[Hexagon] Generate vector min/max for HVX

llvm-svn: 369014

lib/Target/Hexagon/HexagonISelLowering.cpp

@@ -2877,6 +2877,17 @@ HexagonTargetLowering::ReplaceNodeResults(SDNode *N,
   }
 }
 
+SDValue
+HexagonTargetLowering::PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI)
+      const {
+  SDValue Op(N, 0);
+  if (isHvxOperation(Op)) {
+    if (SDValue V = PerformHvxDAGCombine(N, DCI))
+      return V;
+  }
+  return SDValue();
+}
+
 /// Returns relocation base for the given PIC jumptable.
 SDValue
 HexagonTargetLowering::getPICJumpTableRelocBase(SDValue Table,

lib/Target/Hexagon/HexagonISelLowering.h

@@ -223,6 +223,8 @@ namespace HexagonISD {
                         const SmallVectorImpl<SDValue> &OutVals,
                         const SDLoc &dl, SelectionDAG &DAG) const override;
 
+    SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const override;
+
     bool mayBeEmittedAsTailCall(const CallInst *CI) const override;
 
     unsigned getRegisterByName(const char* RegName, EVT VT,
@@ -301,7 +303,8 @@ namespace HexagonISD {
         const AttributeList &FuncAttributes) const override;
 
     bool allowsMisalignedMemoryAccesses(EVT VT, unsigned AddrSpace,
-        unsigned Align, MachineMemOperand::Flags Flags, bool *Fast) const override;
+        unsigned Align, MachineMemOperand::Flags Flags, bool *Fast)
+        const override;
 
     /// Returns relocation base for the given PIC jumptable.
     SDValue getPICJumpTableRelocBase(SDValue Table, SelectionDAG &DAG)
@@ -458,6 +461,8 @@ namespace HexagonISD {
 
     bool isHvxOperation(SDValue Op) const;
     SDValue LowerHvxOperation(SDValue Op, SelectionDAG &DAG) const;
+
+    SDValue PerformHvxDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
   };
 
 } // end namespace llvm

lib/Target/Hexagon/HexagonISelLoweringHVX.cpp

@@ -193,6 +193,8 @@ HexagonTargetLowering::initializeHVXLowering() {
     setOperationAction(ISD::OR,                 BoolV, Legal);
     setOperationAction(ISD::XOR,                BoolV, Legal);
   }
+
+  setTargetDAGCombine(ISD::VSELECT);
 }
 
 SDValue
@@ -1580,6 +1582,28 @@ HexagonTargetLowering::LowerHvxOperation(SDValue Op, SelectionDAG &DAG) const {
   llvm_unreachable("Unhandled HVX operation");
 }
 
+SDValue
+HexagonTargetLowering::PerformHvxDAGCombine(SDNode *N, DAGCombinerInfo &DCI)
+      const {
+  const SDLoc &dl(N);
+  SDValue Op(N, 0);
+
+  unsigned Opc = Op.getOpcode();
+  if (Opc == ISD::VSELECT) {
+    // (vselect (xor x, qtrue), v0, v1) -> (vselect x, v1, v0)
+    SDValue Cond = Op.getOperand(0);
+    if (Cond->getOpcode() == ISD::XOR) {
+      SDValue C0 = Cond.getOperand(0), C1 = Cond.getOperand(1);
+      if (C1->getOpcode() == HexagonISD::QTRUE) {
+        SDValue VSel = DCI.DAG.getNode(ISD::VSELECT, dl, ty(Op), C0,
+                                       Op.getOperand(2), Op.getOperand(1));
+        return VSel;
+      }
+    }
+  }
+  return SDValue();
+}
+
 bool
 HexagonTargetLowering::isHvxOperation(SDValue Op) const {
   // If the type of the result, or any operand type are HVX vector types,

lib/Target/Hexagon/HexagonPatternsHVX.td

@@ -259,6 +259,31 @@ class Vneg1<ValueType VecTy>
 class Vnot<ValueType VecTy>
   : PatFrag<(ops node:$Vs), (xor $Vs, Vneg1<VecTy>)>;
 
+multiclass VMinMax_pat<InstHexagon MinInst, InstHexagon MaxInst, PatFrag CmpOp,
+                       ValueType CmpType, PatFrag CmpPred> {
+  def: Pat<(vselect (CmpType (CmpOp CmpPred:$Vs, CmpPred:$Vt)),
+                             CmpPred:$Vt, CmpPred:$Vs),
+           (MinInst CmpPred:$Vs, CmpPred:$Vt)>;
+  def: Pat<(vselect (CmpType (CmpOp CmpPred:$Vs, CmpPred:$Vt)),
+                             CmpPred:$Vs, CmpPred:$Vt),
+           (MaxInst CmpPred:$Vs, CmpPred:$Vt)>;
+}
+
+let Predicates = [UseHVX] in {
+  let AddedComplexity = 220 in {
+    defm: VMinMax_pat<V6_vminb,  V6_vmaxb,   setgt,  VecQ8,  HVI8>;
+    defm: VMinMax_pat<V6_vminb,  V6_vmaxb,   setge,  VecQ8,  HVI8>;
+    defm: VMinMax_pat<V6_vminub, V6_vmaxub, setugt,  VecQ8,  HVI8>;
+    defm: VMinMax_pat<V6_vminub, V6_vmaxub, setuge,  VecQ8,  HVI8>;
+    defm: VMinMax_pat<V6_vminh,  V6_vmaxh,   setgt, VecQ16, HVI16>;
+    defm: VMinMax_pat<V6_vminh,  V6_vmaxh,   setge, VecQ16, HVI16>;
+    defm: VMinMax_pat<V6_vminuh, V6_vmaxuh, setugt, VecQ16, HVI16>;
+    defm: VMinMax_pat<V6_vminuh, V6_vmaxuh, setuge, VecQ16, HVI16>;
+    defm: VMinMax_pat<V6_vminw,  V6_vmaxw,   setgt, VecQ32, HVI32>;
+    defm: VMinMax_pat<V6_vminw,  V6_vmaxw,   setge, VecQ32, HVI32>;
+  }
+}
+
 let Predicates = [UseHVX] in {
   let AddedComplexity = 200 in {
     def: Pat<(Vnot<VecI8>   HVI8:$Vs), (V6_vnot HvxVR:$Vs)>;

