Index: lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- lib/Target/X86/X86ISelLowering.cpp
+++ lib/Target/X86/X86ISelLowering.cpp
@@ -26791,9 +26791,49 @@
   return Cond;
 }
 
+/// Turn vector tests of the signbit in the form of:
+///   xor (sra X, elt_size(X)-1), -1
+/// into:
+///   pcmpgt X, -1
+static SDValue foldVectorXorShiftIntoCmp(SDNode *N, SelectionDAG &DAG,
+                                         const X86Subtarget &Subtarget) {
+  EVT VT = N->getValueType(0);
+  // TODO: AVX2 can handle 256-bit integer vectors.
+  if (!VT.is128BitVector() || !Subtarget.hasSSE2())
+    return SDValue();
+
+  if (VT == MVT::v2i64 && !Subtarget.hasSSE42())
+    return SDValue();
+
+  // There must be a shift right algebraic before the xor, and the xor must be a
+  // 'not' operation.
+  SDValue Shift = N->getOperand(0);
+  SDValue Ones = N->getOperand(1);
+  if (Shift.getOpcode() != ISD::SRA || !Shift.hasOneUse() ||
+      !ISD::isBuildVectorAllOnes(Ones.getNode()))
+    return SDValue();
+
+  // The shift should be smearing the sign bit across each vector element.
+  auto *ShiftBV = dyn_cast<BuildVectorSDNode>(Shift.getOperand(1));
+  if (!ShiftBV)
+    return SDValue();
+
+  EVT ShiftEltTy = Shift.getValueType().getVectorElementType();
+  auto *ShiftAmt = ShiftBV->getConstantSplatNode();
+  if (!ShiftAmt || ShiftAmt->getZExtValue() != ShiftEltTy.getSizeInBits() - 1)
+    return SDValue();
+
+  // Create a greater-than comparison against -1. We don't use the more obvious
+  // greater-than-or-equal-to-zero because SSE/AVX don't have that instruction.
+  return DAG.getNode(X86ISD::PCMPGT, SDLoc(N), VT, Shift.getOperand(0), Ones);
+}
+
 static SDValue PerformXorCombine(SDNode *N, SelectionDAG &DAG,
                                  TargetLowering::DAGCombinerInfo &DCI,
                                  const X86Subtarget &Subtarget) {
+  if (SDValue Cmp = foldVectorXorShiftIntoCmp(N, DAG, Subtarget))
+    return Cmp;
+
   if (DCI.isBeforeLegalizeOps())
     return SDValue();
 
Index: test/CodeGen/X86/vector-pcmp.ll
===================================================================
--- test/CodeGen/X86/vector-pcmp.ll
+++ test/CodeGen/X86/vector-pcmp.ll
@@ -9,18 +9,14 @@
 define <16 x i8> @pcmpgtb(<16 x i8> %x) {
 ; SSE-LABEL: pcmpgtb:
 ; SSE:       # BB#0:
-; SSE-NEXT:    pxor %xmm1, %xmm1
-; SSE-NEXT:    pcmpgtb %xmm0, %xmm1
-; SSE-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE-NEXT:    pcmpgtb %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: pcmpgtb:
 ; AVX:       # BB#0:
-; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    retq
   %sign = ashr <16 x i8> %x, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
   %not = xor <16 x i8> %sign, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
@@ -30,16 +26,14 @@
 define <8 x i16> @pcmpgtw(<8 x i16> %x) {
 ; SSE-LABEL: pcmpgtw:
 ; SSE:       # BB#0:
-; SSE-NEXT:    psraw $15, %xmm0
 ; SSE-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    pcmpgtw %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: pcmpgtw:
 ; AVX:       # BB#0:
-; AVX-NEXT:    vpsraw $15, %xmm0, %xmm0
 ; AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    retq
   %sign = ashr <8 x i16> %x, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
   %not = xor <8 x i16> %sign, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
@@ -49,16 +43,14 @@
 define <4 x i32> @pcmpgtd(<4 x i32> %x) {
 ; SSE-LABEL: pcmpgtd:
 ; SSE:       # BB#0:
-; SSE-NEXT:    psrad $31, %xmm0
 ; SSE-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    pcmpgtd %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: pcmpgtd:
 ; AVX:       # BB#0:
-; AVX-NEXT:    vpsrad $31, %xmm0, %xmm0
 ; AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    retq
   %sign = ashr <4 x i32> %x, <i32 31, i32 31, i32 31, i32 31>
   %not = xor <4 x i32> %sign, <i32 -1, i32 -1, i32 -1, i32 -1>
@@ -66,20 +58,24 @@
 }
 
 define <2 x i64> @pcmpgtq(<2 x i64> %x) {
-; SSE-LABEL: pcmpgtq:
-; SSE:       # BB#0:
-; SSE-NEXT:    psrad $31, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; SSE-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE-NEXT:    pxor %xmm1, %xmm0
-; SSE-NEXT:    retq
+; SSE2-LABEL: pcmpgtq:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    psrad $31, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
+; SSE2-NEXT:    pxor %xmm1, %xmm0
+; SSE2-NEXT:    retq
+;
+; SSE42-LABEL: pcmpgtq:
+; SSE42:       # BB#0:
+; SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE42-NEXT:    pcmpgtq %xmm1, %xmm0
+; SSE42-NEXT:    retq
 ;
 ; AVX-LABEL: pcmpgtq:
 ; AVX:       # BB#0:
-; AVX-NEXT:    vpsrad $31, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
 ; AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    retq
   %sign = ashr <2 x i64> %x, <i64 63, i64 63>
   %not = xor <2 x i64> %sign, <i64 -1, i64 -1>