diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -5522,6 +5522,23 @@
   return true;
 }
 
+// Match 2 elements of a packed halfword bswap.
+static bool isBSwapHWordPair(SDValue N, MutableArrayRef<SDNode *> Parts) {
+  if (N.getOpcode() == ISD::OR)
+    return isBSwapHWordElement(N.getOperand(0), Parts) &&
+           isBSwapHWordElement(N.getOperand(1), Parts);
+
+  if (N.getOpcode() == ISD::SRL && N.getOperand(0).getOpcode() == ISD::BSWAP) {
+    ConstantSDNode *C = isConstOrConstSplat(N.getOperand(1));
+    if (!C || C->getAPIntValue() != 16)
+      return false;
+    Parts[0] = Parts[1] = N.getOperand(0).getOperand(0).getNode();
+    return true;
+  }
+
+  return false;
+}
+
 /// Match a 32-bit packed halfword bswap. That is
 /// ((x & 0x000000ff) << 8) |
 /// ((x & 0x0000ff00) >> 8) |
@@ -5539,43 +5556,26 @@
     return SDValue();
 
   // Look for either
-  // (or (or (and), (and)), (or (and), (and)))
-  // (or (or (or (and), (and)), (and)), (and))
-  if (N0.getOpcode() != ISD::OR)
-    return SDValue();
-  SDValue N00 = N0.getOperand(0);
-  SDValue N01 = N0.getOperand(1);
+  // (or (bswaphpair), (bswaphpair))
+  // (or (or (bswaphpair), (and)), (and))
+  // (or (or (and), (bswaphpair)), (and))
   SDNode *Parts[4] = {};
 
-  if (N1.getOpcode() == ISD::OR &&
-      N00.getNumOperands() == 2 && N01.getNumOperands() == 2) {
+  if (isBSwapHWordPair(N0, Parts)) {
     // (or (or (and), (and)), (or (and), (and)))
-    if (!isBSwapHWordElement(N00, Parts))
-      return SDValue();
-
-    if (!isBSwapHWordElement(N01, Parts))
-      return SDValue();
-    SDValue N10 = N1.getOperand(0);
-    if (!isBSwapHWordElement(N10, Parts))
-      return SDValue();
-    SDValue N11 = N1.getOperand(1);
-    if (!isBSwapHWordElement(N11, Parts))
+    if (!isBSwapHWordPair(N1, Parts))
       return SDValue();
-  } else {
+  } else if (N0.getOpcode() == ISD::OR) {
     // (or (or (or (and), (and)), (and)), (and))
     if (!isBSwapHWordElement(N1, Parts))
       return SDValue();
-    if (!isBSwapHWordElement(N01, Parts))
-      return SDValue();
-    if (N00.getOpcode() != ISD::OR)
-      return SDValue();
-    SDValue N000 = N00.getOperand(0);
-    if (!isBSwapHWordElement(N000, Parts))
-      return SDValue();
-    SDValue N001 = N00.getOperand(1);
-    if (!isBSwapHWordElement(N001, Parts))
+    SDValue N00 = N0.getOperand(0);
+    SDValue N01 = N0.getOperand(1);
+    if (!(isBSwapHWordElement(N01, Parts) && isBSwapHWordPair(N00, Parts)) &&
+        !(isBSwapHWordElement(N00, Parts) && isBSwapHWordPair(N01, Parts)))
       return SDValue();
-  }
+  } else
+    return SDValue();
 
   // Make sure the parts are all coming from the same node.
   if (Parts[0] != Parts[1] || Parts[0] != Parts[2] || Parts[0] != Parts[3])
diff --git a/llvm/test/CodeGen/X86/bswap_tree.ll b/llvm/test/CodeGen/X86/bswap_tree.ll
--- a/llvm/test/CodeGen/X86/bswap_tree.ll
+++ b/llvm/test/CodeGen/X86/bswap_tree.ll
@@ -79,30 +79,15 @@
 ; CHECK-LABEL: test3:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; CHECK-NEXT:    movl %eax, %ecx
-; CHECK-NEXT:    andl $16711680, %ecx # imm = 0xFF0000
-; CHECK-NEXT:    movl %eax, %edx
-; CHECK-NEXT:    andl $-16777216, %edx # imm = 0xFF000000
-; CHECK-NEXT:    shll $8, %ecx
-; CHECK-NEXT:    shrl $8, %edx
-; CHECK-NEXT:    orl %ecx, %edx
 ; CHECK-NEXT:    bswapl %eax
-; CHECK-NEXT:    shrl $16, %eax
-; CHECK-NEXT:    orl %edx, %eax
+; CHECK-NEXT:    roll $16, %eax
 ; CHECK-NEXT:    retl
 ;
 ; CHECK64-LABEL: test3:
 ; CHECK64:       # %bb.0:
 ; CHECK64-NEXT:    movl %edi, %eax
-; CHECK64-NEXT:    andl $16711680, %eax # imm = 0xFF0000
-; CHECK64-NEXT:    movl %edi, %ecx
-; CHECK64-NEXT:    andl $-16777216, %ecx # imm = 0xFF000000
-; CHECK64-NEXT:    shll $8, %eax
-; CHECK64-NEXT:    shrl $8, %ecx
-; CHECK64-NEXT:    addl %ecx, %eax
-; CHECK64-NEXT:    bswapl %edi
-; CHECK64-NEXT:    shrl $16, %edi
-; CHECK64-NEXT:    orl %edi, %eax
+; CHECK64-NEXT:    bswapl %eax
+; CHECK64-NEXT:    roll $16, %eax
 ; CHECK64-NEXT:    retq
   %byte2 = and i32 %x, 16711680   ; 0x00ff0000
   %byte3 = and i32 %x, 4278190080 ; 0xff000000