[PPC64] Add vector pack/unpack support from ISA 2.07

This patch adds support for the following new instructions in the
Power ISA 2.07:

  vpksdss
  vpksdus
  vpkudus
  vpkudum
  vupkhsw
  vupklsw

These instructions are available through the vec_packs, vec_packsu,
vec_unpackh, and vec_unpackl built-in interfaces.  These are
lane-sensitive instructions, so the built-ins have different
implementations for big- and little-endian, and the instructions must
be marked as killing the vector swap optimization for now.

The first three instructions perform saturating pack operations.  The
fourth performs a modulo pack operation, which means it can be
represented with a vector shuffle, and conversely the appropriate
vector shuffles may cause this instruction to be generated.  The other
instructions are only generated via built-in support for now.

Appropriate tests have been added.

There is a companion patch to clang for the rest of this support.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237499 91177308-0d34-0410-b5e6-96231b3b80d8

include/llvm/IR/IntrinsicsPowerPC.td

@@ -480,6 +480,12 @@ let TargetPrefix = "ppc" in {  // All PPC intrinsics start with "llvm.ppc.".
   def int_ppc_altivec_vpkswus : GCCBuiltin<"__builtin_altivec_vpkswus">,
             Intrinsic<[llvm_v8i16_ty], [llvm_v4i32_ty, llvm_v4i32_ty],
                       [IntrNoMem]>;
+  def int_ppc_altivec_vpksdss : GCCBuiltin<"__builtin_altivec_vpksdss">,
+            Intrinsic<[llvm_v4i32_ty], [llvm_v2i64_ty, llvm_v2i64_ty],
+                      [IntrNoMem]>;
+  def int_ppc_altivec_vpksdus : GCCBuiltin<"__builtin_altivec_vpksdus">,
+            Intrinsic<[llvm_v4i32_ty], [llvm_v2i64_ty, llvm_v2i64_ty],
+                      [IntrNoMem]>;
   // vpkuhum is lowered to a shuffle.
   def int_ppc_altivec_vpkuhus : GCCBuiltin<"__builtin_altivec_vpkuhus">,
             Intrinsic<[llvm_v16i8_ty], [llvm_v8i16_ty, llvm_v8i16_ty],
@@ -488,6 +494,10 @@ let TargetPrefix = "ppc" in {  // All PPC intrinsics start with "llvm.ppc.".
   def int_ppc_altivec_vpkuwus : GCCBuiltin<"__builtin_altivec_vpkuwus">,
             Intrinsic<[llvm_v8i16_ty], [llvm_v4i32_ty, llvm_v4i32_ty],
                       [IntrNoMem]>;
+  // vpkudum is lowered to a shuffle.
+  def int_ppc_altivec_vpkudus : GCCBuiltin<"__builtin_altivec_vpkudus">,
+            Intrinsic<[llvm_v4i32_ty], [llvm_v2i64_ty, llvm_v2i64_ty],
+                      [IntrNoMem]>;
 
   // Unpacks.
   def int_ppc_altivec_vupkhpx : GCCBuiltin<"__builtin_altivec_vupkhpx">,
@@ -496,12 +506,16 @@ let TargetPrefix = "ppc" in {  // All PPC intrinsics start with "llvm.ppc.".
             Intrinsic<[llvm_v8i16_ty], [llvm_v16i8_ty], [IntrNoMem]>;
   def int_ppc_altivec_vupkhsh : GCCBuiltin<"__builtin_altivec_vupkhsh">,
             Intrinsic<[llvm_v4i32_ty], [llvm_v8i16_ty], [IntrNoMem]>;
+  def int_ppc_altivec_vupkhsw : GCCBuiltin<"__builtin_altivec_vupkhsw">,
+            Intrinsic<[llvm_v2i64_ty], [llvm_v4i32_ty], [IntrNoMem]>;
   def int_ppc_altivec_vupklpx : GCCBuiltin<"__builtin_altivec_vupklpx">,
             Intrinsic<[llvm_v4i32_ty], [llvm_v8i16_ty], [IntrNoMem]>;
   def int_ppc_altivec_vupklsb : GCCBuiltin<"__builtin_altivec_vupklsb">,
             Intrinsic<[llvm_v8i16_ty], [llvm_v16i8_ty], [IntrNoMem]>;
   def int_ppc_altivec_vupklsh : GCCBuiltin<"__builtin_altivec_vupklsh">,
             Intrinsic<[llvm_v4i32_ty], [llvm_v8i16_ty], [IntrNoMem]>;
+  def int_ppc_altivec_vupklsw : GCCBuiltin<"__builtin_altivec_vupklsw">,
+            Intrinsic<[llvm_v2i64_ty], [llvm_v4i32_ty], [IntrNoMem]>;
 
 
   // FP <-> integer conversion.

lib/Target/PowerPC/PPCISelLowering.cpp

@@ -1101,7 +1101,7 @@ static bool isConstantOrUndef(int Op, int Val) {
 /// VPKUHUM instruction.
 /// The ShuffleKind distinguishes between big-endian operations with
 /// two different inputs (0), either-endian operations with two identical
-/// inputs (1), and little-endian operantion with two different inputs (2).
+/// inputs (1), and little-endian operations with two different inputs (2).
 /// For the latter, the input operands are swapped (see PPCInstrAltivec.td).
 bool PPC::isVPKUHUMShuffleMask(ShuffleVectorSDNode *N, unsigned ShuffleKind,
                                SelectionDAG &DAG) {
@@ -1132,7 +1132,7 @@ bool PPC::isVPKUHUMShuffleMask(ShuffleVectorSDNode *N, unsigned ShuffleKind,
 /// VPKUWUM instruction.
 /// The ShuffleKind distinguishes between big-endian operations with
 /// two different inputs (0), either-endian operations with two identical
-/// inputs (1), and little-endian operantion with two different inputs (2).
+/// inputs (1), and little-endian operations with two different inputs (2).
 /// For the latter, the input operands are swapped (see PPCInstrAltivec.td).
 bool PPC::isVPKUWUMShuffleMask(ShuffleVectorSDNode *N, unsigned ShuffleKind,
                                SelectionDAG &DAG) {
@@ -1163,6 +1163,49 @@ bool PPC::isVPKUWUMShuffleMask(ShuffleVectorSDNode *N, unsigned ShuffleKind,
   return true;
 }
 
+/// isVPKUDUMShuffleMask - Return true if this is the shuffle mask for a
+/// VPKUDUM instruction.
+/// The ShuffleKind distinguishes between big-endian operations with
+/// two different inputs (0), either-endian operations with two identical
+/// inputs (1), and little-endian operations with two different inputs (2).
+/// For the latter, the input operands are swapped (see PPCInstrAltivec.td).
+bool PPC::isVPKUDUMShuffleMask(ShuffleVectorSDNode *N, unsigned ShuffleKind,
+                               SelectionDAG &DAG) {
+  bool IsLE = DAG.getTarget().getDataLayout()->isLittleEndian();
+  if (ShuffleKind == 0) {
+    if (IsLE)
+      return false;
+    for (unsigned i = 0; i != 16; i += 4)
+      if (!isConstantOrUndef(N->getMaskElt(i  ),  i*2+4) ||
+          !isConstantOrUndef(N->getMaskElt(i+1),  i*2+5) ||
+          !isConstantOrUndef(N->getMaskElt(i+2),  i*2+6) ||
+          !isConstantOrUndef(N->getMaskElt(i+3),  i*2+7))
+        return false;
+  } else if (ShuffleKind == 2) {
+    if (!IsLE)
+      return false;
+    for (unsigned i = 0; i != 16; i += 4)
+      if (!isConstantOrUndef(N->getMaskElt(i  ),  i*2) ||
+          !isConstantOrUndef(N->getMaskElt(i+1),  i*2+1) ||
+          !isConstantOrUndef(N->getMaskElt(i+2),  i*2+2) ||
+          !isConstantOrUndef(N->getMaskElt(i+3),  i*2+3))
+        return false;
+  } else if (ShuffleKind == 1) {
+    unsigned j = IsLE ? 0 : 4;
+    for (unsigned i = 0; i != 8; i += 4)
+      if (!isConstantOrUndef(N->getMaskElt(i  ),  i*2+j)   ||
+          !isConstantOrUndef(N->getMaskElt(i+1),  i*2+j+1) ||
+          !isConstantOrUndef(N->getMaskElt(i+2),  i*2+j+2) ||
+          !isConstantOrUndef(N->getMaskElt(i+3),  i*2+j+3) ||
+          !isConstantOrUndef(N->getMaskElt(i+8),  i*2+j)   ||
+          !isConstantOrUndef(N->getMaskElt(i+9),  i*2+j+1) ||
+          !isConstantOrUndef(N->getMaskElt(i+10), i*2+j+2) ||
+          !isConstantOrUndef(N->getMaskElt(i+11), i*2+j+3))
+        return false;
+  }
+  return true;
+}
+
 /// isVMerge - Common function, used to match vmrg* shuffles.
 ///
 static bool isVMerge(ShuffleVectorSDNode *N, unsigned UnitSize,
@@ -6993,6 +7036,7 @@ SDValue PPCTargetLowering::LowerVECTOR_SHUFFLE(SDValue Op,
         PPC::isSplatShuffleMask(SVOp, 4) ||
         PPC::isVPKUWUMShuffleMask(SVOp, 1, DAG) ||
         PPC::isVPKUHUMShuffleMask(SVOp, 1, DAG) ||
+        PPC::isVPKUDUMShuffleMask(SVOp, 1, DAG) ||
         PPC::isVSLDOIShuffleMask(SVOp, 1, DAG) != -1 ||
         PPC::isVMRGLShuffleMask(SVOp, 1, 1, DAG) ||
         PPC::isVMRGLShuffleMask(SVOp, 2, 1, DAG) ||
@@ -7010,6 +7054,7 @@ SDValue PPCTargetLowering::LowerVECTOR_SHUFFLE(SDValue Op,
   unsigned int ShuffleKind = isLittleEndian ? 2 : 0;
   if (PPC::isVPKUWUMShuffleMask(SVOp, ShuffleKind, DAG) ||
       PPC::isVPKUHUMShuffleMask(SVOp, ShuffleKind, DAG) ||
+      PPC::isVPKUDUMShuffleMask(SVOp, ShuffleKind, DAG) ||
       PPC::isVSLDOIShuffleMask(SVOp, ShuffleKind, DAG) != -1 ||
       PPC::isVMRGLShuffleMask(SVOp, 1, ShuffleKind, DAG) ||
       PPC::isVMRGLShuffleMask(SVOp, 2, ShuffleKind, DAG) ||

lib/Target/PowerPC/PPCISelLowering.h

@@ -357,6 +357,11 @@ namespace llvm {
     bool isVPKUWUMShuffleMask(ShuffleVectorSDNode *N, unsigned ShuffleKind,
                               SelectionDAG &DAG);
 
+    /// isVPKUDUMShuffleMask - Return true if this is the shuffle mask for a
+    /// VPKUDUM instruction.
+    bool isVPKUDUMShuffleMask(ShuffleVectorSDNode *N, unsigned ShuffleKind,
+                              SelectionDAG &DAG);
+
     /// isVMRGLShuffleMask - Return true if this is a shuffle mask suitable for
     /// a VRGL* instruction with the specified unit size (1,2 or 4 bytes).
     bool isVMRGLShuffleMask(ShuffleVectorSDNode *N, unsigned UnitSize,

lib/Target/PowerPC/PPCInstrAltivec.td

@@ -43,6 +43,10 @@ def vpkuwum_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
                               (vector_shuffle node:$lhs, node:$rhs), [{
   return PPC::isVPKUWUMShuffleMask(cast<ShuffleVectorSDNode>(N), 0, *CurDAG);
 }]>;
+def vpkudum_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
+                              (vector_shuffle node:$lhs, node:$rhs), [{
+  return PPC::isVPKUDUMShuffleMask(cast<ShuffleVectorSDNode>(N), 0, *CurDAG);
+}]>;
 def vpkuhum_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
                                     (vector_shuffle node:$lhs, node:$rhs), [{
   return PPC::isVPKUHUMShuffleMask(cast<ShuffleVectorSDNode>(N), 1, *CurDAG);
@@ -51,6 +55,10 @@ def vpkuwum_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
                                     (vector_shuffle node:$lhs, node:$rhs), [{
   return PPC::isVPKUWUMShuffleMask(cast<ShuffleVectorSDNode>(N), 1, *CurDAG);
 }]>;
+def vpkudum_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
+                                    (vector_shuffle node:$lhs, node:$rhs), [{
+  return PPC::isVPKUDUMShuffleMask(cast<ShuffleVectorSDNode>(N), 1, *CurDAG);
+}]>;
 
 // These fragments are provided for little-endian, where the inputs must be
 // swapped for correct semantics.
@@ -62,6 +70,10 @@ def vpkuwum_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
                                       (vector_shuffle node:$lhs, node:$rhs), [{
   return PPC::isVPKUWUMShuffleMask(cast<ShuffleVectorSDNode>(N), 2, *CurDAG);
 }]>;
+def vpkudum_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
+                                      (vector_shuffle node:$lhs, node:$rhs), [{
+  return PPC::isVPKUDUMShuffleMask(cast<ShuffleVectorSDNode>(N), 2, *CurDAG);
+}]>;
 
 def vmrglb_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
                              (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
@@ -1091,6 +1103,29 @@ def VPMSUMD : VX1_Int_Ty<1224, "vpmsumd",
 def VPERMXOR : VA1a_Int_Ty<45, "vpermxor",
                          int_ppc_altivec_crypto_vpermxor, v16i8>;
 
+// Vector doubleword integer pack and unpack.
+def VPKSDSS : VX1_Int_Ty2<1486, "vpksdss", int_ppc_altivec_vpksdss,
+                          v4i32, v2i64>;
+def VPKSDUS : VX1_Int_Ty2<1358, "vpksdus", int_ppc_altivec_vpksdus,
+                          v4i32, v2i64>;
+def VPKUDUM : VXForm_1<1102, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
+                       "vpkudum $vD, $vA, $vB", IIC_VecFP,
+                       [(set v16i8:$vD,
+                         (vpkudum_shuffle v16i8:$vA, v16i8:$vB))]>;
+def VPKUDUS : VX1_Int_Ty2<1230, "vpkudus", int_ppc_altivec_vpkudus,
+                          v4i32, v2i64>;
+def VUPKHSW : VX2_Int_Ty2<1614, "vupkhsw", int_ppc_altivec_vupkhsw,
+                          v2i64, v4i32>;
+def VUPKLSW : VX2_Int_Ty2<1742, "vupklsw", int_ppc_altivec_vupklsw,
+                          v2i64, v4i32>;
+
+// Shuffle patterns for unary and swapped (LE) vector pack modulo.
+def:Pat<(vpkudum_unary_shuffle v16i8:$vA, undef),
+        (VPKUDUM $vA, $vA)>;
+def:Pat<(vpkudum_swapped_shuffle v16i8:$vA, v16i8:$vB),
+        (VPKUDUM $vB, $vA)>;
+
+
 } // end HasP8Altivec
 
 // Crypto instructions (from builtins)

lib/Target/PowerPC/PPCVSXSwapRemoval.cpp

@@ -382,8 +382,12 @@ bool PPCVSXSwapRemoval::gatherVectorInstructions() {
       case PPC::VPKPX:
       case PPC::VPKSHSS:
       case PPC::VPKSHUS:
+      case PPC::VPKSDSS:
+      case PPC::VPKSDUS:
       case PPC::VPKSWSS:
       case PPC::VPKSWUS:
+      case PPC::VPKUDUM:
+      case PPC::VPKUDUS:
       case PPC::VPKUHUM:
       case PPC::VPKUHUS:
       case PPC::VPKUWUM:
@@ -412,9 +416,11 @@ bool PPCVSXSwapRemoval::gatherVectorInstructions() {
       case PPC::VUPKHPX:
       case PPC::VUPKHSB:
       case PPC::VUPKHSH:
+      case PPC::VUPKHSW:
       case PPC::VUPKLPX:
       case PPC::VUPKLSB:
       case PPC::VUPKLSH:
+      case PPC::VUPKLSW:
       case PPC::XXMRGHW:
       case PPC::XXMRGLW:
       case PPC::XXSPLTW:

test/CodeGen/PowerPC/vec_shuffle_p8vector.ll

@@ -0,0 +1,43 @@
+; RUN: llc -mcpu=pwr8 -mtriple=powerpc64-unknown-linux-gnu -mattr=+power8-vector < %s | FileCheck %s
+
+define void @VPKUDUM_unary(<2 x i64>* %A) {
+entry:
+        %tmp = load <2 x i64>, <2 x i64>* %A
+        %tmp2 = bitcast <2 x i64> %tmp to <4 x i32>
+        %tmp3 = extractelement <4 x i32> %tmp2, i32 1
+        %tmp4 = extractelement <4 x i32> %tmp2, i32 3
+        %tmp5 = insertelement <4 x i32> undef, i32 %tmp3, i32 0
+        %tmp6 = insertelement <4 x i32> %tmp5, i32 %tmp4, i32 1
+        %tmp7 = insertelement <4 x i32> %tmp6, i32 %tmp3, i32 2
+        %tmp8 = insertelement <4 x i32> %tmp7, i32 %tmp4, i32 3
+        %tmp9 = bitcast <4 x i32> %tmp8 to <2 x i64>
+        store <2 x i64> %tmp9, <2 x i64>* %A
+        ret void
+}
+
+; CHECK-LABEL: @VPKUDUM_unary
+; CHECK-NOT:   vperm
+; CHECK:       vpkudum
+
+define void @VPKUDUM(<2 x i64>* %A, <2 x i64>* %B) {
+entry:
+        %tmp = load <2 x i64>, <2 x i64>* %A
+        %tmp2 = bitcast <2 x i64> %tmp to <4 x i32>
+        %tmp3 = load <2 x i64>, <2 x i64>* %B
+        %tmp4 = bitcast <2 x i64> %tmp3 to <4 x i32>
+        %tmp5 = extractelement <4 x i32> %tmp2, i32 1
+        %tmp6 = extractelement <4 x i32> %tmp2, i32 3
+        %tmp7 = extractelement <4 x i32> %tmp4, i32 1
+        %tmp8 = extractelement <4 x i32> %tmp4, i32 3
+        %tmp9 = insertelement <4 x i32> undef, i32 %tmp5, i32 0
+        %tmp10 = insertelement <4 x i32> %tmp9, i32 %tmp6, i32 1
+        %tmp11 = insertelement <4 x i32> %tmp10, i32 %tmp7, i32 2
+        %tmp12 = insertelement <4 x i32> %tmp11, i32 %tmp8, i32 3
+        %tmp13 = bitcast <4 x i32> %tmp12 to <2 x i64>
+        store <2 x i64> %tmp13, <2 x i64>* %A
+        ret void
+}
+
+; CHECK-LABEL: @VPKUDUM
+; CHECK-NOT:   vperm
+; CHECK:       vpkudum

test/CodeGen/PowerPC/vec_shuffle_p8vector_le.ll

@@ -0,0 +1,43 @@
+; RUN: llc -mcpu=pwr8 -mtriple=powerpc64le-unknown-linux-gnu -mattr=+power8-vector < %s | FileCheck %s
+
+define void @VPKUDUM_unary(<2 x i64>* %A) {
+entry:
+	%tmp = load <2 x i64>, <2 x i64>* %A
+	%tmp2 = bitcast <2 x i64> %tmp to <4 x i32>
+	%tmp3 = extractelement <4 x i32> %tmp2, i32 0
+	%tmp4 = extractelement <4 x i32> %tmp2, i32 2
+	%tmp5 = insertelement <4 x i32> undef, i32 %tmp3, i32 0
+	%tmp6 = insertelement <4 x i32> %tmp5, i32 %tmp4, i32 1
+	%tmp7 = insertelement <4 x i32> %tmp6, i32 %tmp3, i32 2
+	%tmp8 = insertelement <4 x i32> %tmp7, i32 %tmp4, i32 3
+	%tmp9 = bitcast <4 x i32> %tmp8 to <2 x i64>
+	store <2 x i64> %tmp9, <2 x i64>* %A
+	ret void
+}
+
+; CHECK-LABEL: @VPKUDUM_unary
+; CHECK-NOT:   vperm
+; CHECK:       vpkudum
+
+define void @VPKUDUM(<2 x i64>* %A, <2 x i64>* %B) {
+entry:
+	%tmp = load <2 x i64>, <2 x i64>* %A
+	%tmp2 = bitcast <2 x i64> %tmp to <4 x i32>
+        %tmp3 = load <2 x i64>, <2 x i64>* %B
+        %tmp4 = bitcast <2 x i64> %tmp3 to <4 x i32>
+	%tmp5 = extractelement <4 x i32> %tmp2, i32 0
+	%tmp6 = extractelement <4 x i32> %tmp2, i32 2
+	%tmp7 = extractelement <4 x i32> %tmp4, i32 0
+	%tmp8 = extractelement <4 x i32> %tmp4, i32 2
+	%tmp9 = insertelement <4 x i32> undef, i32 %tmp5, i32 0
+	%tmp10 = insertelement <4 x i32> %tmp9, i32 %tmp6, i32 1
+	%tmp11 = insertelement <4 x i32> %tmp10, i32 %tmp7, i32 2
+	%tmp12 = insertelement <4 x i32> %tmp11, i32 %tmp8, i32 3
+	%tmp13 = bitcast <4 x i32> %tmp12 to <2 x i64>
+	store <2 x i64> %tmp13, <2 x i64>* %A
+	ret void
+}
+
+; CHECK-LABEL: @VPKUDUM
+; CHECK-NOT:   vperm
+; CHECK:       vpkudum

test/MC/Disassembler/PowerPC/ppc64-encoding-p8vector.txt

@@ -0,0 +1,19 @@
+# RUN: llvm-mc --disassemble %s -triple powerpc64-unknown-unknown -mcpu=pwr8 | FileCheck %s
+
+# CHECK: vpksdss 2, 3, 4
+0x10 0x43 0x25 0xce
+
+# CHECK: vpksdus 2, 3, 4
+0x10 0x43 0x25 0x4e
+
+# CHECK: vpkudus 2, 3, 4
+0x10 0x43 0x24 0xce
+
+# CHECK: vpkudum 2, 3, 4
+0x10 0x43 0x24 0x4e
+
+# CHECK: vupkhsw 2, 3
+0x10 0x40 0x1e 0x4e
+
+# CHECK: vupklsw 2, 3
+0x10 0x40 0x1e 0xce

test/MC/PowerPC/ppc64-encoding-p8vector.s

@@ -0,0 +1,26 @@
+# RUN: llvm-mc -triple powerpc64-unknown-unknown --show-encoding %s | FileCheck -check-prefix=CHECK-BE %s
+# RUN: llvm-mc -triple powerpc64le-unknown-unknown --show-encoding %s | FileCheck -check-prefix=CHECK-LE %s
+
+# CHECK-BE: vpksdss 2, 3, 4                 # encoding: [0x10,0x43,0x25,0xce]
+# CHECK-LE: vpksdss 2, 3, 4                 # encoding: [0xce,0x25,0x43,0x10]
+            vpksdss 2, 3, 4
+
+# CHECK-BE: vpksdus 2, 3, 4                 # encoding: [0x10,0x43,0x25,0x4e]
+# CHECK-LE: vpksdus 2, 3, 4                 # encoding: [0x4e,0x25,0x43,0x10]
+            vpksdus 2, 3, 4
+
+# CHECK-BE: vpkudus 2, 3, 4                 # encoding: [0x10,0x43,0x24,0xce]
+# CHECK-LE: vpkudus 2, 3, 4                 # encoding: [0xce,0x24,0x43,0x10]
+            vpkudus 2, 3, 4
+
+# CHECK-BE: vpkudum 2, 3, 4                 # encoding: [0x10,0x43,0x24,0x4e]
+# CHECK-LE: vpkudum 2, 3, 4                 # encoding: [0x4e,0x24,0x43,0x10]
+            vpkudum 2, 3, 4
+
+# CHECK-BE: vupkhsw 2, 3                    # encoding: [0x10,0x40,0x1e,0x4e]
+# CHECK-LE: vupkhsw 2, 3                    # encoding: [0x4e,0x1e,0x40,0x10]
+            vupkhsw 2, 3
+
+# CHECK-BE: vupklsw 2, 3                    # encoding: [0x10,0x40,0x1e,0xce]
+# CHECK-LE: vupklsw 2, 3                    # encoding: [0xce,0x1e,0x40,0x10]
+            vupklsw 2, 3