Index: llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
===================================================================
--- llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -7986,15 +7986,20 @@
     return DAG.getNode(NewOp, dl, VT, Quiet0, Quiet1, Node->getFlags());
   }
 
+  auto IsConstantNonZero = [](SDValue N) {
+    auto *Input = dyn_cast<ConstantFPSDNode>(N);
+    return Input && !Input->isZero();
+  };
   // If the target has FMINIMUM/FMAXIMUM but not FMINNUM/FMAXNUM use that
-  // instead if there are no NaNs.
-  if (Node->getFlags().hasNoNaNs()) {
+  // instead if there are no NaNs and at least one operand isn't +/-0.
+  if (Node->getFlags().hasNoNaNs() &&
+      (IsConstantNonZero(Node->getOperand(0)) ||
+       IsConstantNonZero(Node->getOperand(1)))) {
     unsigned IEEE2018Op =
         Node->getOpcode() == ISD::FMINNUM ? ISD::FMINIMUM : ISD::FMAXIMUM;
-    if (isOperationLegalOrCustom(IEEE2018Op, VT)) {
+    if (isOperationLegalOrCustom(IEEE2018Op, VT))
       return DAG.getNode(IEEE2018Op, dl, VT, Node->getOperand(0),
                          Node->getOperand(1), Node->getFlags());
-    }
   }
 
   if (SDValue SelCC = createSelectForFMINNUM_FMAXNUM(Node, DAG))
Index: llvm/test/CodeGen/ARM/lower-vmax.ll
===================================================================
--- llvm/test/CodeGen/ARM/lower-vmax.ll
+++ llvm/test/CodeGen/ARM/lower-vmax.ll
@@ -1,11 +1,21 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=arm-eabihf -mattr=+neon < %s | FileCheck -check-prefixes=CHECK-NO_NEON %s
 ; RUN: llc -mtriple=arm-eabihf -mattr=+neon,+neonfp < %s | FileCheck -check-prefixes=CHECK-NEON %s
 
 define float @max_f32(float, float) {
-;CHECK-NEON: vmax.f32
-;CHECK-NO_NEON: vcmp.f32
-;CHECK-NO_NEON: vmrs
-;CHECK-NO_NEON: vmovgt.f32
+; CHECK-NO_NEON-LABEL: max_f32:
+; CHECK-NO_NEON:       @ %bb.0:
+; CHECK-NO_NEON-NEXT:    vcmp.f32 s1, s0
+; CHECK-NO_NEON-NEXT:    vmrs APSR_nzcv, fpscr
+; CHECK-NO_NEON-NEXT:    vmovgt.f32 s0, s1
+; CHECK-NO_NEON-NEXT:    mov pc, lr
+;
+; CHECK-NEON-LABEL: max_f32:
+; CHECK-NEON:       @ %bb.0:
+; CHECK-NEON-NEXT:    vcmp.f32 s1, s0
+; CHECK-NEON-NEXT:    vmrs APSR_nzcv, fpscr
+; CHECK-NEON-NEXT:    vmovgt.f32 s0, s1
+; CHECK-NEON-NEXT:    mov pc, lr
   %3 = call nnan float @llvm.maxnum.f32(float %1, float %0)
   ret float %3
 }
@@ -13,10 +23,19 @@
 declare float @llvm.maxnum.f32(float, float) #1
 
 define float @min_f32(float, float) {
-;CHECK-NEON: vmin.f32
-;CHECK-NO_NEON: vcmp.f32
-;CHECK-NO_NEON: vmrs
-;CHECK-NO_NEON: vmovlt.f32
+; CHECK-NO_NEON-LABEL: min_f32:
+; CHECK-NO_NEON:       @ %bb.0:
+; CHECK-NO_NEON-NEXT:    vcmp.f32 s1, s0
+; CHECK-NO_NEON-NEXT:    vmrs APSR_nzcv, fpscr
+; CHECK-NO_NEON-NEXT:    vmovlt.f32 s0, s1
+; CHECK-NO_NEON-NEXT:    mov pc, lr
+;
+; CHECK-NEON-LABEL: min_f32:
+; CHECK-NEON:       @ %bb.0:
+; CHECK-NEON-NEXT:    vcmp.f32 s1, s0
+; CHECK-NEON-NEXT:    vmrs APSR_nzcv, fpscr
+; CHECK-NEON-NEXT:    vmovlt.f32 s0, s1
+; CHECK-NEON-NEXT:    mov pc, lr