test/CodeGen/Hexagon/autohvx/minmax-128b.ll

@@ -0,0 +1,344 @@
+; RUN: llc -march=hexagon < %s | FileCheck %s
+
+; minb
+
+; CHECK: test_00:
+; CHECK: v0.b = vmin(v1.b,v0.b)
+define <128 x i8> @test_00(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp slt <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  ret <128 x i8> %t1
+}
+
+; CHECK: test_01:
+; CHECK: v0.b = vmin(v0.b,v1.b)
+define <128 x i8> @test_01(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp sle <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  ret <128 x i8> %t1
+}
+
+; CHECK: test_02:
+; CHECK: v0.b = vmin(v0.b,v1.b)
+define <128 x i8> @test_02(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp sgt <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v0
+  ret <128 x i8> %t1
+}
+
+; CHECK: test_03:
+; CHECK: v0.b = vmin(v1.b,v0.b)
+define <128 x i8> @test_03(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp sge <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v0
+  ret <128 x i8> %t1
+}
+
+; maxb
+
+; CHECK: test_04:
+; CHECK: v0.b = vmax(v1.b,v0.b)
+define <128 x i8> @test_04(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp slt <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v0
+  ret <128 x i8> %t1
+}
+
+; CHECK: test_05:
+; CHECK: v0.b = vmax(v0.b,v1.b)
+define <128 x i8> @test_05(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp sle <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v0
+  ret <128 x i8> %t1
+}
+
+; CHECK: test_06:
+; CHECK: v0.b = vmax(v0.b,v1.b)
+define <128 x i8> @test_06(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp sgt <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  ret <128 x i8> %t1
+}
+
+; CHECK: test_07:
+; CHECK: v0.b = vmax(v1.b,v0.b)
+define <128 x i8> @test_07(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp sge <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  ret <128 x i8> %t1
+}
+
+; minub
+
+; CHECK: test_08:
+; CHECK: v0.ub = vmin(v1.ub,v0.ub)
+define <128 x i8> @test_08(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp ult <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  ret <128 x i8> %t1
+}
+
+; CHECK: test_09:
+; CHECK: v0.ub = vmin(v0.ub,v1.ub)
+define <128 x i8> @test_09(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp ule <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  ret <128 x i8> %t1
+}
+
+; CHECK: test_0a:
+; CHECK: v0.ub = vmin(v0.ub,v1.ub)
+define <128 x i8> @test_0a(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp ugt <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v0
+  ret <128 x i8> %t1
+}
+
+; CHECK: test_0b:
+; CHECK: v0.ub = vmin(v1.ub,v0.ub)
+define <128 x i8> @test_0b(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp uge <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v0
+  ret <128 x i8> %t1
+}
+
+; maxub
+
+; CHECK: test_0c:
+; CHECK: v0.ub = vmax(v1.ub,v0.ub)
+define <128 x i8> @test_0c(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp ult <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v0
+  ret <128 x i8> %t1
+}
+
+; CHECK: test_0d:
+; CHECK: v0.ub = vmax(v0.ub,v1.ub)
+define <128 x i8> @test_0d(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp ule <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v0
+  ret <128 x i8> %t1
+}
+
+; CHECK: test_0e:
+; CHECK: v0.ub = vmax(v0.ub,v1.ub)
+define <128 x i8> @test_0e(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp ugt <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  ret <128 x i8> %t1
+}
+
+; CHECK: test_0f:
+; CHECK: v0.ub = vmax(v1.ub,v0.ub)
+define <128 x i8> @test_0f(<128 x i8> %v0, <128 x i8> %v1) #0 {
+  %t0 = icmp uge <128 x i8> %v0, %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  ret <128 x i8> %t1
+}
+
+; minh
+
+; CHECK: test_10:
+; CHECK: v0.h = vmin(v1.h,v0.h)
+define <64 x i16> @test_10(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp slt <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  ret <64 x i16> %t1
+}
+
+; CHECK: test_11:
+; CHECK: v0.h = vmin(v0.h,v1.h)
+define <64 x i16> @test_11(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp sle <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  ret <64 x i16> %t1
+}
+
+; CHECK: test_12:
+; CHECK: v0.h = vmin(v0.h,v1.h)
+define <64 x i16> @test_12(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp sgt <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v0
+  ret <64 x i16> %t1
+}
+
+; CHECK: test_13:
+; CHECK: v0.h = vmin(v1.h,v0.h)
+define <64 x i16> @test_13(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp sge <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v0
+  ret <64 x i16> %t1
+}
+
+; maxh
+
+; CHECK: test_14:
+; CHECK: v0.h = vmax(v1.h,v0.h)
+define <64 x i16> @test_14(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp slt <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v0
+  ret <64 x i16> %t1
+}
+
+; CHECK: test_15:
+; CHECK: v0.h = vmax(v0.h,v1.h)
+define <64 x i16> @test_15(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp sle <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v0
+  ret <64 x i16> %t1
+}
+
+; CHECK: test_16:
+; CHECK: v0.h = vmax(v0.h,v1.h)
+define <64 x i16> @test_16(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp sgt <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  ret <64 x i16> %t1
+}
+
+; CHECK: test_17:
+; CHECK: v0.h = vmax(v1.h,v0.h)
+define <64 x i16> @test_17(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp sge <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  ret <64 x i16> %t1
+}
+
+; minuh
+
+; CHECK: test_18:
+; CHECK: v0.uh = vmin(v1.uh,v0.uh)
+define <64 x i16> @test_18(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp ult <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  ret <64 x i16> %t1
+}
+
+; CHECK: test_19:
+; CHECK: v0.uh = vmin(v0.uh,v1.uh)
+define <64 x i16> @test_19(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp ule <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  ret <64 x i16> %t1
+}
+
+; CHECK: test_1a:
+; CHECK: v0.uh = vmin(v0.uh,v1.uh)
+define <64 x i16> @test_1a(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp ugt <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v0
+  ret <64 x i16> %t1
+}
+
+; CHECK: test_1b:
+; CHECK: v0.uh = vmin(v1.uh,v0.uh)
+define <64 x i16> @test_1b(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp uge <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v0
+  ret <64 x i16> %t1
+}
+
+; maxuh
+
+; CHECK: test_1c:
+; CHECK: v0.uh = vmax(v1.uh,v0.uh)
+define <64 x i16> @test_1c(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp ult <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v0
+  ret <64 x i16> %t1
+}
+
+; CHECK: test_1d:
+; CHECK: v0.uh = vmax(v0.uh,v1.uh)
+define <64 x i16> @test_1d(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp ule <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v0
+  ret <64 x i16> %t1
+}
+
+; CHECK: test_1e:
+; CHECK: v0.uh = vmax(v0.uh,v1.uh)
+define <64 x i16> @test_1e(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp ugt <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  ret <64 x i16> %t1
+}
+
+; CHECK: test_1f:
+; CHECK: v0.uh = vmax(v1.uh,v0.uh)
+define <64 x i16> @test_1f(<64 x i16> %v0, <64 x i16> %v1) #0 {
+  %t0 = icmp uge <64 x i16> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  ret <64 x i16> %t1
+}
+
+; minw
+
+; CHECK: test_20:
+; CHECK: v0.w = vmin(v1.w,v0.w)
+define <32 x i32> @test_20(<32 x i32> %v0, <32 x i32> %v1) #0 {
+  %t0 = icmp slt <32 x i32> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v0, <32 x i32> %v1
+  ret <32 x i32> %t1
+}
+
+; CHECK: test_21:
+; CHECK: v0.w = vmin(v0.w,v1.w)
+define <32 x i32> @test_21(<32 x i32> %v0, <32 x i32> %v1) #0 {
+  %t0 = icmp sle <32 x i32> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v0, <32 x i32> %v1
+  ret <32 x i32> %t1
+}
+
+; CHECK: test_22:
+; CHECK: v0.w = vmin(v0.w,v1.w)
+define <32 x i32> @test_22(<32 x i32> %v0, <32 x i32> %v1) #0 {
+  %t0 = icmp sgt <32 x i32> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v1, <32 x i32> %v0
+  ret <32 x i32> %t1
+}
+
+; CHECK: test_23:
+; CHECK: v0.w = vmin(v1.w,v0.w)
+define <32 x i32> @test_23(<32 x i32> %v0, <32 x i32> %v1) #0 {
+  %t0 = icmp sge <32 x i32> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v1, <32 x i32> %v0
+  ret <32 x i32> %t1
+}
+
+; maxw
+
+; CHECK: test_24:
+; CHECK: v0.w = vmax(v1.w,v0.w)
+define <32 x i32> @test_24(<32 x i32> %v0, <32 x i32> %v1) #0 {
+  %t0 = icmp slt <32 x i32> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v1, <32 x i32> %v0
+  ret <32 x i32> %t1
+}
+
+; CHECK: test_25:
+; CHECK: v0.w = vmax(v0.w,v1.w)
+define <32 x i32> @test_25(<32 x i32> %v0, <32 x i32> %v1) #0 {
+  %t0 = icmp sle <32 x i32> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v1, <32 x i32> %v0
+  ret <32 x i32> %t1
+}
+
+; CHECK: test_26:
+; CHECK: v0.w = vmax(v0.w,v1.w)
+define <32 x i32> @test_26(<32 x i32> %v0, <32 x i32> %v1) #0 {
+  %t0 = icmp sgt <32 x i32> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v0, <32 x i32> %v1
+  ret <32 x i32> %t1
+}
+
+; CHECK: test_27:
+; CHECK: v0.w = vmax(v1.w,v0.w)
+define <32 x i32> @test_27(<32 x i32> %v0, <32 x i32> %v1) #0 {
+  %t0 = icmp sge <32 x i32> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v0, <32 x i32> %v1
+  ret <32 x i32> %t1
+}
+
+attributes #0 = { readnone nounwind "target-cpu"="hexagonv62" "target-features"="+hvx,+hvx-length128b" }
+

