Index: lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
===================================================================
--- lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -2418,6 +2418,28 @@
   return nullptr;
 }
 
+/// If we have a masked merge, in the form of:
+///   |        A  |  |B|
+///   ((x ^ y) & M) ^ y
+///    |  D  |
+/// If A has one use, and M is inverted, we can convert it into:
+///    |  D  |
+///   ((x ^ y) & ~M) ^ x
+static Instruction *invertMaskInMaskedMerge(BinaryOperator &I,
+                                            InstCombiner::BuilderTy &Builder) {
+  Value *B, *X, *D, *M;
+  if (!match(&I, m_c_Xor(m_Value(B),
+                         m_OneUse(m_c_And(
+                             m_CombineAnd(m_c_Xor(m_c_Specific(&B), m_Value(X)),
+                                          m_Value(D)),
+                             m_Not(m_Value(M)))))))
+    return nullptr;
+
+  // De-invert the mask and swap the value in B part.
+  Value *NewA = Builder.CreateAnd(D, M);
+  return BinaryOperator::CreateXor(NewA, X);
+}
+
 // FIXME: We use commutative matchers (m_c_*) for some, but not all, matches
 // here. We should standardize that construct where it is needed or choose some
 // other way to ensure that commutated variants of patterns are not missed.
@@ -2468,6 +2490,9 @@
     return BinaryOperator::CreateAnd(X, NotY);
   }
 
+  if (Instruction *Xor = invertMaskInMaskedMerge(I, Builder))
+    return Xor;
+
   // Is this a 'not' (~) fed by a binary operator?
   BinaryOperator *NotVal;
   if (match(&I, m_Not(m_BinOp(NotVal)))) {
Index: test/Transforms/InstCombine/invert-variable-mask-in-masked-merge-scalar.ll
===================================================================
--- test/Transforms/InstCombine/invert-variable-mask-in-masked-merge-scalar.ll
+++ test/Transforms/InstCombine/invert-variable-mask-in-masked-merge-scalar.ll
@@ -8,10 +8,9 @@
 
 define i4 @scalar (i4 %x, i4 %y, i4 %m) {
 ; CHECK-LABEL: @scalar(
-; CHECK-NEXT:    [[IM:%.*]] = xor i4 [[M:%.*]], -1
 ; CHECK-NEXT:    [[N0:%.*]] = xor i4 [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and i4 [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor i4 [[N1]], [[Y]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and i4 [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor i4 [[TMP1]], [[X]]
 ; CHECK-NEXT:    ret i4 [[R]]
 ;
   %im = xor i4 %m, -1
@@ -32,10 +31,9 @@
 
 define i4 @c_1_0_0 (i4 %x, i4 %y, i4 %m) {
 ; CHECK-LABEL: @c_1_0_0(
-; CHECK-NEXT:    [[IM:%.*]] = xor i4 [[M:%.*]], -1
 ; CHECK-NEXT:    [[N0:%.*]] = xor i4 [[Y:%.*]], [[X:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and i4 [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor i4 [[N1]], [[Y]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and i4 [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor i4 [[TMP1]], [[X]]
 ; CHECK-NEXT:    ret i4 [[R]]
 ;
   %im = xor i4 %m, -1
@@ -47,10 +45,9 @@
 
 define i4 @c_0_1_0 (i4 %x, i4 %y, i4 %m) {
 ; CHECK-LABEL: @c_0_1_0(
-; CHECK-NEXT:    [[IM:%.*]] = xor i4 [[M:%.*]], -1
 ; CHECK-NEXT:    [[N0:%.*]] = xor i4 [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and i4 [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor i4 [[N1]], [[X]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and i4 [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor i4 [[TMP1]], [[Y]]
 ; CHECK-NEXT:    ret i4 [[R]]
 ;
   %im = xor i4 %m, -1
@@ -62,12 +59,11 @@
 
 define i4 @c_0_0_1 (i4 %m) {
 ; CHECK-LABEL: @c_0_0_1(
-; CHECK-NEXT:    [[IM:%.*]] = xor i4 [[M:%.*]], -1
 ; CHECK-NEXT:    [[X:%.*]] = call i4 @gen4()
 ; CHECK-NEXT:    [[Y:%.*]] = call i4 @gen4()
 ; CHECK-NEXT:    [[N0:%.*]] = xor i4 [[X]], [[Y]]
-; CHECK-NEXT:    [[N1:%.*]] = and i4 [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor i4 [[Y]], [[N1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and i4 [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor i4 [[TMP1]], [[X]]
 ; CHECK-NEXT:    ret i4 [[R]]
 ;
   %im = xor i4 %m, -1
@@ -81,10 +77,9 @@
 