   %3 = call nnan float @llvm.minnum.f32(float %1, float %0)
   ret float %3
 }
Index: llvm/test/CodeGen/WebAssembly/f32.ll
===================================================================
--- llvm/test/CodeGen/WebAssembly/f32.ll
+++ llvm/test/CodeGen/WebAssembly/f32.ll
@@ -207,14 +207,29 @@
 ; CHECK-LABEL: fminnum32_intrinsic:
 ; CHECK:         .functype fminnum32_intrinsic (f32, f32) -> (f32)
 ; CHECK-NEXT:  # %bb.0:
-; CHECK-NEXT:    local.get $push2=, 0
-; CHECK-NEXT:    local.get $push1=, 1
-; CHECK-NEXT:    f32.min $push0=, $pop2, $pop1
-; CHECK-NEXT:    return $pop0
+; CHECK-NEXT:    local.get $push5=, 0
+; CHECK-NEXT:    local.get $push4=, 1
+; CHECK-NEXT:    local.get $push3=, 0
+; CHECK-NEXT:    local.get $push2=, 1
+; CHECK-NEXT:    f32.lt $push0=, $pop3, $pop2
+; CHECK-NEXT:    f32.select $push1=, $pop5, $pop4, $pop0
+; CHECK-NEXT:    return $pop1
   %a = call nnan float @llvm.minnum.f32(float %x, float %y)
   ret float %a
 }
 
+define float @fminnum32_non_zero_intrinsic(float %x) {
+; CHECK-LABEL: fminnum32_non_zero_intrinsic:
+; CHECK:         .functype fminnum32_non_zero_intrinsic (f32) -> (f32)
+; CHECK-NEXT:  # %bb.0:
+; CHECK-NEXT:    local.get $push2=, 0
+; CHECK-NEXT:    f32.const $push0=, -0x1p0
+; CHECK-NEXT:    f32.min $push1=, $pop2, $pop0
+; CHECK-NEXT:    return $pop1
+  %a = call nnan float @llvm.minnum.f32(float %x, float -1.0)
+  ret float %a
+}
+
 declare float @llvm.maximum.f32(float, float)
 define float @fmax32_intrinsic(float %x, float %y) {
 ; CHECK-LABEL: fmax32_intrinsic:
@@ -233,14 +248,44 @@
 ; CHECK-LABEL: fmaxnum32_intrinsic:
 ; CHECK:         .functype fmaxnum32_intrinsic (f32, f32) -> (f32)
 ; CHECK-NEXT:  # %bb.0:
-; CHECK-NEXT:    local.get $push2=, 0
-; CHECK-NEXT:    local.get $push1=, 1
-; CHECK-NEXT:    f32.max $push0=, $pop2, $pop1
-; CHECK-NEXT:    return $pop0
+; CHECK-NEXT:    local.get $push5=, 0
+; CHECK-NEXT:    local.get $push4=, 1
+; CHECK-NEXT:    local.get $push3=, 0
+; CHECK-NEXT:    local.get $push2=, 1
+; CHECK-NEXT:    f32.gt $push0=, $pop3, $pop2
+; CHECK-NEXT:    f32.select $push1=, $pop5, $pop4, $pop0
+; CHECK-NEXT:    return $pop1
   %a = call nnan float @llvm.maxnum.f32(float %x, float %y)
   ret float %a
 }
 
+define float @fmaxnum32_zero_intrinsic(float %x) {
+; CHECK-LABEL: fmaxnum32_zero_intrinsic:
+; CHECK:         .functype fmaxnum32_zero_intrinsic (f32) -> (f32)
+; CHECK-NEXT:  # %bb.0:
+; CHECK-NEXT:    local.get $push5=, 0
+; CHECK-NEXT:    f32.const $push0=, 0x0p0
+; CHECK-NEXT:    local.get $push4=, 0
+; CHECK-NEXT:    f32.const $push3=, 0x0p0
+; CHECK-NEXT:    f32.gt $push1=, $pop4, $pop3
+; CHECK-NEXT:    f32.select $push2=, $pop5, $pop0, $pop1
+; CHECK-NEXT:    return $pop2
+  %a = call nnan float @llvm.maxnum.f32(float %x, float 0.0)
+  ret float %a
+}
+
+define float @fmaxnum32_non_zero_intrinsic(float %x) {
+; CHECK-LABEL: fmaxnum32_non_zero_intrinsic:
+; CHECK:         .functype fmaxnum32_non_zero_intrinsic (f32) -> (f32)
+; CHECK-NEXT:  # %bb.0:
+; CHECK-NEXT:    local.get $push2=, 0