test/CodeGen/Hexagon/autohvx/minmax-64b.ll

@@ -0,0 +1,344 @@
+; RUN: llc -march=hexagon < %s | FileCheck %s
+
+; minb
+
+; CHECK: test_00:
+; CHECK: v0.b = vmin(v1.b,v0.b)
+define <64 x i8> @test_00(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp slt <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  ret <64 x i8> %t1
+}
+
+; CHECK: test_01:
+; CHECK: v0.b = vmin(v0.b,v1.b)
+define <64 x i8> @test_01(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp sle <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  ret <64 x i8> %t1
+}
+
+; CHECK: test_02:
+; CHECK: v0.b = vmin(v0.b,v1.b)
+define <64 x i8> @test_02(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp sgt <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v0
+  ret <64 x i8> %t1
+}
+
+; CHECK: test_03:
+; CHECK: v0.b = vmin(v1.b,v0.b)
+define <64 x i8> @test_03(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp sge <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v0
+  ret <64 x i8> %t1
+}
+
+; maxb
+
+; CHECK: test_04:
+; CHECK: v0.b = vmax(v1.b,v0.b)
+define <64 x i8> @test_04(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp slt <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v0
+  ret <64 x i8> %t1
+}
+
+; CHECK: test_05:
+; CHECK: v0.b = vmax(v0.b,v1.b)
+define <64 x i8> @test_05(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp sle <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v0
+  ret <64 x i8> %t1
+}
+
+; CHECK: test_06:
+; CHECK: v0.b = vmax(v0.b,v1.b)
+define <64 x i8> @test_06(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp sgt <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  ret <64 x i8> %t1
+}
+
+; CHECK: test_07:
+; CHECK: v0.b = vmax(v1.b,v0.b)
+define <64 x i8> @test_07(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp sge <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  ret <64 x i8> %t1
+}
+
+; minub
+
+; CHECK: test_08:
+; CHECK: v0.ub = vmin(v1.ub,v0.ub)
+define <64 x i8> @test_08(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp ult <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  ret <64 x i8> %t1
+}
+
+; CHECK: test_09:
+; CHECK: v0.ub = vmin(v0.ub,v1.ub)
+define <64 x i8> @test_09(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp ule <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  ret <64 x i8> %t1
+}
+
+; CHECK: test_0a:
+; CHECK: v0.ub = vmin(v0.ub,v1.ub)
+define <64 x i8> @test_0a(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp ugt <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v0
+  ret <64 x i8> %t1
+}
+
+; CHECK: test_0b:
+; CHECK: v0.ub = vmin(v1.ub,v0.ub)
+define <64 x i8> @test_0b(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp uge <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v0
+  ret <64 x i8> %t1
+}
+
+; maxub
+
+; CHECK: test_0c:
+; CHECK: v0.ub = vmax(v1.ub,v0.ub)
+define <64 x i8> @test_0c(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp ult <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v0
+  ret <64 x i8> %t1
+}
+
+; CHECK: test_0d:
+; CHECK: v0.ub = vmax(v0.ub,v1.ub)
+define <64 x i8> @test_0d(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp ule <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v0
+  ret <64 x i8> %t1
+}
+
+; CHECK: test_0e:
+; CHECK: v0.ub = vmax(v0.ub,v1.ub)
+define <64 x i8> @test_0e(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp ugt <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  ret <64 x i8> %t1
+}
+
+; CHECK: test_0f:
+; CHECK: v0.ub = vmax(v1.ub,v0.ub)
+define <64 x i8> @test_0f(<64 x i8> %v0, <64 x i8> %v1) #0 {
+  %t0 = icmp uge <64 x i8> %v0, %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  ret <64 x i8> %t1
+}
+
+; minh
+
+; CHECK: test_10:
+; CHECK: v0.h = vmin(v1.h,v0.h)
+define <32 x i16> @test_10(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp slt <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  ret <32 x i16> %t1
+}
+
+; CHECK: test_11:
+; CHECK: v0.h = vmin(v0.h,v1.h)
+define <32 x i16> @test_11(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp sle <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  ret <32 x i16> %t1
+}
+
+; CHECK: test_12:
+; CHECK: v0.h = vmin(v0.h,v1.h)
+define <32 x i16> @test_12(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp sgt <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v0
+  ret <32 x i16> %t1
+}
+
+; CHECK: test_13:
+; CHECK: v0.h = vmin(v1.h,v0.h)
+define <32 x i16> @test_13(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp sge <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v0
+  ret <32 x i16> %t1
+}
+
+; maxh
+
+; CHECK: test_14:
+; CHECK: v0.h = vmax(v1.h,v0.h)
+define <32 x i16> @test_14(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp slt <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v0
+  ret <32 x i16> %t1
+}
+
+; CHECK: test_15:
+; CHECK: v0.h = vmax(v0.h,v1.h)
+define <32 x i16> @test_15(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp sle <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v0
+  ret <32 x i16> %t1
+}
+
+; CHECK: test_16:
+; CHECK: v0.h = vmax(v0.h,v1.h)
+define <32 x i16> @test_16(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp sgt <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  ret <32 x i16> %t1
+}
+
+; CHECK: test_17:
+; CHECK: v0.h = vmax(v1.h,v0.h)
+define <32 x i16> @test_17(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp sge <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  ret <32 x i16> %t1
+}
+
+; minuh
+
+; CHECK: test_18:
+; CHECK: v0.uh = vmin(v1.uh,v0.uh)
+define <32 x i16> @test_18(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp ult <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  ret <32 x i16> %t1
+}
+
+; CHECK: test_19:
+; CHECK: v0.uh = vmin(v0.uh,v1.uh)
+define <32 x i16> @test_19(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp ule <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  ret <32 x i16> %t1
+}
+
+; CHECK: test_1a:
+; CHECK: v0.uh = vmin(v0.uh,v1.uh)
+define <32 x i16> @test_1a(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp ugt <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v0
+  ret <32 x i16> %t1
+}
+
+; CHECK: test_1b:
+; CHECK: v0.uh = vmin(v1.uh,v0.uh)
+define <32 x i16> @test_1b(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp uge <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v0
+  ret <32 x i16> %t1
+}
+
+; maxuh
+
+; CHECK: test_1c:
+; CHECK: v0.uh = vmax(v1.uh,v0.uh)
+define <32 x i16> @test_1c(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp ult <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v0
+  ret <32 x i16> %t1
+}
+
+; CHECK: test_1d:
+; CHECK: v0.uh = vmax(v0.uh,v1.uh)
+define <32 x i16> @test_1d(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp ule <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v0
+  ret <32 x i16> %t1
+}
+
+; CHECK: test_1e:
+; CHECK: v0.uh = vmax(v0.uh,v1.uh)
+define <32 x i16> @test_1e(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp ugt <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  ret <32 x i16> %t1
+}
+
+; CHECK: test_1f:
+; CHECK: v0.uh = vmax(v1.uh,v0.uh)
+define <32 x i16> @test_1f(<32 x i16> %v0, <32 x i16> %v1) #0 {
+  %t0 = icmp uge <32 x i16> %v0, %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  ret <32 x i16> %t1
+}
+
+; minw
+
+; CHECK: test_20:
+; CHECK: v0.w = vmin(v1.w,v0.w)
+define <16 x i32> @test_20(<16 x i32> %v0, <16 x i32> %v1) #0 {
+  %t0 = icmp slt <16 x i32> %v0, %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v0, <16 x i32> %v1
+  ret <16 x i32> %t1
+}
+
+; CHECK: test_21:
+; CHECK: v0.w = vmin(v0.w,v1.w)
+define <16 x i32> @test_21(<16 x i32> %v0, <16 x i32> %v1) #0 {
+  %t0 = icmp sle <16 x i32> %v0, %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v0, <16 x i32> %v1
+  ret <16 x i32> %t1
+}
+
+; CHECK: test_22:
+; CHECK: v0.w = vmin(v0.w,v1.w)
+define <16 x i32> @test_22(<16 x i32> %v0, <16 x i32> %v1) #0 {
+  %t0 = icmp sgt <16 x i32> %v0, %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v1, <16 x i32> %v0
+  ret <16 x i32> %t1
+}
+
+; CHECK: test_23:
+; CHECK: v0.w = vmin(v1.w,v0.w)
+define <16 x i32> @test_23(<16 x i32> %v0, <16 x i32> %v1) #0 {
+  %t0 = icmp sge <16 x i32> %v0, %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v1, <16 x i32> %v0
+  ret <16 x i32> %t1
+}
+
+; maxw
+
+; CHECK: test_24:
+; CHECK: v0.w = vmax(v1.w,v0.w)
+define <16 x i32> @test_24(<16 x i32> %v0, <16 x i32> %v1) #0 {
+  %t0 = icmp slt <16 x i32> %v0, %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v1, <16 x i32> %v0
+  ret <16 x i32> %t1
+}
+
+; CHECK: test_25:
+; CHECK: v0.w = vmax(v0.w,v1.w)
+define <16 x i32> @test_25(<16 x i32> %v0, <16 x i32> %v1) #0 {
+  %t0 = icmp sle <16 x i32> %v0, %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v1, <16 x i32> %v0
+  ret <16 x i32> %t1
+}
+
+; CHECK: test_26:
+; CHECK: v0.w = vmax(v0.w,v1.w)
+define <16 x i32> @test_26(<16 x i32> %v0, <16 x i32> %v1) #0 {
+  %t0 = icmp sgt <16 x i32> %v0, %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v0, <16 x i32> %v1
+  ret <16 x i32> %t1
+}
+
+; CHECK: test_27:
+; CHECK: v0.w = vmax(v1.w,v0.w)
+define <16 x i32> @test_27(<16 x i32> %v0, <16 x i32> %v1) #0 {
+  %t0 = icmp sge <16 x i32> %v0, %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v0, <16 x i32> %v1
+  ret <16 x i32> %t1
+}
+
+attributes #0 = { readnone nounwind "target-cpu"="hexagonv62" "target-features"="+hvx,+hvx-length64b" }
+

