Ease contention on this lock by noticing that all writes to the VTs array will

be of (dynamically) constant values, so races on it are immaterial.  We just need
to ensure that at least one write has completed before return the pointer into it.

With this change, parllc exhibits essentially no overhead on 403.gcc.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79708 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 5 insertions, 2 deletions

diff --git a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index f5de81b..91f7914 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -5022,12 +5022,15 @@ static ManagedStatic<sys::SmartMutex<true> > VTMutex;
 /// getValueTypeList - Return a pointer to the specified value type.
 ///
 const EVT *SDNode::getValueTypeList(EVT VT) {
-  sys::SmartScopedLock<true> Lock(*VTMutex);
   if (VT.isExtended()) {
     return &(*EVTs->insert(VT).first);
   } else {
+    // All writes to this location will have the same value, so it's ok
+    // to race on it.  We only need to ensure that at least one write has
+    // succeeded before we return the pointer into the array.
     VTs[VT.getSimpleVT().SimpleTy] = VT;
-    return &VTs[VT.getSimpleVT().SimpleTy];
+    sys::MemoryFence();
+    return VTs + VT.getSimpleVT().SimpleTy;
   }
 }