diff --git a/llvm/include/llvm/CodeGen/SlotIndexes.h b/llvm/include/llvm/CodeGen/SlotIndexes.h
--- a/llvm/include/llvm/CodeGen/SlotIndexes.h
+++ b/llvm/include/llvm/CodeGen/SlotIndexes.h
@@ -215,8 +215,12 @@
     }
 
     /// Return the scaled distance from this index to the given one, where all
-    /// slots on the same instruction have zero distance.
-    int getInstrDistance(SlotIndex other) const {
+    /// slots on the same instruction have zero distance, assuming that the slot
+    /// indices are packed as densely as possible. There are normally gaps
+    /// between instructions, so this assumption often doesn't hold. This
+    /// results in this function often returning a value greater than the actual
+    /// instruction distance.
+    int getApproxInstrDistance(SlotIndex other) const {
       return (other.listEntry()->getIndex() - listEntry()->getIndex())
         / Slot_Count;
     }
diff --git a/llvm/lib/CodeGen/RegAllocGreedy.cpp b/llvm/lib/CodeGen/RegAllocGreedy.cpp
--- a/llvm/lib/CodeGen/RegAllocGreedy.cpp
+++ b/llvm/lib/CodeGen/RegAllocGreedy.cpp
@@ -330,12 +330,12 @@
       // are singly defined, this produces optimal coloring in the absence of
       // global interference and other constraints.
       if (!ReverseLocalAssignment)
-        Prio = LI.beginIndex().getInstrDistance(Indexes->getLastIndex());
+        Prio = LI.beginIndex().getApproxInstrDistance(Indexes->getLastIndex());
       else {
         // Allocating bottom up may allow many short LRGs to be assigned first
         // to one of the cheap registers. This could be much faster for very
         // large blocks on targets with many physical registers.
-        Prio = Indexes->getZeroIndex().getInstrDistance(LI.endIndex());
+        Prio = Indexes->getZeroIndex().getApproxInstrDistance(LI.endIndex());
       }
     } else {
       // Allocate global and split ranges in long->short order. Long ranges that