test/CodeGen/Hexagon/autohvx/vector-compare-128b.ll

@@ -4,91 +4,91 @@
 
 ; CHECK-LABEL: test_00:
 ; CHECK: q[[Q000:[0-3]]] = vcmp.eq(v0.b,v1.b)
-; CHECK: v0 = vmux(q[[Q000]],v0,v1)
+; CHECK: v0 = vmux(q[[Q000]],v1,v2)
-define <128 x i8> @test_00(<128 x i8> %v0, <128 x i8> %v1) #0 {
+define <128 x i8> @test_00(<128 x i8> %v0, <128 x i8> %v1, <128 x i8> %v2) #0 {
   %t0 = icmp eq <128 x i8> %v0, %v1
-  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v2
   ret <128 x i8> %t1
 }
 
 ; CHECK-LABEL: test_01:
 ; CHECK: q[[Q010:[0-3]]] = vcmp.eq(v0.b,v1.b)
-; CHECK: v0 = vmux(q[[Q010]],v1,v0)
+; CHECK: v0 = vmux(q[[Q010]],v2,v1)
-define <128 x i8> @test_01(<128 x i8> %v0, <128 x i8> %v1) #0 {
+define <128 x i8> @test_01(<128 x i8> %v0, <128 x i8> %v1, <128 x i8> %v2) #0 {
   %t0 = icmp ne <128 x i8> %v0, %v1
-  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v2
   ret <128 x i8> %t1
 }
 
 ; CHECK-LABEL: test_02:
 ; CHECK: q[[Q020:[0-3]]] = vcmp.gt(v1.b,v0.b)
-; CHECK: v0 = vmux(q[[Q020]],v0,v1)
+; CHECK: v0 = vmux(q[[Q020]],v1,v2)
-define <128 x i8> @test_02(<128 x i8> %v0, <128 x i8> %v1) #0 {
+define <128 x i8> @test_02(<128 x i8> %v0, <128 x i8> %v1, <128 x i8> %v2) #0 {
   %t0 = icmp slt <128 x i8> %v0, %v1
-  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v2
   ret <128 x i8> %t1
 }
 
 ; CHECK-LABEL: test_03:
 ; CHECK: q[[Q030:[0-3]]] = vcmp.gt(v0.b,v1.b)
-; CHECK: v0 = vmux(q[[Q030]],v1,v0)
+; CHECK: v0 = vmux(q[[Q030]],v2,v1)
-define <128 x i8> @test_03(<128 x i8> %v0, <128 x i8> %v1) #0 {
+define <128 x i8> @test_03(<128 x i8> %v0, <128 x i8> %v1, <128 x i8> %v2) #0 {
   %t0 = icmp sle <128 x i8> %v0, %v1
-  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v2
   ret <128 x i8> %t1
 }
 
 ; CHECK-LABEL: test_04:
 ; CHECK: q[[Q040:[0-3]]] = vcmp.gt(v0.b,v1.b)
-; CHECK: v0 = vmux(q[[Q040]],v0,v1)
+; CHECK: v0 = vmux(q[[Q040]],v1,v2)
-define <128 x i8> @test_04(<128 x i8> %v0, <128 x i8> %v1) #0 {
+define <128 x i8> @test_04(<128 x i8> %v0, <128 x i8> %v1, <128 x i8> %v2) #0 {
   %t0 = icmp sgt <128 x i8> %v0, %v1
-  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v2
   ret <128 x i8> %t1
 }
 
 ; CHECK-LABEL: test_05:
 ; CHECK: q[[Q050:[0-3]]] = vcmp.gt(v1.b,v0.b)
-; CHECK: v0 = vmux(q[[Q050]],v1,v0)
+; CHECK: v0 = vmux(q[[Q050]],v2,v1)
-define <128 x i8> @test_05(<128 x i8> %v0, <128 x i8> %v1) #0 {
+define <128 x i8> @test_05(<128 x i8> %v0, <128 x i8> %v1, <128 x i8> %v2) #0 {
   %t0 = icmp sge <128 x i8> %v0, %v1
-  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v2
   ret <128 x i8> %t1
 }
 
 ; CHECK-LABEL: test_06:
 ; CHECK: q[[Q060:[0-3]]] = vcmp.gt(v1.ub,v0.ub)
-; CHECK: v0 = vmux(q[[Q060]],v0,v1)
+; CHECK: v0 = vmux(q[[Q060]],v1,v2)
-define <128 x i8> @test_06(<128 x i8> %v0, <128 x i8> %v1) #0 {
+define <128 x i8> @test_06(<128 x i8> %v0, <128 x i8> %v1, <128 x i8> %v2) #0 {
   %t0 = icmp ult <128 x i8> %v0, %v1
-  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v2
   ret <128 x i8> %t1
 }
 
 ; CHECK-LABEL: test_07:
 ; CHECK: q[[Q070:[0-3]]] = vcmp.gt(v0.ub,v1.ub)
-; CHECK: v0 = vmux(q[[Q070]],v1,v0)
+; CHECK: v0 = vmux(q[[Q070]],v2,v1)
-define <128 x i8> @test_07(<128 x i8> %v0, <128 x i8> %v1) #0 {
+define <128 x i8> @test_07(<128 x i8> %v0, <128 x i8> %v1, <128 x i8> %v2) #0 {
   %t0 = icmp ule <128 x i8> %v0, %v1
-  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v2
   ret <128 x i8> %t1
 }
 
 ; CHECK-LABEL: test_08:
 ; CHECK: q[[Q080:[0-3]]] = vcmp.gt(v0.ub,v1.ub)
-; CHECK: v0 = vmux(q[[Q080]],v0,v1)
+; CHECK: v0 = vmux(q[[Q080]],v1,v2)
-define <128 x i8> @test_08(<128 x i8> %v0, <128 x i8> %v1) #0 {
+define <128 x i8> @test_08(<128 x i8> %v0, <128 x i8> %v1, <128 x i8> %v2) #0 {
   %t0 = icmp ugt <128 x i8> %v0, %v1
-  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v2
   ret <128 x i8> %t1
 }
 
 ; CHECK-LABEL: test_09:
 ; CHECK: q[[Q090:[0-3]]] = vcmp.gt(v1.ub,v0.ub)
-; CHECK: v0 = vmux(q[[Q090]],v1,v0)
+; CHECK: v0 = vmux(q[[Q090]],v2,v1)
-define <128 x i8> @test_09(<128 x i8> %v0, <128 x i8> %v1) #0 {
+define <128 x i8> @test_09(<128 x i8> %v0, <128 x i8> %v1, <128 x i8> %v2) #0 {
   %t0 = icmp uge <128 x i8> %v0, %v1
-  %t1 = select <128 x i1> %t0, <128 x i8> %v0, <128 x i8> %v1
+  %t1 = select <128 x i1> %t0, <128 x i8> %v1, <128 x i8> %v2
   ret <128 x i8> %t1
 }
 
@@ -195,91 +195,91 @@ define <128 x i8> @test_0i(<128 x i8> %v0, <128 x i8> %v1, <128 x i8> %v2) #0 {
 
 ; CHECK-LABEL: test_10:
 ; CHECK: q[[Q100:[0-3]]] = vcmp.eq(v0.h,v1.h)
-; CHECK: v0 = vmux(q[[Q100]],v0,v1)
+; CHECK: v0 = vmux(q[[Q100]],v1,v2)
-define <64 x i16> @test_10(<64 x i16> %v0, <64 x i16> %v1) #0 {
+define <64 x i16> @test_10(<64 x i16> %v0, <64 x i16> %v1, <64 x i16> %v2) #0 {
   %t0 = icmp eq <64 x i16> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v2
   ret <64 x i16> %t1
 }
 
