointer, getting Caller could have invalidated
+      // it.
+      CalleeFI = &CachedFunctionInfo[Callee];
+    }
 
     // Don't inline a callee with dynamic alloca into a caller without them.
     // Functions containing dynamic alloca's are inefficient in various ways;
@@ -426,6 +431,8 @@
     return;
   }
   
+  // Since CalleeMetrics were already calculated, we know that the CallerMetrics
+  // reference isn't invalidated: both were in the DenseMap.  
   CallerMetrics.NeverInline |= CalleeMetrics.NeverInline;
   CallerMetrics.usesDynamicAlloca |= CalleeMetrics.usesDynamicAlloca;