[CUDA] Use the multi-element remove function in EraseUnwantedCUDAMatches.

[jlebar]

Summary

Bug pointed out by Benjamin Kramer in r264008. I think the bug is
benign because by the time this is called, we should only have at most
two overloads to consider (either a host and a device overload, or a
host+device overload, but not all three).

[tra]

LGTM.

cfe/trunk/lib/Sema/SemaCUDA.cpp

static void EraseUnwantedCUDAMatchesImpl(
   };
 
   // Find the best call preference among the functions in Matches.
   Sema::CUDAFunctionPreference BestCFP = GetCFP(*std::max_element(
       Matches.begin(), Matches.end(),
       [&](const T &M1, const T &M2) { return GetCFP(M1) < GetCFP(M2); }));
 
   // Erase all functions with lower priority.
-  Matches.erase(llvm::remove_if(
-      Matches, [&](const T &Match) { return GetCFP(Match) < BestCFP; }));
+  Matches.erase(
+      llvm::remove_if(Matches,
+                      [&](const T &Match) { return GetCFP(Match) < BestCFP; }),
+      Matches.end());
 }
 
 void Sema::EraseUnwantedCUDAMatches(const FunctionDecl *Caller,
                                     SmallVectorImpl<FunctionDecl *> &Matches){
   EraseUnwantedCUDAMatchesImpl<FunctionDecl *>(
       *this, Caller, Matches, [](const FunctionDecl *item) { return item; });
 }