 ; CHECK-LABEL: test_11:
 ; CHECK: q[[Q110:[0-3]]] = vcmp.eq(v0.h,v1.h)
-; CHECK: v0 = vmux(q[[Q110]],v1,v0)
+; CHECK: v0 = vmux(q[[Q110]],v2,v1)
-define <64 x i16> @test_11(<64 x i16> %v0, <64 x i16> %v1) #0 {
+define <64 x i16> @test_11(<64 x i16> %v0, <64 x i16> %v1, <64 x i16> %v2) #0 {
   %t0 = icmp ne <64 x i16> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v2
   ret <64 x i16> %t1
 }
 
 ; CHECK-LABEL: test_12:
 ; CHECK: q[[Q120:[0-3]]] = vcmp.gt(v1.h,v0.h)
-; CHECK: v0 = vmux(q[[Q120]],v0,v1)
+; CHECK: v0 = vmux(q[[Q120]],v1,v2)
-define <64 x i16> @test_12(<64 x i16> %v0, <64 x i16> %v1) #0 {
+define <64 x i16> @test_12(<64 x i16> %v0, <64 x i16> %v1, <64 x i16> %v2) #0 {
   %t0 = icmp slt <64 x i16> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v2
   ret <64 x i16> %t1
 }
 
 ; CHECK-LABEL: test_13:
 ; CHECK: q[[Q130:[0-3]]] = vcmp.gt(v0.h,v1.h)
-; CHECK: v0 = vmux(q[[Q130]],v1,v0)
+; CHECK: v0 = vmux(q[[Q130]],v2,v1)
-define <64 x i16> @test_13(<64 x i16> %v0, <64 x i16> %v1) #0 {
+define <64 x i16> @test_13(<64 x i16> %v0, <64 x i16> %v1, <64 x i16> %v2) #0 {
   %t0 = icmp sle <64 x i16> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v2
   ret <64 x i16> %t1
 }
 
 ; CHECK-LABEL: test_14:
 ; CHECK: q[[Q140:[0-3]]] = vcmp.gt(v0.h,v1.h)
-; CHECK: v0 = vmux(q[[Q140]],v0,v1)
+; CHECK: v0 = vmux(q[[Q140]],v1,v2)
-define <64 x i16> @test_14(<64 x i16> %v0, <64 x i16> %v1) #0 {
+define <64 x i16> @test_14(<64 x i16> %v0, <64 x i16> %v1, <64 x i16> %v2) #0 {
   %t0 = icmp sgt <64 x i16> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v2
   ret <64 x i16> %t1
 }
 
 ; CHECK-LABEL: test_15:
 ; CHECK: q[[Q150:[0-3]]] = vcmp.gt(v1.h,v0.h)
-; CHECK: v0 = vmux(q[[Q150]],v1,v0)
+; CHECK: v0 = vmux(q[[Q150]],v2,v1)
-define <64 x i16> @test_15(<64 x i16> %v0, <64 x i16> %v1) #0 {
+define <64 x i16> @test_15(<64 x i16> %v0, <64 x i16> %v1, <64 x i16> %v2) #0 {
   %t0 = icmp sge <64 x i16> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v2
   ret <64 x i16> %t1
 }
 
 ; CHECK-LABEL: test_16:
 ; CHECK: q[[Q160:[0-3]]] = vcmp.gt(v1.uh,v0.uh)
-; CHECK: v0 = vmux(q[[Q160]],v0,v1)
+; CHECK: v0 = vmux(q[[Q160]],v1,v2)
-define <64 x i16> @test_16(<64 x i16> %v0, <64 x i16> %v1) #0 {
+define <64 x i16> @test_16(<64 x i16> %v0, <64 x i16> %v1, <64 x i16> %v2) #0 {
   %t0 = icmp ult <64 x i16> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v2
   ret <64 x i16> %t1
 }
 
 ; CHECK-LABEL: test_17:
 ; CHECK: q[[Q170:[0-3]]] = vcmp.gt(v0.uh,v1.uh)
-; CHECK: v0 = vmux(q[[Q170]],v1,v0)
+; CHECK: v0 = vmux(q[[Q170]],v2,v1)
-define <64 x i16> @test_17(<64 x i16> %v0, <64 x i16> %v1) #0 {
+define <64 x i16> @test_17(<64 x i16> %v0, <64 x i16> %v1, <64 x i16> %v2) #0 {
   %t0 = icmp ule <64 x i16> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v2
   ret <64 x i16> %t1
 }
 
 ; CHECK-LABEL: test_18:
 ; CHECK: q[[Q180:[0-3]]] = vcmp.gt(v0.uh,v1.uh)
-; CHECK: v0 = vmux(q[[Q180]],v0,v1)
+; CHECK: v0 = vmux(q[[Q180]],v1,v2)
-define <64 x i16> @test_18(<64 x i16> %v0, <64 x i16> %v1) #0 {
+define <64 x i16> @test_18(<64 x i16> %v0, <64 x i16> %v1, <64 x i16> %v2) #0 {
   %t0 = icmp ugt <64 x i16> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v2
   ret <64 x i16> %t1
 }
 
 ; CHECK-LABEL: test_19:
 ; CHECK: q[[Q190:[0-3]]] = vcmp.gt(v1.uh,v0.uh)
-; CHECK: v0 = vmux(q[[Q190]],v1,v0)
+; CHECK: v0 = vmux(q[[Q190]],v2,v1)
-define <64 x i16> @test_19(<64 x i16> %v0, <64 x i16> %v1) #0 {
+define <64 x i16> @test_19(<64 x i16> %v0, <64 x i16> %v1, <64 x i16> %v2) #0 {
   %t0 = icmp uge <64 x i16> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i16> %v0, <64 x i16> %v1
+  %t1 = select <64 x i1> %t0, <64 x i16> %v1, <64 x i16> %v2
   ret <64 x i16> %t1
 }
 
@@ -386,91 +386,91 @@ define <64 x i16> @test_1i(<64 x i16> %v0, <64 x i16> %v1, <64 x i16> %v2) #0 {
 
 ; CHECK-LABEL: test_20:
 ; CHECK: q[[Q200:[0-3]]] = vcmp.eq(v0.w,v1.w)
-; CHECK: v0 = vmux(q[[Q200]],v0,v1)
+; CHECK: v0 = vmux(q[[Q200]],v1,v2)
-define <32 x i32> @test_20(<32 x i32> %v0, <32 x i32> %v1) #0 {
+define <32 x i32> @test_20(<32 x i32> %v0, <32 x i32> %v1, <32 x i32> %v2) #0 {
   %t0 = icmp eq <32 x i32> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i32> %v0, <32 x i32> %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v1, <32 x i32> %v2
   ret <32 x i32> %t1
 }
 
 ; CHECK-LABEL: test_21:
 ; CHECK: q[[Q210:[0-3]]] = vcmp.eq(v0.w,v1.w)
-; CHECK: v0 = vmux(q[[Q210]],v1,v0)
+; CHECK: v0 = vmux(q[[Q210]],v2,v1)
-define <32 x i32> @test_21(<32 x i32> %v0, <32 x i32> %v1) #0 {
+define <32 x i32> @test_21(<32 x i32> %v0, <32 x i32> %v1, <32 x i32> %v2) #0 {
   %t0 = icmp ne <32 x i32> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i32> %v0, <32 x i32> %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v1, <32 x i32> %v2
   ret <32 x i32> %t1
 }
 
 ; CHECK-LABEL: test_22:
 ; CHECK: q[[Q220:[0-3]]] = vcmp.gt(v1.w,v0.w)
-; CHECK: v0 = vmux(q[[Q220]],v0,v1)
+; CHECK: v0 = vmux(q[[Q220]],v1,v2)
-define <32 x i32> @test_22(<32 x i32> %v0, <32 x i32> %v1) #0 {
+define <32 x i32> @test_22(<32 x i32> %v0, <32 x i32> %v1, <32 x i32> %v2) #0 {
   %t0 = icmp slt <32 x i32> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i32> %v0, <32 x i32> %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v1, <32 x i32> %v2
   ret <32 x i32> %t1
 }
 
 ; CHECK-LABEL: test_23:
 ; CHECK: q[[Q230:[0-3]]] = vcmp.gt(v0.w,v1.w)
-; CHECK: v0 = vmux(q[[Q230]],v1,v0)
+; CHECK: v0 = vmux(q[[Q230]],v2,v1)
-define <32 x i32> @test_23(<32 x i32> %v0, <32 x i32> %v1) #0 {
+define <32 x i32> @test_23(<32 x i32> %v0, <32 x i32> %v1, <32 x i32> %v2) #0 {
   %t0 = icmp sle <32 x i32> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i32> %v0, <32 x i32> %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v1, <32 x i32> %v2
   ret <32 x i32> %t1
 }
 
