Index: llvm/lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- llvm/lib/Target/X86/X86ISelLowering.cpp
+++ llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -30266,10 +30266,12 @@
   SDLoc DL(Op);
   uint64_t SizeInBits = VT.getScalarSizeInBits();
   APInt PreferredZero = APInt::getZero(SizeInBits);
+  APInt OppositeZero = PreferredZero;
   EVT IVT = VT.changeTypeToInteger();
   X86ISD::NodeType MinMaxOp;
   if (Op.getOpcode() == ISD::FMAXIMUM) {
     MinMaxOp = X86ISD::FMAX;
+    OppositeZero.setSignBit();
   } else {
     PreferredZero.setSignBit();
     MinMaxOp = X86ISD::FMIN;
@@ -30294,7 +30296,7 @@
   // We check if any of operands is NaN and return NaN. Then we check if any of
   // operands is zero or negative zero (for fmaximum and fminimum respectively)
   // to ensure the correct zero is returned.
-  auto IsPreferredZero = [PreferredZero](SDValue Op) {
+  auto MatchesZero = [](SDValue Op, APInt PreferredZero) {
     Op = peekThroughBitcasts(Op);
     if (auto *CstOp = dyn_cast<ConstantFPSDNode>(Op))
       return CstOp->getValueAPF().bitcastToAPInt() == PreferredZero;
@@ -30323,11 +30325,12 @@
                           DAG.isKnownNeverZeroFloat(X) ||
                           DAG.isKnownNeverZeroFloat(Y);
   SDValue NewX, NewY;
-  if (IgnoreSignedZero || IsPreferredZero(Y)) {
+  if (IgnoreSignedZero || MatchesZero(Y, PreferredZero) ||
+      MatchesZero(X, OppositeZero)) {
     // Operands are already in right order or order does not matter.
     NewX = X;
     NewY = Y;
-  } else if (IsPreferredZero(X)) {
+  } else if (MatchesZero(X, PreferredZero) || MatchesZero(Y, OppositeZero)) {
     NewX = Y;
     NewY = X;
   } else if (!VT.isVector() && (VT == MVT::f16 || Subtarget.hasDQI()) &&
Index: llvm/test/CodeGen/X86/fminimum-fmaximum.ll
===================================================================
--- llvm/test/CodeGen/X86/fminimum-fmaximum.ll
+++ llvm/test/CodeGen/X86/fminimum-fmaximum.ll
@@ -1042,42 +1042,21 @@
 define <2 x double> @test_fminimum_vector_zero(<2 x double> %x) {
 ; SSE2-LABEL: test_fminimum_vector_zero:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movaps %xmm0, %xmm1
-; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[3,3]
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    pcmpgtd %xmm1, %xmm2
-; SSE2-NEXT:    movaps %xmm0, %xmm1
-; SSE2-NEXT:    andps %xmm2, %xmm1
-; SSE2-NEXT:    andnps %xmm0, %xmm2
-; SSE2-NEXT:    movaps %xmm2, %xmm3
-; SSE2-NEXT:    minpd %xmm1, %xmm3
-; SSE2-NEXT:    movaps %xmm2, %xmm0
-; SSE2-NEXT:    cmpunordpd %xmm2, %xmm0
-; SSE2-NEXT:    andpd %xmm0, %xmm2
-; SSE2-NEXT:    andnpd %xmm3, %xmm0
-; SSE2-NEXT:    orpd %xmm2, %xmm0
+; SSE2-NEXT:    xorpd %xmm1, %xmm1
+; SSE2-NEXT:    minpd %xmm0, %xmm1
+; SSE2-NEXT:    movapd %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; AVX-LABEL: test_fminimum_vector_zero:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm1
-; AVX-NEXT:    vpand %xmm0, %xmm1, %xmm2
-; AVX-NEXT:    vpandn %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vminpd %xmm2, %xmm0, %xmm1
-; AVX-NEXT:    vcmpunordpd %xmm0, %xmm0, %xmm2
-; AVX-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
+; AVX-NEXT:    vminpd %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; X86-LABEL: test_fminimum_vector_zero:
 ; X86:       # %bb.0:
-; X86-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; X86-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm1
-; X86-NEXT:    vpand %xmm0, %xmm1, %xmm2
-; X86-NEXT:    vpandn %xmm0, %xmm1, %xmm0
-; X86-NEXT:    vminpd %xmm2, %xmm0, %xmm1
-; X86-NEXT:    vcmpunordpd %xmm0, %xmm0, %xmm2
-; X86-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
+; X86-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
+; X86-NEXT:    vminpd %xmm0, %xmm1, %xmm0
