Index: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
===================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -4823,7 +4823,8 @@
 // reduces to a rotate in direction shift2 by Pos or (equivalently) a rotate
 // in direction shift1 by Neg.  The range [0, EltSize) means that we only need
 // to consider shift amounts with defined behavior.
-static bool matchRotateSub(SDValue Pos, SDValue Neg, unsigned EltSize) {
+static bool matchRotateSub(SDValue Pos, SDValue Neg, unsigned EltSize,
+                           SelectionDAG &DAG) {
   // If EltSize is a power of 2 then:
   //
   //  (a) (Pos == 0 ? 0 : EltSize - Pos) == (EltSize - Pos) & (EltSize - 1)
@@ -4858,9 +4859,13 @@
   unsigned MaskLoBits = 0;
   if (Neg.getOpcode() == ISD::AND && isPowerOf2_64(EltSize)) {
     if (ConstantSDNode *NegC = isConstOrConstSplat(Neg.getOperand(1))) {
-      if (NegC->getAPIntValue() == EltSize - 1) {
+      KnownBits Known;
+      DAG.computeKnownBits(Neg.getOperand(0), Known);
+      unsigned Bits = Log2_64(EltSize);
+      if (NegC->getAPIntValue().getActiveBits() <= Bits &&
+          ((NegC->getAPIntValue() | Known.Zero).countTrailingOnes() >= Bits)) {
         Neg = Neg.getOperand(0);
-        MaskLoBits = Log2_64(EltSize);
+        MaskLoBits = Bits;
       }
     }
   }
@@ -4875,10 +4880,16 @@
 
   // On the RHS of [A], if Pos is Pos' & (EltSize - 1), just replace Pos with
   // Pos'.  The truncation is redundant for the purpose of the equality.
-  if (MaskLoBits && Pos.getOpcode() == ISD::AND)
-    if (ConstantSDNode *PosC = isConstOrConstSplat(Pos.getOperand(1)))
-      if (PosC->getAPIntValue() == EltSize - 1)
+  if (MaskLoBits && Pos.getOpcode() == ISD::AND) {
+    if (ConstantSDNode *PosC = isConstOrConstSplat(Pos.getOperand(1))) {
+      KnownBits Known;
+      DAG.computeKnownBits(Pos.getOperand(0), Known);
+      if (PosC->getAPIntValue().getActiveBits() <= MaskLoBits &&
+          ((PosC->getAPIntValue() | Known.Zero).countTrailingOnes() >=
+           MaskLoBits))
         Pos = Pos.getOperand(0);
+    }
+  }
 
   // The condition we need is now:
   //
@@ -4934,7 +4945,7 @@
   //          (srl x, (*ext y))) ->
   //   (rotr x, y) or (rotl x, (sub 32, y))
   EVT VT = Shifted.getValueType();
-  if (matchRotateSub(InnerPos, InnerNeg, VT.getScalarSizeInBits())) {
+  if (matchRotateSub(InnerPos, InnerNeg, VT.getScalarSizeInBits(), DAG)) {
     bool HasPos = TLI.isOperationLegalOrCustom(PosOpcode, VT);
     return DAG.getNode(HasPos ? PosOpcode : NegOpcode, DL, VT, Shifted,
                        HasPos ? Pos : Neg).getNode();
Index: llvm/trunk/test/CodeGen/X86/combine-rotates.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/combine-rotates.ll
+++ llvm/trunk/test/CodeGen/X86/combine-rotates.ll
@@ -61,27 +61,14 @@
 define <4 x i32> @rotate_demanded_bits(<4 x i32>, <4 x i32>) {
 ; XOP-LABEL: rotate_demanded_bits:
 ; XOP:       # %bb.0:
-; XOP-NEXT:    vmovdqa {{.*#+}} xmm2 = [30,30,30,30]
-; XOP-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; XOP-NEXT:    vpshld %xmm1, %xmm0, %xmm3
-; XOP-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; XOP-NEXT:    vpsubd %xmm1, %xmm4, %xmm1
-; XOP-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; XOP-NEXT:    vpsubd %xmm1, %xmm4, %xmm1
-; XOP-NEXT:    vpshld %xmm1, %xmm0, %xmm0
-; XOP-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; XOP-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; XOP-NEXT:    vprotd %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; AVX512-LABEL: rotate_demanded_bits:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [30,30,30,30]