+; CHECK-NEXT:    f32.const $push0=, 0x1p0
+; CHECK-NEXT:    f32.max $push1=, $pop2, $pop0
+; CHECK-NEXT:    return $pop1
+  %a = call nnan float @llvm.maxnum.f32(float %x, float 1.0)
+  ret float %a
+}
+
 define float @fma32(float %a, float %b, float %c) {
 ; CHECK-LABEL: fma32:
 ; CHECK:         .functype fma32 (f32, f32, f32) -> (f32)
Index: llvm/test/CodeGen/WebAssembly/f64.ll
===================================================================
--- llvm/test/CodeGen/WebAssembly/f64.ll
+++ llvm/test/CodeGen/WebAssembly/f64.ll
@@ -202,6 +202,49 @@
   ret double %a
 }
 
+declare double @llvm.minnum.f64(double, double)
+define double @fminnum64_intrinsic(double %x, double %y) {
+; CHECK-LABEL: fminnum64_intrinsic:
+; CHECK:         .functype fminnum64_intrinsic (f64, f64) -> (f64)
+; CHECK-NEXT:  # %bb.0:
+; CHECK-NEXT:    local.get $push5=, 0
+; CHECK-NEXT:    local.get $push4=, 1
+; CHECK-NEXT:    local.get $push3=, 0
+; CHECK-NEXT:    local.get $push2=, 1
+; CHECK-NEXT:    f64.lt $push0=, $pop3, $pop2
+; CHECK-NEXT:    f64.select $push1=, $pop5, $pop4, $pop0
+; CHECK-NEXT:    return $pop1
+  %a = call nnan double @llvm.minnum.f64(double %x, double %y)
+  ret double %a
+}
+
+define double @fminnum64_zero_intrinsic(double %x) {
+; CHECK-LABEL: fminnum64_zero_intrinsic:
+; CHECK:         .functype fminnum64_zero_intrinsic (f64) -> (f64)
+; CHECK-NEXT:  # %bb.0:
+; CHECK-NEXT:    local.get $push5=, 0
+; CHECK-NEXT:    f64.const $push0=, -0x0p0
+; CHECK-NEXT:    local.get $push4=, 0
+; CHECK-NEXT:    f64.const $push3=, -0x0p0
+; CHECK-NEXT:    f64.lt $push1=, $pop4, $pop3
+; CHECK-NEXT:    f64.select $push2=, $pop5, $pop0, $pop1
+; CHECK-NEXT:    return $pop2
+  %a = call nnan double @llvm.minnum.f64(double %x, double -0.0)
+  ret double %a
+}
+
+define double @fminnum64_non_zero_intrinsic(double %x) {
+; CHECK-LABEL: fminnum64_non_zero_intrinsic:
+; CHECK:         .functype fminnum64_non_zero_intrinsic (f64) -> (f64)
+; CHECK-NEXT:  # %bb.0:
+; CHECK-NEXT:    local.get $push2=, 0
+; CHECK-NEXT:    f64.const $push0=, -0x1p0
+; CHECK-NEXT:    f64.min $push1=, $pop2, $pop0
+; CHECK-NEXT:    return $pop1
+  %a = call nnan double @llvm.minnum.f64(double %x, double -1.0)
+  ret double %a
+}
+
 declare double @llvm.maximum.f64(double, double)
 define double @fmax64_intrinsic(double %x, double %y) {
 ; CHECK-LABEL: fmax64_intrinsic:
@@ -215,6 +258,49 @@
   ret double %a
 }
 
+declare double @llvm.maxnum.f64(double, double)
+define double@fmaxnum64_intrinsic(double %x, double %y) {
+; CHECK-LABEL: fmaxnum64_intrinsic:
+; CHECK:         .functype fmaxnum64_intrinsic (f64, f64) -> (f64)
+; CHECK-NEXT:  # %bb.0:
+; CHECK-NEXT:    local.get $push5=, 0
+; CHECK-NEXT:    local.get $push4=, 1
+; CHECK-NEXT:    local.get $push3=, 0
+; CHECK-NEXT:    local.get $push2=, 1
+; CHECK-NEXT:    f64.gt $push0=, $pop3, $pop2
+; CHECK-NEXT:    f64.select $push1=, $pop5, $pop4, $pop0
+; CHECK-NEXT:    return $pop1
+  %a = call nnan double @llvm.maxnum.f64(double %x, double %y)
+  ret double %a
