diff --git a/lib/CodeGen/MachineCSE.cpp b/lib/CodeGen/MachineCSE.cpp
index 0143d0fbea2..fc3f5943d71 100644
--- a/lib/CodeGen/MachineCSE.cpp
+++ b/lib/CodeGen/MachineCSE.cpp
@@ -24,7 +24,6 @@
 #include "llvm/ADT/ScopedHashTable.h"
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/ADT/Statistic.h"
-#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 
 using namespace llvm;
@@ -33,6 +32,7 @@ STATISTIC(NumCoalesces, "Number of copies coalesced");
 STATISTIC(NumCSEs,      "Number of common subexpression eliminated");
 STATISTIC(NumPhysCSEs,
           "Number of physreg referencing common subexpr eliminated");
+STATISTIC(NumCommutes,  "Number of copies coalesced after commuting");
 
 namespace {
   class MachineCSE : public MachineFunctionPass {
@@ -368,7 +368,22 @@ bool MachineCSE::ProcessBlock(MachineBasicBlock *MBB) {
         FoundCSE = VNT.count(MI);
       }
     }
-    // FIXME: commute commutable instructions?
+
+    // Commute commutable instructions.
+    bool Commuted = false;
+    if (!FoundCSE && MI->getDesc().isCommutable()) {
+      MachineInstr *NewMI = TII->commuteInstruction(MI);
+      if (NewMI) {
+        Commuted = true;
+        FoundCSE = VNT.count(NewMI);
+        if (NewMI != MI)
+          // New instruction. It doesn't need to be kept.
+          NewMI->eraseFromParent();
+        else if (!FoundCSE)
+          // MI was changed but it didn't help, commute it back!
+          (void)TII->commuteInstruction(MI);
+      }
+    }
 
     // If the instruction defines physical registers and the values *may* be
     // used, then it's not safe to replace it with a common subexpression.
@@ -430,6 +445,8 @@ bool MachineCSE::ProcessBlock(MachineBasicBlock *MBB) {
       ++NumCSEs;
       if (!PhysRefs.empty())
         ++NumPhysCSEs;
+      if (Commuted)
+        ++NumCommutes;
     } else {
       DEBUG(dbgs() << "*** Not profitable, avoid CSE!\n");
       VNT.insert(MI, CurrVN++);
diff --git a/test/CodeGen/X86/machine-cse.ll b/test/CodeGen/X86/machine-cse.ll
index a8afdc84c51..e284776ed02 100644
--- a/test/CodeGen/X86/machine-cse.ll
+++ b/test/CodeGen/X86/machine-cse.ll
@@ -37,3 +37,43 @@ bb3:
 declare void @bar(i32*)
 
 declare fastcc i8* @foo(%struct.s2*) nounwind
+
+; rdar://8773371
+
+declare void @printf(...) nounwind
+
+define void @commute(i32 %test_case, i32 %scale) nounwind ssp {
+; CHECK: commute:
+entry:
+  switch i32 %test_case, label %sw.bb307 [
+    i32 1, label %sw.bb
+    i32 2, label %sw.bb
+    i32 3, label %sw.bb
+  ]
+
+sw.bb:                                            ; preds = %entry, %entry, %entry
+  %mul = mul nsw i32 %test_case, 3
+  %mul20 = mul nsw i32 %mul, %scale
+  br i1 undef, label %if.end34, label %sw.bb307
+
+if.end34:                                         ; preds = %sw.bb
+; CHECK: %if.end34
+; CHECK: imull
+; CHECK: leal
+; CHECK-NOT: imull
+  tail call void (...)* @printf(i32 %test_case, i32 %mul20) nounwind
+  %tmp = mul i32 %scale, %test_case
+  %tmp752 = mul i32 %tmp, 3
+  %tmp753 = zext i32 %tmp752 to i64
+  br label %bb.nph743.us
+
+for.body53.us:                                    ; preds = %bb.nph743.us, %for.body53.us
+  %exitcond = icmp eq i64 undef, %tmp753
+  br i1 %exitcond, label %bb.nph743.us, label %for.body53.us
+
+bb.nph743.us:                                     ; preds = %for.body53.us, %if.end34
+  br label %for.body53.us
+
+sw.bb307:                                         ; preds = %sw.bb, %entry
+  ret void
+}