-; AVX512-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; AVX512-NEXT:    vpsllvd %xmm1, %xmm0, %xmm3
-; AVX512-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512-NEXT:    vpsubd %xmm1, %xmm4, %xmm1
-; AVX512-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; AVX512-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; AVX512-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm1, %xmm1
+; AVX512-NEXT:    vprolvd %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    retq
   %3 = and <4 x i32> %1, <i32 30, i32 30, i32 30, i32 30>
   %4 = shl <4 x i32> %0, %3
@@ -117,28 +104,15 @@
 ; XOP-LABEL: rotate_demanded_bits_3:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vpaddd %xmm1, %xmm1, %xmm1
-; XOP-NEXT:    vmovdqa {{.*#+}} xmm2 = [30,30,30,30]
-; XOP-NEXT:    vpand %xmm2, %xmm1, %xmm3
-; XOP-NEXT:    vpshld %xmm3, %xmm0, %xmm3
-; XOP-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; XOP-NEXT:    vpsubd %xmm1, %xmm4, %xmm1
-; XOP-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; XOP-NEXT:    vpsubd %xmm1, %xmm4, %xmm1
-; XOP-NEXT:    vpshld %xmm1, %xmm0, %xmm0
-; XOP-NEXT:    vpor %xmm0, %xmm3, %xmm0
+; XOP-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; XOP-NEXT:    vprotd %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; AVX512-LABEL: rotate_demanded_bits_3:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpaddd %xmm1, %xmm1, %xmm1
-; AVX512-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [30,30,30,30]
-; AVX512-NEXT:    vpand %xmm2, %xmm1, %xmm3
-; AVX512-NEXT:    vpsllvd %xmm3, %xmm0, %xmm3
-; AVX512-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512-NEXT:    vpsubd %xmm1, %xmm4, %xmm1
-; AVX512-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; AVX512-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpor %xmm0, %xmm3, %xmm0
+; AVX512-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm1, %xmm1
+; AVX512-NEXT:    vprolvd %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    retq
   %3 = shl <4 x i32> %1, <i32 1, i32 1, i32 1, i32 1>
   %4 = and <4 x i32> %3, <i32 30, i32 30, i32 30, i32 30>
Index: llvm/trunk/test/CodeGen/X86/rotate4.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/rotate4.ll
+++ llvm/trunk/test/CodeGen/X86/rotate4.ll
@@ -284,15 +284,9 @@
 define i32 @rotate_demanded_bits(i32, i32) {
 ; CHECK-LABEL: rotate_demanded_bits:
 ; CHECK:       # %bb.0:
+; CHECK-NEXT:    andb $30, %sil
 ; CHECK-NEXT:    movl %esi, %ecx
-; CHECK-NEXT:    andl $30, %ecx
-; CHECK-NEXT:    movl %edi, %eax
-; CHECK-NEXT:    shll %cl, %eax
-; CHECK-NEXT:    negl %ecx
-; CHECK-NEXT:    andb $30, %cl
-; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
-; CHECK-NEXT:    shrl %cl, %edi
-; CHECK-NEXT:    orl %eax, %edi
+; CHECK-NEXT:    roll %cl, %edi
 ; CHECK-NEXT:    movl %edi, %eax
 ; CHECK-NEXT:    retq
   %3 = and i32 %1, 30
@@ -324,16 +318,10 @@
 define i32 @rotate_demanded_bits_3(i32, i32) {
 ; CHECK-LABEL: rotate_demanded_bits_3:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    addl %esi, %esi
-; CHECK-NEXT:    movl %esi, %ecx
-; CHECK-NEXT:    andb $30, %cl
-; CHECK-NEXT:    movl %edi, %eax
-; CHECK-NEXT:    shll %cl, %eax
-; CHECK-NEXT:    negl %esi
+; CHECK-NEXT:    addb %sil, %sil
 ; CHECK-NEXT:    andb $30, %sil
 ; CHECK-NEXT:    movl %esi, %ecx
-; CHECK-NEXT:    shrl %cl, %edi
-; CHECK-NEXT:    orl %eax, %edi
+; CHECK-NEXT:    roll %cl, %edi
 ; CHECK-NEXT:    movl %edi, %eax
 ; CHECK-NEXT:    retq
   %3 = shl i32 %1, 1