 ; X86-NEXT:    retl
   %r = call <2 x double> @llvm.minimum.v2f64(<2 x double> %x, <2 x double> <double 0., double 0.>)
   ret <2 x double> %r
@@ -1086,54 +1065,27 @@
 define <4 x float> @test_fmaximum_vector_signed_zero(<4 x float> %x) {
 ; SSE2-LABEL: test_fmaximum_vector_signed_zero:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]
-; SSE2-NEXT:    movdqa %xmm1, %xmm2
-; SSE2-NEXT:    pand %xmm0, %xmm2
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    pcmpgtd %xmm0, %xmm3
-; SSE2-NEXT:    movdqa %xmm3, %xmm4
-; SSE2-NEXT:    pandn %xmm0, %xmm4
-; SSE2-NEXT:    por %xmm2, %xmm4
-; SSE2-NEXT:    pand %xmm3, %xmm0
-; SSE2-NEXT:    pandn %xmm1, %xmm3
-; SSE2-NEXT:    por %xmm3, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    maxps %xmm4, %xmm1
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    cmpunordps %xmm0, %xmm2
-; SSE2-NEXT:    andps %xmm2, %xmm0
-; SSE2-NEXT:    andnps %xmm1, %xmm2
-; SSE2-NEXT:    orps %xmm2, %xmm0
+; SSE2-NEXT:    movaps {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]
+; SSE2-NEXT:    maxps %xmm0, %xmm1
+; SSE2-NEXT:    movaps %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; AVX1-LABEL: test_fmaximum_vector_signed_zero:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vmovaps {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]
-; AVX1-NEXT:    vblendvps %xmm0, %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vblendvps %xmm0, %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vmaxps %xmm2, %xmm0, %xmm1
-; AVX1-NEXT:    vcmpunordps %xmm0, %xmm0, %xmm2
-; AVX1-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vmaxps %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX512-LABEL: test_fmaximum_vector_signed_zero:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vbroadcastss {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]
-; AVX512-NEXT:    vblendvps %xmm0, %xmm1, %xmm0, %xmm2
-; AVX512-NEXT:    vblendvps %xmm0, %xmm0, %xmm1, %xmm0
-; AVX512-NEXT:    vmaxps %xmm2, %xmm0, %xmm1
-; AVX512-NEXT:    vcmpunordps %xmm0, %xmm0, %xmm2
-; AVX512-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; AVX512-NEXT:    vmaxps %xmm0, %xmm1, %xmm0
 ; AVX512-NEXT:    retq
 ;
 ; X86-LABEL: test_fmaximum_vector_signed_zero:
 ; X86:       # %bb.0:
 ; X86-NEXT:    vmovaps {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]
-; X86-NEXT:    vblendvps %xmm0, %xmm1, %xmm0, %xmm2
-; X86-NEXT:    vblendvps %xmm0, %xmm0, %xmm1, %xmm0
-; X86-NEXT:    vmaxps %xmm2, %xmm0, %xmm1
-; X86-NEXT:    vcmpunordps %xmm0, %xmm0, %xmm2
-; X86-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; X86-NEXT:    vmaxps %xmm0, %xmm1, %xmm0
 ; X86-NEXT:    retl
   %r = call <4 x float> @llvm.maximum.v4f32(<4 x float> %x, <4 x float> <float -0., float -0., float -0., float -0.>)
   ret <4 x float> %r
@@ -1265,42 +1217,21 @@
 define <2 x double> @test_fminimum_vector_zero_first(<2 x double> %x) {
 ; SSE2-LABEL: test_fminimum_vector_zero_first:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movaps %xmm0, %xmm1
-; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[3,3]
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    pcmpgtd %xmm1, %xmm2
-; SSE2-NEXT:    movaps %xmm0, %xmm1
-; SSE2-NEXT:    andps %xmm2, %xmm1
-; SSE2-NEXT:    andnps %xmm0, %xmm2
-; SSE2-NEXT:    movaps %xmm2, %xmm3
-; SSE2-NEXT:    minpd %xmm1, %xmm3
-; SSE2-NEXT:    movaps %xmm2, %xmm0
-; SSE2-NEXT:    cmpunordpd %xmm2, %xmm0
-; SSE2-NEXT:    andpd %xmm0, %xmm2
-; SSE2-NEXT:    andnpd %xmm3, %xmm0
-; SSE2-NEXT:    orpd %xmm2, %xmm0
+; SSE2-NEXT:    xorpd %xmm1, %xmm1
+; SSE2-NEXT:    minpd %xmm0, %xmm1
+; SSE2-NEXT:    movapd %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; AVX-LABEL: test_fminimum_vector_zero_first:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm1
-; AVX-NEXT:    vpand %xmm0, %xmm1, %xmm2
-; AVX-NEXT:    vpandn %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vminpd %xmm2, %xmm0, %xmm1
-; AVX-NEXT:    vcmpunordpd %xmm0, %xmm0, %xmm2
-; AVX-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
+; AVX-NEXT:    vminpd %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; X86-LABEL: test_fminimum_vector_zero_first:
 ; X86:       # %bb.0:
-; X86-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; X86-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm1
-; X86-NEXT:    vpand %xmm0, %xmm1, %xmm2
-; X86-NEXT:    vpandn %xmm0, %xmm1, %xmm0
-; X86-NEXT:    vminpd %xmm2, %xmm0, %xmm1
-; X86-NEXT:    vcmpunordpd %xmm0, %xmm0, %xmm2
-; X86-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
+; X86-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
+; X86-NEXT:    vminpd %xmm0, %xmm1, %xmm0
 ; X86-NEXT:    retl
   %r = call <2 x double> @llvm.minimum.v2f64(<2 x double> <double 0., double 0.>, <2 x double> %x)
   ret <2 x double> %r
@@ -1338,54 +1269,27 @@
 define <4 x float> @test_fmaximum_vector_signed_zero_first(<4 x float> %x) {
 ; SSE2-LABEL: test_fmaximum_vector_signed_zero_first:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]
-; SSE2-NEXT:    movdqa %xmm1, %xmm2
-; SSE2-NEXT:    pand %xmm0, %xmm2
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    pcmpgtd %xmm0, %xmm3
-; SSE2-NEXT:    movdqa %xmm3, %xmm4
-; SSE2-NEXT:    pandn %xmm0, %xmm4
-; SSE2-NEXT:    por %xmm2, %xmm4
-; SSE2-NEXT:    pand %xmm3, %xmm0
-; SSE2-NEXT:    pandn %xmm1, %xmm3
-; SSE2-NEXT:    por %xmm3, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    maxps %xmm4, %xmm1
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    cmpunordps %xmm0, %xmm2
-; SSE2-NEXT:    andps %xmm2, %xmm0
-; SSE2-NEXT:    andnps %xmm1, %xmm2
-; SSE2-NEXT:    orps %xmm2, %xmm0
+; SSE2-NEXT:    movaps {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]
+; SSE2-NEXT:    maxps %xmm0, %xmm1
+; SSE2-NEXT:    movaps %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; AVX1-LABEL: test_fmaximum_vector_signed_zero_first:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vmovaps {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]
-; AVX1-NEXT:    vblendvps %xmm0, %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vblendvps %xmm0, %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vmaxps %xmm2, %xmm0, %xmm1
-; AVX1-NEXT:    vcmpunordps %xmm0, %xmm0, %xmm2
-; AVX1-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vmaxps %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX512-LABEL: test_fmaximum_vector_signed_zero_first:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vbroadcastss {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]
-; AVX512-NEXT:    vblendvps %xmm0, %xmm1, %xmm0, %xmm2
-; AVX512-NEXT:    vblendvps %xmm0, %xmm0, %xmm1, %xmm0
-; AVX512-NEXT:    vmaxps %xmm2, %xmm0, %xmm1
-; AVX512-NEXT:    vcmpunordps %xmm0, %xmm0, %xmm2
-; AVX512-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; AVX512-NEXT:    vmaxps %xmm0, %xmm1, %xmm0
 ; AVX512-NEXT:    retq
 ;
 ; X86-LABEL: test_fmaximum_vector_signed_zero_first:
 ; X86:       # %bb.0:
 ; X86-NEXT:    vmovaps {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]
-; X86-NEXT:    vblendvps %xmm0, %xmm1, %xmm0, %xmm2
-; X86-NEXT:    vblendvps %xmm0, %xmm0, %xmm1, %xmm0
-; X86-NEXT:    vmaxps %xmm2, %xmm0, %xmm1
-; X86-NEXT:    vcmpunordps %xmm0, %xmm0, %xmm2
-; X86-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; X86-NEXT:    vmaxps %xmm0, %xmm1, %xmm0
 ; X86-NEXT:    retl
   %r = call <4 x float> @llvm.maximum.v4f32(<4 x float> <float -0., float -0., float -0., float -0.>, <4 x float> %x)
   ret <4 x float> %r