+}
+
+define double @fmaxnum64_zero_intrinsic(double %x) {
+; CHECK-LABEL: fmaxnum64_zero_intrinsic:
+; CHECK:         .functype fmaxnum64_zero_intrinsic (f64) -> (f64)
+; CHECK-NEXT:  # %bb.0:
+; CHECK-NEXT:    local.get $push5=, 0
+; CHECK-NEXT:    f64.const $push0=, 0x0p0
+; CHECK-NEXT:    local.get $push4=, 0
+; CHECK-NEXT:    f64.const $push3=, 0x0p0
+; CHECK-NEXT:    f64.gt $push1=, $pop4, $pop3
+; CHECK-NEXT:    f64.select $push2=, $pop5, $pop0, $pop1
+; CHECK-NEXT:    return $pop2
+  %a = call nnan double @llvm.maxnum.f64(double %x, double 0.0)
+  ret double %a
+}
+
+define double @fmaxnum64_non_zero_intrinsic(double %x) {
+; CHECK-LABEL: fmaxnum64_non_zero_intrinsic:
+; CHECK:         .functype fmaxnum64_non_zero_intrinsic (f64) -> (f64)
+; CHECK-NEXT:  # %bb.0:
+; CHECK-NEXT:    local.get $push2=, 0
+; CHECK-NEXT:    f64.const $push0=, 0x1p0
+; CHECK-NEXT:    f64.max $push1=, $pop2, $pop0
+; CHECK-NEXT:    return $pop1
+  %a = call nnan double @llvm.maxnum.f64(double %x, double 1.0)
+  ret double %a
+}
+
 define double @fma64(double %a, double %b, double %c) {
 ; CHECK-LABEL: fma64:
 ; CHECK:         .functype fma64 (f64, f64, f64) -> (f64)
Index: llvm/test/CodeGen/WebAssembly/simd-arith.ll
===================================================================
--- llvm/test/CodeGen/WebAssembly/simd-arith.ll
+++ llvm/test/CodeGen/WebAssembly/simd-arith.ll
@@ -13422,13 +13422,67 @@
 ; SIMD128-LABEL: minnum_intrinsic_v4f32:
 ; SIMD128:         .functype minnum_intrinsic_v4f32 (v128, v128) -> (v128)
 ; SIMD128-NEXT:  # %bb.0:
-; SIMD128-NEXT:    f32x4.min $push0=, $0, $1
-; SIMD128-NEXT:    return $pop0
+; SIMD128-NEXT:    f32x4.extract_lane $push27=, $0, 0
+; SIMD128-NEXT:    local.tee $push26=, $3=, $pop27
+; SIMD128-NEXT:    f32x4.extract_lane $push25=, $1, 0
+; SIMD128-NEXT:    local.tee $push24=, $2=, $pop25
+; SIMD128-NEXT:    f32.lt $push2=, $3, $2
+; SIMD128-NEXT:    f32.select $push3=, $pop26, $pop24, $pop2
+; SIMD128-NEXT:    f32x4.splat $push4=, $pop3
+; SIMD128-NEXT:    f32x4.extract_lane $push23=, $0, 1
+; SIMD128-NEXT:    local.tee $push22=, $3=, $pop23
+; SIMD128-NEXT:    f32x4.extract_lane $push21=, $1, 1
+; SIMD128-NEXT:    local.tee $push20=, $2=, $pop21
+; SIMD128-NEXT:    f32.lt $push0=, $3, $2
+; SIMD128-NEXT:    f32.select $push1=, $pop22, $pop20, $pop0
+; SIMD128-NEXT:    f32x4.replace_lane $push5=, $pop4, 1, $pop1
+; SIMD128-NEXT:    f32x4.extract_lane $push19=, $0, 2
+; SIMD128-NEXT:    local.tee $push18=, $3=, $pop19
+; SIMD128-NEXT:    f32x4.extract_lane $push17=, $1, 2
+; SIMD128-NEXT:    local.tee $push16=, $2=, $pop17
+; SIMD128-NEXT:    f32.lt $push6=, $3, $2
+; SIMD128-NEXT:    f32.select $push7=, $pop18, $pop16, $pop6
+; SIMD128-NEXT:    f32x4.replace_lane $push8=, $pop5, 2, $pop7
+; SIMD128-NEXT:    f32x4.extract_lane $push15=, $0, 3
+; SIMD128-NEXT:    local.tee $push14=, $3=, $pop15
+; SIMD128-NEXT:    f32x4.extract_lane $push13=, $1, 3
+; SIMD128-NEXT:    local.tee $push12=, $2=, $pop13
+; SIMD128-NEXT:    f32.lt $push9=, $3, $2