 ; CHECK-LABEL: test_24:
 ; CHECK: q[[Q240:[0-3]]] = vcmp.gt(v0.w,v1.w)
-; CHECK: v0 = vmux(q[[Q240]],v0,v1)
+; CHECK: v0 = vmux(q[[Q240]],v1,v2)
-define <32 x i32> @test_24(<32 x i32> %v0, <32 x i32> %v1) #0 {
+define <32 x i32> @test_24(<32 x i32> %v0, <32 x i32> %v1, <32 x i32> %v2) #0 {
   %t0 = icmp sgt <32 x i32> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i32> %v0, <32 x i32> %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v1, <32 x i32> %v2
   ret <32 x i32> %t1
 }
 
 ; CHECK-LABEL: test_25:
 ; CHECK: q[[Q250:[0-3]]] = vcmp.gt(v1.w,v0.w)
-; CHECK: v0 = vmux(q[[Q250]],v1,v0)
+; CHECK: v0 = vmux(q[[Q250]],v2,v1)
-define <32 x i32> @test_25(<32 x i32> %v0, <32 x i32> %v1) #0 {
+define <32 x i32> @test_25(<32 x i32> %v0, <32 x i32> %v1, <32 x i32> %v2) #0 {
   %t0 = icmp sge <32 x i32> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i32> %v0, <32 x i32> %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v1, <32 x i32> %v2
   ret <32 x i32> %t1
 }
 
 ; CHECK-LABEL: test_26:
 ; CHECK: q[[Q260:[0-3]]] = vcmp.gt(v1.uw,v0.uw)
-; CHECK: v0 = vmux(q[[Q260]],v0,v1)
+; CHECK: v0 = vmux(q[[Q260]],v1,v2)
-define <32 x i32> @test_26(<32 x i32> %v0, <32 x i32> %v1) #0 {
+define <32 x i32> @test_26(<32 x i32> %v0, <32 x i32> %v1, <32 x i32> %v2) #0 {
   %t0 = icmp ult <32 x i32> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i32> %v0, <32 x i32> %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v1, <32 x i32> %v2
   ret <32 x i32> %t1
 }
 
 ; CHECK-LABEL: test_27:
 ; CHECK: q[[Q270:[0-3]]] = vcmp.gt(v0.uw,v1.uw)
-; CHECK: v0 = vmux(q[[Q270]],v1,v0)
+; CHECK: v0 = vmux(q[[Q270]],v2,v1)
-define <32 x i32> @test_27(<32 x i32> %v0, <32 x i32> %v1) #0 {
+define <32 x i32> @test_27(<32 x i32> %v0, <32 x i32> %v1, <32 x i32> %v2) #0 {
   %t0 = icmp ule <32 x i32> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i32> %v0, <32 x i32> %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v1, <32 x i32> %v2
   ret <32 x i32> %t1
 }
 
 ; CHECK-LABEL: test_28:
 ; CHECK: q[[Q280:[0-3]]] = vcmp.gt(v0.uw,v1.uw)
-; CHECK: v0 = vmux(q[[Q280]],v0,v1)
+; CHECK: v0 = vmux(q[[Q280]],v1,v2)
-define <32 x i32> @test_28(<32 x i32> %v0, <32 x i32> %v1) #0 {
+define <32 x i32> @test_28(<32 x i32> %v0, <32 x i32> %v1, <32 x i32> %v2) #0 {
   %t0 = icmp ugt <32 x i32> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i32> %v0, <32 x i32> %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v1, <32 x i32> %v2
   ret <32 x i32> %t1
 }
 
 ; CHECK-LABEL: test_29:
 ; CHECK: q[[Q290:[0-3]]] = vcmp.gt(v1.uw,v0.uw)
-; CHECK: v0 = vmux(q[[Q290]],v1,v0)
+; CHECK: v0 = vmux(q[[Q290]],v2,v1)
-define <32 x i32> @test_29(<32 x i32> %v0, <32 x i32> %v1) #0 {
+define <32 x i32> @test_29(<32 x i32> %v0, <32 x i32> %v1, <32 x i32> %v2) #0 {
   %t0 = icmp uge <32 x i32> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i32> %v0, <32 x i32> %v1
+  %t1 = select <32 x i1> %t0, <32 x i32> %v1, <32 x i32> %v2
   ret <32 x i32> %t1
 }
 

test/CodeGen/Hexagon/autohvx/vector-compare-64b.ll

@@ -4,91 +4,91 @@
 
 ; CHECK-LABEL: test_00:
 ; CHECK: q[[Q000:[0-3]]] = vcmp.eq(v0.b,v1.b)
-; CHECK: v0 = vmux(q[[Q000]],v0,v1)
+; CHECK: v0 = vmux(q[[Q000]],v1,v2)
-define <64 x i8> @test_00(<64 x i8> %v0, <64 x i8> %v1) #0 {
+define <64 x i8> @test_00(<64 x i8> %v0, <64 x i8> %v1, <64 x i8> %v2) #0 {
   %t0 = icmp eq <64 x i8> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v2
   ret <64 x i8> %t1
 }
 
 ; CHECK-LABEL: test_01:
 ; CHECK: q[[Q010:[0-3]]] = vcmp.eq(v0.b,v1.b)
-; CHECK: v0 = vmux(q[[Q010]],v1,v0)
+; CHECK: v0 = vmux(q[[Q010]],v2,v1)
-define <64 x i8> @test_01(<64 x i8> %v0, <64 x i8> %v1) #0 {
+define <64 x i8> @test_01(<64 x i8> %v0, <64 x i8> %v1, <64 x i8> %v2) #0 {
   %t0 = icmp ne <64 x i8> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v2
   ret <64 x i8> %t1
 }
 
 ; CHECK-LABEL: test_02:
 ; CHECK: q[[Q020:[0-3]]] = vcmp.gt(v1.b,v0.b)
-; CHECK: v0 = vmux(q[[Q020]],v0,v1)
+; CHECK: v0 = vmux(q[[Q020]],v1,v2)
-define <64 x i8> @test_02(<64 x i8> %v0, <64 x i8> %v1) #0 {
+define <64 x i8> @test_02(<64 x i8> %v0, <64 x i8> %v1, <64 x i8> %v2) #0 {
   %t0 = icmp slt <64 x i8> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v2
   ret <64 x i8> %t1
 }
 
 ; CHECK-LABEL: test_03:
 ; CHECK: q[[Q030:[0-3]]] = vcmp.gt(v0.b,v1.b)
-; CHECK: v0 = vmux(q[[Q030]],v1,v0)
+; CHECK: v0 = vmux(q[[Q030]],v2,v1)
-define <64 x i8> @test_03(<64 x i8> %v0, <64 x i8> %v1) #0 {
+define <64 x i8> @test_03(<64 x i8> %v0, <64 x i8> %v1, <64 x i8> %v2) #0 {
   %t0 = icmp sle <64 x i8> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v2
   ret <64 x i8> %t1
 }
 
 ; CHECK-LABEL: test_04:
 ; CHECK: q[[Q040:[0-3]]] = vcmp.gt(v0.b,v1.b)
-; CHECK: v0 = vmux(q[[Q040]],v0,v1)
+; CHECK: v0 = vmux(q[[Q040]],v1,v2)
-define <64 x i8> @test_04(<64 x i8> %v0, <64 x i8> %v1) #0 {
+define <64 x i8> @test_04(<64 x i8> %v0, <64 x i8> %v1, <64 x i8> %v2) #0 {
   %t0 = icmp sgt <64 x i8> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v2
   ret <64 x i8> %t1
 }
 
 ; CHECK-LABEL: test_05:
 ; CHECK: q[[Q050:[0-3]]] = vcmp.gt(v1.b,v0.b)
-; CHECK: v0 = vmux(q[[Q050]],v1,v0)
+; CHECK: v0 = vmux(q[[Q050]],v2,v1)
-define <64 x i8> @test_05(<64 x i8> %v0, <64 x i8> %v1) #0 {
+define <64 x i8> @test_05(<64 x i8> %v0, <64 x i8> %v1, <64 x i8> %v2) #0 {
   %t0 = icmp sge <64 x i8> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v2
   ret <64 x i8> %t1
 }
 
 ; CHECK-LABEL: test_06:
 ; CHECK: q[[Q060:[0-3]]] = vcmp.gt(v1.ub,v0.ub)
-; CHECK: v0 = vmux(q[[Q060]],v0,v1)
+; CHECK: v0 = vmux(q[[Q060]],v1,v2)
-define <64 x i8> @test_06(<64 x i8> %v0, <64 x i8> %v1) #0 {
+define <64 x i8> @test_06(<64 x i8> %v0, <64 x i8> %v1, <64 x i8> %v2) #0 {
   %t0 = icmp ult <64 x i8> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v2
   ret <64 x i8> %t1
 }
 
 ; CHECK-LABEL: test_07:
 ; CHECK: q[[Q070:[0-3]]] = vcmp.gt(v0.ub,v1.ub)
-; CHECK: v0 = vmux(q[[Q070]],v1,v0)
+; CHECK: v0 = vmux(q[[Q070]],v2,v1)
-define <64 x i8> @test_07(<64 x i8> %v0, <64 x i8> %v1) #0 {
+define <64 x i8> @test_07(<64 x i8> %v0, <64 x i8> %v1, <64 x i8> %v2) #0 {
   %t0 = icmp ule <64 x i8> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v2
   ret <64 x i8> %t1
 }
 
 ; CHECK-LABEL: test_08:
 ; CHECK: q[[Q080:[0-3]]] = vcmp.gt(v0.ub,v1.ub)
-; CHECK: v0 = vmux(q[[Q080]],v0,v1)
+; CHECK: v0 = vmux(q[[Q080]],v1,v2)
-define <64 x i8> @test_08(<64 x i8> %v0, <64 x i8> %v1) #0 {
+define <64 x i8> @test_08(<64 x i8> %v0, <64 x i8> %v1, <64 x i8> %v2) #0 {
   %t0 = icmp ugt <64 x i8> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v2
   ret <64 x i8> %t1
 }
 
 ; CHECK-LABEL: test_09:
 ; CHECK: q[[Q090:[0-3]]] = vcmp.gt(v1.ub,v0.ub)
-; CHECK: v0 = vmux(q[[Q090]],v1,v0)
+; CHECK: v0 = vmux(q[[Q090]],v2,v1)
-define <64 x i8> @test_09(<64 x i8> %v0, <64 x i8> %v1) #0 {
+define <64 x i8> @test_09(<64 x i8> %v0, <64 x i8> %v1, <64 x i8> %v2) #0 {
   %t0 = icmp uge <64 x i8> %v0, %v1
-  %t1 = select <64 x i1> %t0, <64 x i8> %v0, <64 x i8> %v1
+  %t1 = select <64 x i1> %t0, <64 x i8> %v1, <64 x i8> %v2
   ret <64 x i8> %t1
 }
 
@@ -196,91 +196,91 @@ define <64 x i8> @test_0i(<64 x i8> %v0, <64 x i8> %v1, <64 x i8> %v2) #0 {
 
 ; CHECK-LABEL: test_10:
 ; CHECK: q[[Q100:[0-3]]] = vcmp.eq(v0.h,v1.h)
-; CHECK: v0 = vmux(q[[Q100]],v0,v1)
+; CHECK: v0 = vmux(q[[Q100]],v1,v2)
-define <32 x i16> @test_10(<32 x i16> %v0, <32 x i16> %v1) #0 {
+define <32 x i16> @test_10(<32 x i16> %v0, <32 x i16> %v1, <32 x i16> %v2) #0 {
   %t0 = icmp eq <32 x i16> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v2
   ret <32 x i16> %t1
 }
 
 ; CHECK-LABEL: test_11:
 ; CHECK: q[[Q110:[0-3]]] = vcmp.eq(v0.h,v1.h)
-; CHECK: v0 = vmux(q[[Q110]],v1,v0)
+; CHECK: v0 = vmux(q[[Q110]],v2,v1)
-define <32 x i16> @test_11(<32 x i16> %v0, <32 x i16> %v1) #0 {
+define <32 x i16> @test_11(<32 x i16> %v0, <32 x i16> %v1, <32 x i16> %v2) #0 {
   %t0 = icmp ne <32 x i16> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v2
   ret <32 x i16> %t1
 }
 
 ; CHECK-LABEL: test_12:
 ; CHECK: q[[Q120:[0-3]]] = vcmp.gt(v1.h,v0.h)
-; CHECK: v0 = vmux(q[[Q120]],v0,v1)
+; CHECK: v0 = vmux(q[[Q120]],v1,v2)
-define <32 x i16> @test_12(<32 x i16> %v0, <32 x i16> %v1) #0 {
+define <32 x i16> @test_12(<32 x i16> %v0, <32 x i16> %v1, <32 x i16> %v2) #0 {
   %t0 = icmp slt <32 x i16> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v2
   ret <32 x i16> %t1
 }
 
 ; CHECK-LABEL: test_13:
 ; CHECK: q[[Q130:[0-3]]] = vcmp.gt(v0.h,v1.h)
-; CHECK: v0 = vmux(q[[Q130]],v1,v0)
+; CHECK: v0 = vmux(q[[Q130]],v2,v1)
-define <32 x i16> @test_13(<32 x i16> %v0, <32 x i16> %v1) #0 {
+define <32 x i16> @test_13(<32 x i16> %v0, <32 x i16> %v1, <32 x i16> %v2) #0 {
   %t0 = icmp sle <32 x i16> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v2
   ret <32 x i16> %t1
 }
 
 ; CHECK-LABEL: test_14:
 ; CHECK: q[[Q140:[0-3]]] = vcmp.gt(v0.h,v1.h)
-; CHECK: v0 = vmux(q[[Q140]],v0,v1)
+; CHECK: v0 = vmux(q[[Q140]],v1,v2)
-define <32 x i16> @test_14(<32 x i16> %v0, <32 x i16> %v1) #0 {
+define <32 x i16> @test_14(<32 x i16> %v0, <32 x i16> %v1, <32 x i16> %v2) #0 {
   %t0 = icmp sgt <32 x i16> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v2
   ret <32 x i16> %t1
 }
 
 ; CHECK-LABEL: test_15:
 ; CHECK: q[[Q150:[0-3]]] = vcmp.gt(v1.h,v0.h)
-; CHECK: v0 = vmux(q[[Q150]],v1,v0)
+; CHECK: v0 = vmux(q[[Q150]],v2,v1)
-define <32 x i16> @test_15(<32 x i16> %v0, <32 x i16> %v1) #0 {
+define <32 x i16> @test_15(<32 x i16> %v0, <32 x i16> %v1, <32 x i16> %v2) #0 {
   %t0 = icmp sge <32 x i16> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v2
   ret <32 x i16> %t1
 }
 
 ; CHECK-LABEL: test_16:
 ; CHECK: q[[Q160:[0-3]]] = vcmp.gt(v1.uh,v0.uh)
-; CHECK: v0 = vmux(q[[Q160]],v0,v1)
+; CHECK: v0 = vmux(q[[Q160]],v1,v2)
-define <32 x i16> @test_16(<32 x i16> %v0, <32 x i16> %v1) #0 {
+define <32 x i16> @test_16(<32 x i16> %v0, <32 x i16> %v1, <32 x i16> %v2) #0 {
   %t0 = icmp ult <32 x i16> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v2
   ret <32 x i16> %t1
 }
 
 ; CHECK-LABEL: test_17:
 ; CHECK: q[[Q170:[0-3]]] = vcmp.gt(v0.uh,v1.uh)
-; CHECK: v0 = vmux(q[[Q170]],v1,v0)
+; CHECK: v0 = vmux(q[[Q170]],v2,v1)
-define <32 x i16> @test_17(<32 x i16> %v0, <32 x i16> %v1) #0 {
+define <32 x i16> @test_17(<32 x i16> %v0, <32 x i16> %v1, <32 x i16> %v2) #0 {
   %t0 = icmp ule <32 x i16> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v2
   ret <32 x i16> %t1
 }
 
 ; CHECK-LABEL: test_18:
 ; CHECK: q[[Q180:[0-3]]] = vcmp.gt(v0.uh,v1.uh)
-; CHECK: v0 = vmux(q[[Q180]],v0,v1)
+; CHECK: v0 = vmux(q[[Q180]],v1,v2)
-define <32 x i16> @test_18(<32 x i16> %v0, <32 x i16> %v1) #0 {
+define <32 x i16> @test_18(<32 x i16> %v0, <32 x i16> %v1, <32 x i16> %v2) #0 {
   %t0 = icmp ugt <32 x i16> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v2
   ret <32 x i16> %t1
 }
 
 ; CHECK-LABEL: test_19:
 ; CHECK: q[[Q190:[0-3]]] = vcmp.gt(v1.uh,v0.uh)
-; CHECK: v0 = vmux(q[[Q190]],v1,v0)
+; CHECK: v0 = vmux(q[[Q190]],v2,v1)
-define <32 x i16> @test_19(<32 x i16> %v0, <32 x i16> %v1) #0 {
+define <32 x i16> @test_19(<32 x i16> %v0, <32 x i16> %v1, <32 x i16> %v2) #0 {
   %t0 = icmp uge <32 x i16> %v0, %v1
-  %t1 = select <32 x i1> %t0, <32 x i16> %v0, <32 x i16> %v1
+  %t1 = select <32 x i1> %t0, <32 x i16> %v1, <32 x i16> %v2
   ret <32 x i16> %t1
 }
 
@@ -387,91 +387,91 @@ define <32 x i16> @test_1i(<32 x i16> %v0, <32 x i16> %v1, <32 x i16> %v2) #0 {
 
 ; CHECK-LABEL: test_20:
 ; CHECK: q[[Q200:[0-3]]] = vcmp.eq(v0.w,v1.w)
-; CHECK: v0 = vmux(q[[Q200]],v0,v1)
+; CHECK: v0 = vmux(q[[Q200]],v1,v2)
-define <16 x i32> @test_20(<16 x i32> %v0, <16 x i32> %v1) #0 {
+define <16 x i32> @test_20(<16 x i32> %v0, <16 x i32> %v1, <16 x i32> %v2) #0 {
   %t0 = icmp eq <16 x i32> %v0, %v1
-  %t1 = select <16 x i1> %t0, <16 x i32> %v0, <16 x i32> %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v1, <16 x i32> %v2
   ret <16 x i32> %t1
 }
 
 ; CHECK-LABEL: test_21:
 ; CHECK: q[[Q210:[0-3]]] = vcmp.eq(v0.w,v1.w)
-; CHECK: v0 = vmux(q[[Q210]],v1,v0)
+; CHECK: v0 = vmux(q[[Q210]],v2,v1)
-define <16 x i32> @test_21(<16 x i32> %v0, <16 x i32> %v1) #0 {
+define <16 x i32> @test_21(<16 x i32> %v0, <16 x i32> %v1, <16 x i32> %v2) #0 {
   %t0 = icmp ne <16 x i32> %v0, %v1
-  %t1 = select <16 x i1> %t0, <16 x i32> %v0, <16 x i32> %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v1, <16 x i32> %v2
   ret <16 x i32> %t1
 }
 
 ; CHECK-LABEL: test_22:
 ; CHECK: q[[Q220:[0-3]]] = vcmp.gt(v1.w,v0.w)
-; CHECK: v0 = vmux(q[[Q220]],v0,v1)
+; CHECK: v0 = vmux(q[[Q220]],v1,v2)
-define <16 x i32> @test_22(<16 x i32> %v0, <16 x i32> %v1) #0 {
+define <16 x i32> @test_22(<16 x i32> %v0, <16 x i32> %v1, <16 x i32> %v2) #0 {
   %t0 = icmp slt <16 x i32> %v0, %v1
-  %t1 = select <16 x i1> %t0, <16 x i32> %v0, <16 x i32> %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v1, <16 x i32> %v2
   ret <16 x i32> %t1
 }
 
 ; CHECK-LABEL: test_23:
 ; CHECK: q[[Q230:[0-3]]] = vcmp.gt(v0.w,v1.w)
-; CHECK: v0 = vmux(q[[Q230]],v1,v0)
+; CHECK: v0 = vmux(q[[Q230]],v2,v1)
-define <16 x i32> @test_23(<16 x i32> %v0, <16 x i32> %v1) #0 {
+define <16 x i32> @test_23(<16 x i32> %v0, <16 x i32> %v1, <16 x i32> %v2) #0 {
   %t0 = icmp sle <16 x i32> %v0, %v1
-  %t1 = select <16 x i1> %t0, <16 x i32> %v0, <16 x i32> %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v1, <16 x i32> %v2
   ret <16 x i32> %t1
 }
 
 ; CHECK-LABEL: test_24:
 ; CHECK: q[[Q240:[0-3]]] = vcmp.gt(v0.w,v1.w)
-; CHECK: v0 = vmux(q[[Q240]],v0,v1)
+; CHECK: v0 = vmux(q[[Q240]],v1,v2)
-define <16 x i32> @test_24(<16 x i32> %v0, <16 x i32> %v1) #0 {
+define <16 x i32> @test_24(<16 x i32> %v0, <16 x i32> %v1, <16 x i32> %v2) #0 {
   %t0 = icmp sgt <16 x i32> %v0, %v1
-  %t1 = select <16 x i1> %t0, <16 x i32> %v0, <16 x i32> %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v1, <16 x i32> %v2
   ret <16 x i32> %t1
 }
 
 ; CHECK-LABEL: test_25:
 ; CHECK: q[[Q250:[0-3]]] = vcmp.gt(v1.w,v0.w)
-; CHECK: v0 = vmux(q[[Q250]],v1,v0)
+; CHECK: v0 = vmux(q[[Q250]],v2,v1)
-define <16 x i32> @test_25(<16 x i32> %v0, <16 x i32> %v1) #0 {
+define <16 x i32> @test_25(<16 x i32> %v0, <16 x i32> %v1, <16 x i32> %v2) #0 {
   %t0 = icmp sge <16 x i32> %v0, %v1
-  %t1 = select <16 x i1> %t0, <16 x i32> %v0, <16 x i32> %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v1, <16 x i32> %v2
   ret <16 x i32> %t1
 }
 
 ; CHECK-LABEL: test_26:
 ; CHECK: q[[Q260:[0-3]]] = vcmp.gt(v1.uw,v0.uw)
-; CHECK: v0 = vmux(q[[Q260]],v0,v1)
+; CHECK: v0 = vmux(q[[Q260]],v1,v2)
-define <16 x i32> @test_26(<16 x i32> %v0, <16 x i32> %v1) #0 {
+define <16 x i32> @test_26(<16 x i32> %v0, <16 x i32> %v1, <16 x i32> %v2) #0 {
   %t0 = icmp ult <16 x i32> %v0, %v1
-  %t1 = select <16 x i1> %t0, <16 x i32> %v0, <16 x i32> %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v1, <16 x i32> %v2
   ret <16 x i32> %t1
 }
 
 ; CHECK-LABEL: test_27:
 ; CHECK: q[[Q270:[0-3]]] = vcmp.gt(v0.uw,v1.uw)
-; CHECK: v0 = vmux(q[[Q270]],v1,v0)
+; CHECK: v0 = vmux(q[[Q270]],v2,v1)
-define <16 x i32> @test_27(<16 x i32> %v0, <16 x i32> %v1) #0 {
+define <16 x i32> @test_27(<16 x i32> %v0, <16 x i32> %v1, <16 x i32> %v2) #0 {
   %t0 = icmp ule <16 x i32> %v0, %v1
-  %t1 = select <16 x i1> %t0, <16 x i32> %v0, <16 x i32> %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v1, <16 x i32> %v2
   ret <16 x i32> %t1
 }
 
 ; CHECK-LABEL: test_28:
 ; CHECK: q[[Q280:[0-3]]] = vcmp.gt(v0.uw,v1.uw)
-; CHECK: v0 = vmux(q[[Q280]],v0,v1)
+; CHECK: v0 = vmux(q[[Q280]],v1,v2)
-define <16 x i32> @test_28(<16 x i32> %v0, <16 x i32> %v1) #0 {
+define <16 x i32> @test_28(<16 x i32> %v0, <16 x i32> %v1, <16 x i32> %v2) #0 {
   %t0 = icmp ugt <16 x i32> %v0, %v1
-  %t1 = select <16 x i1> %t0, <16 x i32> %v0, <16 x i32> %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v1, <16 x i32> %v2
   ret <16 x i32> %t1
 }
 
 ; CHECK-LABEL: test_29:
 ; CHECK: q[[Q290:[0-3]]] = vcmp.gt(v1.uw,v0.uw)
-; CHECK: v0 = vmux(q[[Q290]],v1,v0)
+; CHECK: v0 = vmux(q[[Q290]],v2,v1)
-define <16 x i32> @test_29(<16 x i32> %v0, <16 x i32> %v1) #0 {
+define <16 x i32> @test_29(<16 x i32> %v0, <16 x i32> %v1, <16 x i32> %v2) #0 {
   %t0 = icmp uge <16 x i32> %v0, %v1
-  %t1 = select <16 x i1> %t0, <16 x i32> %v0, <16 x i32> %v1
+  %t1 = select <16 x i1> %t0, <16 x i32> %v1, <16 x i32> %v2
   ret <16 x i32> %t1
 }