 define i4 @c_1_1_0 (i4 %x, i4 %y, i4 %m) {
 ; CHECK-LABEL: @c_1_1_0(
-; CHECK-NEXT:    [[IM:%.*]] = xor i4 [[M:%.*]], -1
 ; CHECK-NEXT:    [[N0:%.*]] = xor i4 [[Y:%.*]], [[X:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and i4 [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor i4 [[N1]], [[X]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and i4 [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor i4 [[TMP1]], [[Y]]
 ; CHECK-NEXT:    ret i4 [[R]]
 ;
   %im = xor i4 %m, -1
@@ -96,11 +91,10 @@
 
 define i4 @c_1_0_1 (i4 %x, i4 %m) {
 ; CHECK-LABEL: @c_1_0_1(
-; CHECK-NEXT:    [[IM:%.*]] = xor i4 [[M:%.*]], -1
 ; CHECK-NEXT:    [[Y:%.*]] = call i4 @gen4()
 ; CHECK-NEXT:    [[N0:%.*]] = xor i4 [[Y]], [[X:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and i4 [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor i4 [[Y]], [[N1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and i4 [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor i4 [[TMP1]], [[X]]
 ; CHECK-NEXT:    ret i4 [[R]]
 ;
   %im = xor i4 %m, -1
@@ -113,11 +107,10 @@
 
 define i4 @c_0_1_1 (i4 %y, i4 %m) {
 ; CHECK-LABEL: @c_0_1_1(
-; CHECK-NEXT:    [[IM:%.*]] = xor i4 [[M:%.*]], -1
 ; CHECK-NEXT:    [[X:%.*]] = call i4 @gen4()
 ; CHECK-NEXT:    [[N0:%.*]] = xor i4 [[X]], [[Y:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and i4 [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor i4 [[X]], [[N1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and i4 [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor i4 [[TMP1]], [[Y]]
 ; CHECK-NEXT:    ret i4 [[R]]
 ;
   %im = xor i4 %m, -1
@@ -130,12 +123,11 @@
 
 define i4 @c_1_1_1 (i4 %m) {
 ; CHECK-LABEL: @c_1_1_1(
-; CHECK-NEXT:    [[IM:%.*]] = xor i4 [[M:%.*]], -1
 ; CHECK-NEXT:    [[X:%.*]] = call i4 @gen4()
 ; CHECK-NEXT:    [[Y:%.*]] = call i4 @gen4()
 ; CHECK-NEXT:    [[N0:%.*]] = xor i4 [[Y]], [[X]]
-; CHECK-NEXT:    [[N1:%.*]] = and i4 [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor i4 [[X]], [[N1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and i4 [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor i4 [[TMP1]], [[Y]]
 ; CHECK-NEXT:    ret i4 [[R]]
 ;
   %im = xor i4 %m, -1
Index: test/Transforms/InstCombine/invert-variable-mask-in-masked-merge-vector.ll
===================================================================
--- test/Transforms/InstCombine/invert-variable-mask-in-masked-merge-vector.ll
+++ test/Transforms/InstCombine/invert-variable-mask-in-masked-merge-vector.ll
@@ -8,10 +8,9 @@
 
 define <2 x i4> @vector (<2 x i4> %x, <2 x i4> %y, <2 x i4> %m) {
 ; CHECK-LABEL: @vector(
-; CHECK-NEXT:    [[IM:%.*]] = xor <2 x i4> [[M:%.*]], <i4 -1, i4 -1>
 ; CHECK-NEXT:    [[N0:%.*]] = xor <2 x i4> [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and <2 x i4> [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[N1]], [[Y]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and <2 x i4> [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[TMP1]], [[X]]
 ; CHECK-NEXT:    ret <2 x i4> [[R]]
 ;
   %im = xor <2 x i4> %m, <i4 -1, i4 -1>
@@ -23,10 +22,9 @@
 