+; SIMD128-NEXT:    f32.select $push10=, $pop14, $pop12, $pop9
+; SIMD128-NEXT:    f32x4.replace_lane $push11=, $pop8, 3, $pop10
+; SIMD128-NEXT:    return $pop11
 ;
 ; SIMD128-FAST-LABEL: minnum_intrinsic_v4f32:
 ; SIMD128-FAST:         .functype minnum_intrinsic_v4f32 (v128, v128) -> (v128)
 ; SIMD128-FAST-NEXT:  # %bb.0:
-; SIMD128-FAST-NEXT:    f32x4.min $push0=, $0, $1
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push27=, $0, 0
+; SIMD128-FAST-NEXT:    local.tee $push26=, $3=, $pop27
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push25=, $1, 0
+; SIMD128-FAST-NEXT:    local.tee $push24=, $2=, $pop25
+; SIMD128-FAST-NEXT:    f32.lt $push3=, $3, $2
+; SIMD128-FAST-NEXT:    f32.select $push4=, $pop26, $pop24, $pop3
+; SIMD128-FAST-NEXT:    f32x4.splat $push5=, $pop4
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push23=, $0, 1
+; SIMD128-FAST-NEXT:    local.tee $push22=, $3=, $pop23
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push21=, $1, 1
+; SIMD128-FAST-NEXT:    local.tee $push20=, $2=, $pop21
+; SIMD128-FAST-NEXT:    f32.lt $push1=, $3, $2
+; SIMD128-FAST-NEXT:    f32.select $push2=, $pop22, $pop20, $pop1
+; SIMD128-FAST-NEXT:    f32x4.replace_lane $push6=, $pop5, 1, $pop2
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push19=, $0, 2
+; SIMD128-FAST-NEXT:    local.tee $push18=, $3=, $pop19
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push17=, $1, 2
+; SIMD128-FAST-NEXT:    local.tee $push16=, $2=, $pop17
+; SIMD128-FAST-NEXT:    f32.lt $push7=, $3, $2
+; SIMD128-FAST-NEXT:    f32.select $push8=, $pop18, $pop16, $pop7
+; SIMD128-FAST-NEXT:    f32x4.replace_lane $push9=, $pop6, 2, $pop8
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push15=, $0, 3
+; SIMD128-FAST-NEXT:    local.tee $push14=, $3=, $pop15
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push13=, $1, 3
+; SIMD128-FAST-NEXT:    local.tee $push12=, $2=, $pop13
+; SIMD128-FAST-NEXT:    f32.lt $push10=, $3, $2
+; SIMD128-FAST-NEXT:    f32.select $push11=, $pop14, $pop12, $pop10
+; SIMD128-FAST-NEXT:    f32x4.replace_lane $push0=, $pop9, 3, $pop11
 ; SIMD128-FAST-NEXT:    return $pop0
 ;
 ; NO-SIMD128-LABEL: minnum_intrinsic_v4f32:
@@ -13516,13 +13570,67 @@
 ; SIMD128-LABEL: maxnum_intrinsic_v4f32:
 ; SIMD128:         .functype maxnum_intrinsic_v4f32 (v128, v128) -> (v128)
 ; SIMD128-NEXT:  # %bb.0:
-; SIMD128-NEXT:    f32x4.max $push0=, $0, $1
-; SIMD128-NEXT:    return $pop0
+; SIMD128-NEXT:    f32x4.extract_lane $push27=, $0, 0
+; SIMD128-NEXT:    local.tee $push26=, $3=, $pop27
+; SIMD128-NEXT:    f32x4.extract_lane $push25=, $1, 0
+; SIMD128-NEXT:    local.tee $push24=, $2=, $pop25
+; SIMD128-NEXT:    f32.gt $push2=, $3, $2
+; SIMD128-NEXT:    f32.select $push3=, $pop26, $pop24, $pop2
+; SIMD128-NEXT:    f32x4.splat $push4=, $pop3
+; SIMD128-NEXT:    f32x4.extract_lane $push23=, $0, 1
+; SIMD128-NEXT:    local.tee $push22=, $3=, $pop23
+; SIMD128-NEXT:    f32x4.extract_lane $push21=, $1, 1
+; SIMD128-NEXT:    local.tee $push20=, $2=, $pop21