 define <3 x i4> @vector_undef (<3 x i4> %x, <3 x i4> %y, <3 x i4> %m) {
 ; CHECK-LABEL: @vector_undef(
-; CHECK-NEXT:    [[IM:%.*]] = xor <3 x i4> [[M:%.*]], <i4 -1, i4 undef, i4 -1>
 ; CHECK-NEXT:    [[N0:%.*]] = xor <3 x i4> [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and <3 x i4> [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor <3 x i4> [[N1]], [[Y]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and <3 x i4> [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor <3 x i4> [[TMP1]], [[X]]
 ; CHECK-NEXT:    ret <3 x i4> [[R]]
 ;
   %im = xor <3 x i4> %m, <i4 -1, i4 undef, i4 -1>
@@ -47,10 +45,9 @@
 
 define <2 x i4> @c_1_0_0 (<2 x i4> %x, <2 x i4> %y, <2 x i4> %m) {
 ; CHECK-LABEL: @c_1_0_0(
-; CHECK-NEXT:    [[IM:%.*]] = xor <2 x i4> [[M:%.*]], <i4 -1, i4 -1>
 ; CHECK-NEXT:    [[N0:%.*]] = xor <2 x i4> [[Y:%.*]], [[X:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and <2 x i4> [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[N1]], [[Y]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and <2 x i4> [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[TMP1]], [[X]]
 ; CHECK-NEXT:    ret <2 x i4> [[R]]
 ;
   %im = xor <2 x i4> %m, <i4 -1, i4 -1>
@@ -62,10 +59,9 @@
 
 define <2 x i4> @c_0_1_0 (<2 x i4> %x, <2 x i4> %y, <2 x i4> %m) {
 ; CHECK-LABEL: @c_0_1_0(
-; CHECK-NEXT:    [[IM:%.*]] = xor <2 x i4> [[M:%.*]], <i4 -1, i4 -1>
 ; CHECK-NEXT:    [[N0:%.*]] = xor <2 x i4> [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and <2 x i4> [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[N1]], [[X]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and <2 x i4> [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[TMP1]], [[Y]]
 ; CHECK-NEXT:    ret <2 x i4> [[R]]
 ;
   %im = xor <2 x i4> %m, <i4 -1, i4 -1>
@@ -77,12 +73,11 @@
 
 define <2 x i4> @c_0_0_1 (<2 x i4> %m) {
 ; CHECK-LABEL: @c_0_0_1(
-; CHECK-NEXT:    [[IM:%.*]] = xor <2 x i4> [[M:%.*]], <i4 -1, i4 -1>
 ; CHECK-NEXT:    [[X:%.*]] = call <2 x i4> @gen4()
 ; CHECK-NEXT:    [[Y:%.*]] = call <2 x i4> @gen4()
 ; CHECK-NEXT:    [[N0:%.*]] = xor <2 x i4> [[X]], [[Y]]
-; CHECK-NEXT:    [[N1:%.*]] = and <2 x i4> [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[Y]], [[N1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and <2 x i4> [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[TMP1]], [[X]]
 ; CHECK-NEXT:    ret <2 x i4> [[R]]
 ;
   %im = xor <2 x i4> %m, <i4 -1, i4 -1>
@@ -96,10 +91,9 @@
 
 define <2 x i4> @c_1_1_0 (<2 x i4> %x, <2 x i4> %y, <2 x i4> %m) {
 ; CHECK-LABEL: @c_1_1_0(
-; CHECK-NEXT:    [[IM:%.*]] = xor <2 x i4> [[M:%.*]], <i4 -1, i4 -1>
 ; CHECK-NEXT:    [[N0:%.*]] = xor <2 x i4> [[Y:%.*]], [[X:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and <2 x i4> [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[N1]], [[X]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and <2 x i4> [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[TMP1]], [[Y]]
 ; CHECK-NEXT:    ret <2 x i4> [[R]]
 ;
   %im = xor <2 x i4> %m, <i4 -1, i4 -1>
@@ -111,11 +105,10 @@
 
 define <2 x i4> @c_1_0_1 (<2 x i4> %x, <2 x i4> %m) {
 ; CHECK-LABEL: @c_1_0_1(
-; CHECK-NEXT:    [[IM:%.*]] = xor <2 x i4> [[M:%.*]], <i4 -1, i4 -1>
 ; CHECK-NEXT:    [[Y:%.*]] = call <2 x i4> @gen4()
 ; CHECK-NEXT:    [[N0:%.*]] = xor <2 x i4> [[Y]], [[X:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and <2 x i4> [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[Y]], [[N1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and <2 x i4> [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[TMP1]], [[X]]
 ; CHECK-NEXT:    ret <2 x i4> [[R]]
 ;
   %im = xor <2 x i4> %m, <i4 -1, i4 -1>
@@ -128,11 +121,10 @@
 
 define <2 x i4> @c_0_1_1 (<2 x i4> %y, <2 x i4> %m) {
 ; CHECK-LABEL: @c_0_1_1(
-; CHECK-NEXT:    [[IM:%.*]] = xor <2 x i4> [[M:%.*]], <i4 -1, i4 -1>
 ; CHECK-NEXT:    [[X:%.*]] = call <2 x i4> @gen4()
 ; CHECK-NEXT:    [[N0:%.*]] = xor <2 x i4> [[X]], [[Y:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and <2 x i4> [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[X]], [[N1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and <2 x i4> [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[TMP1]], [[Y]]
 ; CHECK-NEXT:    ret <2 x i4> [[R]]
 ;
   %im = xor <2 x i4> %m, <i4 -1, i4 -1>
@@ -145,12 +137,11 @@
 
 define <2 x i4> @c_1_1_1 (<2 x i4> %m) {
 ; CHECK-LABEL: @c_1_1_1(
-; CHECK-NEXT:    [[IM:%.*]] = xor <2 x i4> [[M:%.*]], <i4 -1, i4 -1>
 ; CHECK-NEXT:    [[X:%.*]] = call <2 x i4> @gen4()
 ; CHECK-NEXT:    [[Y:%.*]] = call <2 x i4> @gen4()
 ; CHECK-NEXT:    [[N0:%.*]] = xor <2 x i4> [[Y]], [[X]]
-; CHECK-NEXT:    [[N1:%.*]] = and <2 x i4> [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[X]], [[N1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and <2 x i4> [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor <2 x i4> [[TMP1]], [[Y]]
 ; CHECK-NEXT:    ret <2 x i4> [[R]]
 ;
   %im = xor <2 x i4> %m, <i4 -1, i4 -1>
Index: test/Transforms/InstCombine/masked-merge-with-xors.ll
===================================================================
--- test/Transforms/InstCombine/masked-merge-with-xors.ll
+++ test/Transforms/InstCombine/masked-merge-with-xors.ll
@@ -5,10 +5,9 @@
 
 define i32 @f_0(i32 %x, i32 %y, i32 %m) {
 ; CHECK-LABEL: @f_0(
-; CHECK-NEXT:    [[IM:%.*]] = xor i32 [[M:%.*]], -1
 ; CHECK-NEXT:    [[N0:%.*]] = xor i32 [[Y:%.*]], [[X:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and i32 [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor i32 [[N1]], [[Y]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor i32 [[TMP1]], [[X]]
 ; CHECK-NEXT:    ret i32 [[R]]
 ;
   %im = xor i32 %m, -1
@@ -21,10 +20,9 @@
 define i32 @f_1(i32 %x, i32 %y, i32 %m) {
 ; CHECK-LABEL: @f_1(
 ; CHECK-NEXT:    [[IX:%.*]] = xor i32 [[X:%.*]], -1
-; CHECK-NEXT:    [[IM:%.*]] = xor i32 [[M:%.*]], -1
 ; CHECK-NEXT:    [[N0:%.*]] = xor i32 [[IX]], [[Y:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and i32 [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor i32 [[N1]], [[Y]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor i32 [[TMP1]], [[IX]]
 ; CHECK-NEXT:    ret i32 [[R]]
 ;
   %ix = xor i32 %x, -1
@@ -43,10 +41,9 @@
 define i32 @f_2(i32 %x, i32 %y, i32 %m) {
 ; CHECK-LABEL: @f_2(
 ; CHECK-NEXT:    [[IY:%.*]] = xor i32 [[Y:%.*]], -1
-; CHECK-NEXT:    [[IM:%.*]] = xor i32 [[M:%.*]], -1
 ; CHECK-NEXT:    [[N0:%.*]] = xor i32 [[IY]], [[X:%.*]]
-; CHECK-NEXT:    [[N1:%.*]] = and i32 [[N0]], [[IM]]
-; CHECK-NEXT:    [[R:%.*]] = xor i32 [[N1]], [[IY]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[N0]], [[M:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = xor i32 [[TMP1]], [[X]]
 ; CHECK-NEXT:    ret i32 [[R]]
 ;
   %iy = xor i32 %y, -1