+; SIMD128-NEXT:    f32.gt $push0=, $3, $2
+; SIMD128-NEXT:    f32.select $push1=, $pop22, $pop20, $pop0
+; SIMD128-NEXT:    f32x4.replace_lane $push5=, $pop4, 1, $pop1
+; SIMD128-NEXT:    f32x4.extract_lane $push19=, $0, 2
+; SIMD128-NEXT:    local.tee $push18=, $3=, $pop19
+; SIMD128-NEXT:    f32x4.extract_lane $push17=, $1, 2
+; SIMD128-NEXT:    local.tee $push16=, $2=, $pop17
+; SIMD128-NEXT:    f32.gt $push6=, $3, $2
+; SIMD128-NEXT:    f32.select $push7=, $pop18, $pop16, $pop6
+; SIMD128-NEXT:    f32x4.replace_lane $push8=, $pop5, 2, $pop7
+; SIMD128-NEXT:    f32x4.extract_lane $push15=, $0, 3
+; SIMD128-NEXT:    local.tee $push14=, $3=, $pop15
+; SIMD128-NEXT:    f32x4.extract_lane $push13=, $1, 3
+; SIMD128-NEXT:    local.tee $push12=, $2=, $pop13
+; SIMD128-NEXT:    f32.gt $push9=, $3, $2
+; SIMD128-NEXT:    f32.select $push10=, $pop14, $pop12, $pop9
+; SIMD128-NEXT:    f32x4.replace_lane $push11=, $pop8, 3, $pop10
+; SIMD128-NEXT:    return $pop11
 ;
 ; SIMD128-FAST-LABEL: maxnum_intrinsic_v4f32:
 ; SIMD128-FAST:         .functype maxnum_intrinsic_v4f32 (v128, v128) -> (v128)
 ; SIMD128-FAST-NEXT:  # %bb.0:
-; SIMD128-FAST-NEXT:    f32x4.max $push0=, $0, $1
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push27=, $0, 0
+; SIMD128-FAST-NEXT:    local.tee $push26=, $3=, $pop27
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push25=, $1, 0
+; SIMD128-FAST-NEXT:    local.tee $push24=, $2=, $pop25
+; SIMD128-FAST-NEXT:    f32.gt $push3=, $3, $2
+; SIMD128-FAST-NEXT:    f32.select $push4=, $pop26, $pop24, $pop3
+; SIMD128-FAST-NEXT:    f32x4.splat $push5=, $pop4
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push23=, $0, 1
+; SIMD128-FAST-NEXT:    local.tee $push22=, $3=, $pop23
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push21=, $1, 1
+; SIMD128-FAST-NEXT:    local.tee $push20=, $2=, $pop21
+; SIMD128-FAST-NEXT:    f32.gt $push1=, $3, $2
+; SIMD128-FAST-NEXT:    f32.select $push2=, $pop22, $pop20, $pop1
+; SIMD128-FAST-NEXT:    f32x4.replace_lane $push6=, $pop5, 1, $pop2
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push19=, $0, 2
+; SIMD128-FAST-NEXT:    local.tee $push18=, $3=, $pop19
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push17=, $1, 2
+; SIMD128-FAST-NEXT:    local.tee $push16=, $2=, $pop17
+; SIMD128-FAST-NEXT:    f32.gt $push7=, $3, $2
+; SIMD128-FAST-NEXT:    f32.select $push8=, $pop18, $pop16, $pop7
+; SIMD128-FAST-NEXT:    f32x4.replace_lane $push9=, $pop6, 2, $pop8
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push15=, $0, 3
+; SIMD128-FAST-NEXT:    local.tee $push14=, $3=, $pop15
+; SIMD128-FAST-NEXT:    f32x4.extract_lane $push13=, $1, 3
+; SIMD128-FAST-NEXT:    local.tee $push12=, $2=, $pop13
+; SIMD128-FAST-NEXT:    f32.gt $push10=, $3, $2
+; SIMD128-FAST-NEXT:    f32.select $push11=, $pop14, $pop12, $pop10
+; SIMD128-FAST-NEXT:    f32x4.replace_lane $push0=, $pop9, 3, $pop11
 ; SIMD128-FAST-NEXT:    return $pop0
 ;
 ; NO-SIMD128-LABEL: maxnum_intrinsic_v4f32: