[SelectionDAG] Add a signed integer absolute ISD node

[RKSimon]

Summary

Reduced version of D26357 - based on the discussion on llvm-dev about canonicalization of UMIN/UMAX/SMIN/SMAX as well as ABS I've reduced that patch to just the ABS ISD node (with x86/sse support) to improve basic combines and lowering.

AFAICT ARM/AArch64, PowerPC and NVPTX all have similar instructions so I'd like to make this a generic opcode and move us away from the hard coded tablegen patterns which makes it tricky to match more complex patterns.

At the moment this patch doesn't attempt legalization (and I only create an ABS node if its legal/custom) - I can add the legalization code from D26357 if people think it useful at this stage - it will let us do some extra combines, knownbits handling etc.

[mkuper]

I think this is a good idea, in terms of the usual trade-off (we lose the opportunity to combine the xor with other logic, simplifying it, etc.) Especially since we only introduce this post-legalization.
But I'd like to hear what other people think too.

include/llvm/CodeGen/ISDOpcodes.h

343

Is this the right behavior for a target-independent intrinsic? That is, do the other (non-x86) platforms that have an ABS instruction have the same behavior?

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

5502

:-(
(I was really surprised by this, but looks like we have this pattern everywhere.)

lib/Target/X86/X86ISelLowering.cpp

20853

Is there a reason not to use ISD::ABS explicitly here? I think it'd be clearer.

[RKSimon]

Thanks Michael - does anyone from a non-X86 target have any comments?

include/llvm/CodeGen/ISDOpcodes.h

343

AFAICT this is the norm and matches what we do for constant folding of the std::abs() function as well, although technically its undefined if the result isn't representable.

ARM NEON has the same functionality with its basic vabs_* intrinsic, and has a saturating version vqabs_* as well.
PowerPC has vec_abs / vec_abss.
Not sure about the other targets?

Saturating patterns can probably be best be handled by abs(smax(v, INT_MIN + 1)) style canonicalizations? But that would still have to be target specific.

lib/Target/X86/X86ISelLowering.cpp

20853

That's me trying to be too generic - will fix.

[RKSimon]

rebased and tidied up X86 LowerABS

[RKSimon]

@hfinkel @jmolloy Is this approach compatible with PowerPC and ARM? Cheers.

[jmolloy]

Hi,

Having asked around: The way we define this is that the VABS instruction takes a signed integer and outputs an unsigned integer, getting around this problem.

However, I believe the output of VABS(INT_MIN) is indeed bit-identical to INT_MIN.

+ @jgreenhalgh to confirm I haven't mangled his explanation.

James

[jgreenhalgh]

In D29639#682183, @jmolloy wrote:

Hi,

Having asked around: The way we define this is that the VABS instruction takes a signed integer and outputs an unsigned integer, getting around this problem.

However, I believe the output of VABS(INT_MIN) is indeed bit-identical to INT_MIN.

+ @jgreenhalgh to confirm I haven't mangled his explanation.

Only a minor tweak on a pedantic point - the instruction doesn't really have an idea of signed/unsigned - just bits. The effect is as if you were calculating in an unsigned value of the same number of bits as the input.

That doesn't hold for the intrinsics which are all signed -> signed. But the behaviour there is ABS(INT_MIN) -> INT_MIN, so the behaviour in this patch looks fine for ARM/AArch64 Advanced SIMD.

[RKSimon]

Rebased and made it explicit in the ISDOpcodes.h that the input and result of ISD::ABS have the same bitwidth.

[RKSimon]

Does anyone have any other comments? What about legalization - should I add it as part of this patch or just keep to the legal cases for now?

[RKSimon]

ping?

[RKSimon]

ping^2

[efriedma]

Would it make sense to use ISD::ABS over combineIntegerAbs in X86ISelLowering.cpp?

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

5509

There are a lot of other folds we could do here (e.g. abs(-x) -> abs(x)), but maybe not worth worrying about?

[RKSimon]

In D29639#699976, @efriedma wrote:

Would it make sense to use ISD::ABS over combineIntegerAbs in X86ISelLowering.cpp?

Yes, although probably as a followup once the legalization support is in place.

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

5509

Definitely scope for further combines, but I'd rather see them in the wild before adding them.

[kparzysz]

In D29639#678538, @RKSimon wrote:

Thanks Michael - does anyone from a non-X86 target have any comments?

Hexagon has integer ABS for 32- and 64-bit integers, and for both of them abs(min_int) == min_int, so this works for us. FWIW, we also have a saturating abs for 32-bit integers.

[efriedma]

LGTM.

It looks like we have consensus that this is generally useful, and the implementation looks fine.

include/llvm/CodeGen/ISDOpcodes.h

enum NodeType {
 
     /// [US]{MIN/MAX} - Binary minimum or maximum or signed or unsigned
     /// integers.
     SMIN, SMAX, UMIN, UMAX,
 
     /// Bitwise operators - logical and, logical or, logical xor.
     AND, OR, XOR,
 
+    /// ABS - Determine the unsigned absolute value of a signed integer value.
+    /// Note: A value of INT_MIN will return INT_MIN, no saturation or overflow
+    /// is performed.
+    ABS,
+
     /// Shift and rotation operations.  After legalization, the type of the
     /// shift amount is known to be TLI.getShiftAmountTy().  Before legalization
     /// the shift amount can be any type, but care must be taken to ensure it is
     /// large enough.  TLI.getShiftAmountTy() is i8 on some targets, but before
     /// legalization, types like i1024 can occur and i8 doesn't have enough bits
     /// to represent the shift amount.
     /// When the 1st operand is a vector, the shift amount must be in the same
     /// type. (TLI.getShiftAmountTy() will return the same type when the input

include/llvm/Target/TargetSelectionDAG.td

 def smax       : SDNode<"ISD::SMAX"      , SDTIntBinOp,
                                   [SDNPCommutative, SDNPAssociative]>;
 def umin       : SDNode<"ISD::UMIN"      , SDTIntBinOp,
                                   [SDNPCommutative, SDNPAssociative]>;
 def umax       : SDNode<"ISD::UMAX"      , SDTIntBinOp,
                                   [SDNPCommutative, SDNPAssociative]>;
 
 def sext_inreg : SDNode<"ISD::SIGN_EXTEND_INREG", SDTExtInreg>;
+def abs        : SDNode<"ISD::ABS"        , SDTIntUnaryOp>;
 def bitreverse : SDNode<"ISD::BITREVERSE" , SDTIntUnaryOp>;
 def bswap      : SDNode<"ISD::BSWAP"      , SDTIntUnaryOp>;
 def ctlz       : SDNode<"ISD::CTLZ"       , SDTIntUnaryOp>;
 def cttz       : SDNode<"ISD::CTTZ"       , SDTIntUnaryOp>;
 def ctpop      : SDNode<"ISD::CTPOP"      , SDTIntUnaryOp>;
 def ctlz_zero_undef : SDNode<"ISD::CTLZ_ZERO_UNDEF", SDTIntUnaryOp>;
 def cttz_zero_undef : SDNode<"ISD::CTTZ_ZERO_UNDEF", SDTIntUnaryOp>;
 def sext       : SDNode<"ISD::SIGN_EXTEND", SDTIntExtendOp>;

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

private:
     SDValue visitOR(SDNode *N);
     SDValue visitORLike(SDValue N0, SDValue N1, SDNode *LocReference);
     SDValue visitXOR(SDNode *N);
     SDValue SimplifyVBinOp(SDNode *N);
     SDValue visitSHL(SDNode *N);
     SDValue visitSRA(SDNode *N);
     SDValue visitSRL(SDNode *N);
     SDValue visitRotate(SDNode *N);
+    SDValue visitABS(SDNode *N);
     SDValue visitBSWAP(SDNode *N);
     SDValue visitBITREVERSE(SDNode *N);
     SDValue visitCTLZ(SDNode *N);
     SDValue visitCTLZ_ZERO_UNDEF(SDNode *N);
     SDValue visitCTTZ(SDNode *N);
     SDValue visitCTTZ_ZERO_UNDEF(SDNode *N);
     SDValue visitCTPOP(SDNode *N);
     SDValue visitSELECT(SDNode *N);
SDValue DAGCombiner::visit(SDNode *N) {
   case ISD::AND:                return visitAND(N);
   case ISD::OR:                 return visitOR(N);
   case ISD::XOR:                return visitXOR(N);
   case ISD::SHL:                return visitSHL(N);
   case ISD::SRA:                return visitSRA(N);
   case ISD::SRL:                return visitSRL(N);
   case ISD::ROTR:
   case ISD::ROTL:               return visitRotate(N);
+  case ISD::ABS:                return visitABS(N);
   case ISD::BSWAP:              return visitBSWAP(N);
   case ISD::BITREVERSE:         return visitBITREVERSE(N);
   case ISD::CTLZ:               return visitCTLZ(N);
   case ISD::CTLZ_ZERO_UNDEF:    return visitCTLZ_ZERO_UNDEF(N);
   case ISD::CTTZ:               return visitCTTZ(N);
   case ISD::CTTZ_ZERO_UNDEF:    return visitCTTZ_ZERO_UNDEF(N);
   case ISD::CTPOP:              return visitCTPOP(N);
   case ISD::SELECT:             return visitSELECT(N);
if (N1C && N0.getOpcode() == ISD::XOR) {
     }
     if (const ConstantSDNode *N01C = getAsNonOpaqueConstant(N0.getOperand(1))) {
       SDLoc DL(N);
       return DAG.getNode(ISD::XOR, DL, VT, N0.getOperand(0),
                          DAG.getConstant(N1C->getAPIntValue() ^
                                          N01C->getAPIntValue(), DL, VT));
     }
   }
+
+  // fold Y = sra (X, size(X)-1); xor (add (X, Y), Y) -> (abs X)
+  unsigned OpSizeInBits = VT.getScalarSizeInBits();
+  if (N0.getOpcode() == ISD::ADD && N0.getOperand(1) == N1 &&
+      N1.getOpcode() == ISD::SRA && N1.getOperand(0) == N0.getOperand(0) &&
+      TLI.isOperationLegalOrCustom(ISD::ABS, VT)) {
+    if (ConstantSDNode *C = isConstOrConstSplat(N1.getOperand(1)))
+      if (C->getAPIntValue() == (OpSizeInBits - 1))
+        return DAG.getNode(ISD::ABS, SDLoc(N), VT, N0.getOperand(0));
+  }
+
   // fold (xor x, x) -> 0
   if (N0 == N1)
     return tryFoldToZero(SDLoc(N), TLI, VT, DAG, LegalOperations, LegalTypes);
 
   // fold (xor (shl 1, x), -1) -> (rotl ~1, x)
   // Here is a concrete example of this equivalence:
   // i16   x ==  14
   // i16 shl ==   1 << 14  == 16384 == 0b0100000000000000
else if (Use->getOpcode() == ISD::TRUNCATE && Use->hasOneUse()) {
       if (Use->getOpcode() == ISD::BRCOND)
         AddToWorklist(Use);
     }
   }
 
   return SDValue();
 }
 
+SDValue DAGCombiner::visitABS(SDNode *N) {
+  SDValue N0 = N->getOperand(0);
+  EVT VT = N->getValueType(0);
+
+  // fold (abs c1) -> c2
+  if (DAG.isConstantIntBuildVectorOrConstantInt(N0))
+    return DAG.getNode(ISD::ABS, SDLoc(N), VT, N0);
+  // fold (abs (abs x)) -> (abs x)
+  if (N0.getOpcode() == ISD::ABS)
+    return N0;
+  // fold (abs x) -> x iff not-negative
+  if (DAG.SignBitIsZero(N0))
+    return N0;
+  return SDValue();
+}
+
 SDValue DAGCombiner::visitBSWAP(SDNode *N) {
   SDValue N0 = N->getOperand(0);
   EVT VT = N->getValueType(0);
 
   // fold (bswap c1) -> c2
   if (DAG.isConstantIntBuildVectorOrConstantInt(N0))
     return DAG.getNode(ISD::BSWAP, SDLoc(N), VT, N0);
   // fold (bswap (bswap x)) -> x

lib/CodeGen/SelectionDAG/SelectionDAG.cpp

case ISD::BITCAST:
         return getConstantFP(APFloat(APFloat::IEEEhalf(), Val), DL, VT);
       if (VT == MVT::f32 && C->getValueType(0) == MVT::i32)
         return getConstantFP(APFloat(APFloat::IEEEsingle(), Val), DL, VT);
       if (VT == MVT::f64 && C->getValueType(0) == MVT::i64)
         return getConstantFP(APFloat(APFloat::IEEEdouble(), Val), DL, VT);
       if (VT == MVT::f128 && C->getValueType(0) == MVT::i128)
         return getConstantFP(APFloat(APFloat::IEEEquad(), Val), DL, VT);
       break;
+    case ISD::ABS:
+      return getConstant(Val.abs(), DL, VT, C->isTargetOpcode(),
+                         C->isOpaque());
     case ISD::BITREVERSE:
       return getConstant(Val.reverseBits(), DL, VT, C->isTargetOpcode(),
                          C->isOpaque());
     case ISD::BSWAP:
       return getConstant(Val.byteSwap(), DL, VT, C->isTargetOpcode(),
                          C->isOpaque());
     case ISD::CTPOP:
       return getConstant(Val.countPopulation(), DL, VT, C->isTargetOpcode(),
if (BV->isConstant()) {
       case ISD::FTRUNC:
       case ISD::FFLOOR:
       case ISD::FP_EXTEND:
       case ISD::FP_TO_SINT:
       case ISD::FP_TO_UINT:
       case ISD::TRUNCATE:
       case ISD::UINT_TO_FP:
       case ISD::SINT_TO_FP:
+      case ISD::ABS:
       case ISD::BITREVERSE:
       case ISD::BSWAP:
       case ISD::CTLZ:
       case ISD::CTLZ_ZERO_UNDEF:
       case ISD::CTTZ:
       case ISD::CTTZ_ZERO_UNDEF:
       case ISD::CTPOP: {
         SDValue Ops = { Operand };
if (OpOpcode == ISD::ZERO_EXTEND || OpOpcode == ISD::SIGN_EXTEND ||
             .bitsLT(VT.getScalarType()))
         return getNode(OpOpcode, DL, VT, Operand.getNode()->getOperand(0));
       if (Operand.getNode()->getOperand(0).getValueType().bitsGT(VT))
         return getNode(ISD::TRUNCATE, DL, VT, Operand.getNode()->getOperand(0));
       return Operand.getNode()->getOperand(0);
     }
     if (OpOpcode == ISD::UNDEF)
       return getUNDEF(VT);
       break;
+  case ISD::ABS:
+    assert(VT.isInteger() && VT == Operand.getValueType() &&
+           "Invalid ABS!");
+    if (OpOpcode == ISD::UNDEF)
+      return getUNDEF(VT);
+    break;
   case ISD::BSWAP:
     assert(VT.isInteger() && VT == Operand.getValueType() &&
            "Invalid BSWAP!");
     assert((VT.getScalarSizeInBits() % 16 == 0) &&
            "BSWAP types must be a multiple of 16 bits!");
     if (OpOpcode == ISD::UNDEF)
       return getUNDEF(VT);
     break;

lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp

#endif
   case ISD::DEBUGTRAP:                  return "debugtrap";
   case ISD::LIFETIME_START:             return "lifetime.start";
   case ISD::LIFETIME_END:               return "lifetime.end";
   case ISD::GC_TRANSITION_START:        return "gc_transition.start";
   case ISD::GC_TRANSITION_END:          return "gc_transition.end";
   case ISD::GET_DYNAMIC_AREA_OFFSET:    return "get.dynamic.area.offset";
 
   // Bit manipulation
+  case ISD::ABS:                        return "abs";
   case ISD::BITREVERSE:                 return "bitreverse";
   case ISD::BSWAP:                      return "bswap";
   case ISD::CTPOP:                      return "ctpop";
   case ISD::CTTZ:                       return "cttz";
   case ISD::CTTZ_ZERO_UNDEF:            return "cttz_zero_undef";
   case ISD::CTLZ:                       return "ctlz";
   case ISD::CTLZ_ZERO_UNDEF:            return "ctlz_zero_undef";
 

lib/CodeGen/TargetLoweringBase.cpp

for (MVT VT : MVT::all_valuetypes()) {
     setOperationAction(ISD::FMAXNUM, VT, Expand);
     setOperationAction(ISD::FMINNAN, VT, Expand);
     setOperationAction(ISD::FMAXNAN, VT, Expand);
     setOperationAction(ISD::FMAD, VT, Expand);
     setOperationAction(ISD::SMIN, VT, Expand);
     setOperationAction(ISD::SMAX, VT, Expand);
     setOperationAction(ISD::UMIN, VT, Expand);
     setOperationAction(ISD::UMAX, VT, Expand);
+    setOperationAction(ISD::ABS, VT, Expand);
 
     // Overflow operations default to expand
     setOperationAction(ISD::SADDO, VT, Expand);
     setOperationAction(ISD::SSUBO, VT, Expand);
     setOperationAction(ISD::UADDO, VT, Expand);
     setOperationAction(ISD::USUBO, VT, Expand);
     setOperationAction(ISD::SMULO, VT, Expand);
     setOperationAction(ISD::UMULO, VT, Expand);

lib/Target/X86/X86ISelLowering.h

enum NodeType : unsigned {
       /// Integer horizontal add/sub.
       HADD,
       HSUB,
 
       /// Floating point horizontal add/sub.
       FHADD,
       FHSUB,
 
-      // Integer absolute value
-      ABS,
-
       // Detect Conflicts Within a Vector
       CONFLICT,
 
       /// Floating point max and min.
       FMAX, FMIN,
 
       /// Commutative FMIN and FMAX.
       FMAXC, FMINC,

lib/Target/X86/X86ISelLowering.cpp

if (!Subtarget.useSoftFloat() && Subtarget.hasSSE2()) {
     for (auto VT : { MVT::v4i32, MVT::v2i64 }) {
       setOperationAction(ISD::SRL, VT, Custom);
       setOperationAction(ISD::SHL, VT, Custom);
       setOperationAction(ISD::SRA, VT, Custom);
     }
   }
 
   if (!Subtarget.useSoftFloat() && Subtarget.hasSSSE3()) {
+    setOperationAction(ISD::ABS,                MVT::v16i8, Legal);
+    setOperationAction(ISD::ABS,                MVT::v8i16, Legal);
+    setOperationAction(ISD::ABS,                MVT::v4i32, Legal);
     setOperationAction(ISD::BITREVERSE,         MVT::v16i8, Custom);
     setOperationAction(ISD::CTLZ,               MVT::v16i8, Custom);
     setOperationAction(ISD::CTLZ,               MVT::v8i16, Custom);
     setOperationAction(ISD::CTLZ,               MVT::v4i32, Custom);
     setOperationAction(ISD::CTLZ,               MVT::v2i64, Custom);
   }
 
   if (!Subtarget.useSoftFloat() && Subtarget.hasSSE41()) {
if (!Subtarget.useSoftFloat() && Subtarget.hasFp256()) {
     setOperationAction(ISD::SMUL_LOHI, MVT::v8i32,  Custom);
 
     setOperationAction(ISD::MULHU,     MVT::v16i16, HasInt256 ? Legal : Custom);
     setOperationAction(ISD::MULHS,     MVT::v16i16, HasInt256 ? Legal : Custom);
     setOperationAction(ISD::MULHU,     MVT::v32i8,  Custom);
     setOperationAction(ISD::MULHS,     MVT::v32i8,  Custom);
 
     for (auto VT : { MVT::v32i8, MVT::v16i16, MVT::v8i32 }) {
+      setOperationAction(ISD::ABS,  VT, HasInt256 ? Legal : Custom);
       setOperationAction(ISD::SMAX, VT, HasInt256 ? Legal : Custom);
       setOperationAction(ISD::UMAX, VT, HasInt256 ? Legal : Custom);
       setOperationAction(ISD::SMIN, VT, HasInt256 ? Legal : Custom);
       setOperationAction(ISD::UMIN, VT, HasInt256 ? Legal : Custom);
     }
 
     if (HasInt256) {
       setOperationAction(ISD::SIGN_EXTEND_VECTOR_INREG, MVT::v4i64,  Custom);
if (Subtarget.hasDQI()) {
       if (Subtarget.hasVLX()) {
         // Fast v2f32 SINT_TO_FP( v2i32 ) custom conversion.
         setOperationAction(ISD::SINT_TO_FP,    MVT::v2f32, Custom);
         setOperationAction(ISD::FP_TO_SINT,    MVT::v2f32, Custom);
         setOperationAction(ISD::FP_TO_UINT,    MVT::v2f32, Custom);
       }
     }
     if (Subtarget.hasVLX()) {
+      setOperationAction(ISD::ABS,              MVT::v4i64, Legal);
+      setOperationAction(ISD::ABS,              MVT::v2i64, Legal);
       setOperationAction(ISD::SINT_TO_FP,       MVT::v8i32, Legal);
       setOperationAction(ISD::UINT_TO_FP,       MVT::v8i32, Legal);
       setOperationAction(ISD::FP_TO_SINT,       MVT::v8i32, Legal);
       setOperationAction(ISD::FP_TO_UINT,       MVT::v8i32, Legal);
       setOperationAction(ISD::SINT_TO_FP,       MVT::v4i32, Legal);
       setOperationAction(ISD::FP_TO_SINT,       MVT::v4i32, Legal);
       setOperationAction(ISD::FP_TO_UINT,       MVT::v4i32, Legal);
       setOperationAction(ISD::ZERO_EXTEND,      MVT::v4i32, Custom);
if (!Subtarget.useSoftFloat() && Subtarget.hasAVX512()) {
     setOperationAction(ISD::SUB,                MVT::v8i1,  Custom);
     setOperationAction(ISD::SUB,                MVT::v16i1, Custom);
     setOperationAction(ISD::MUL,                MVT::v8i1,  Custom);
     setOperationAction(ISD::MUL,                MVT::v16i1, Custom);
 
     setOperationAction(ISD::MUL,                MVT::v16i32, Legal);
 
     for (auto VT : { MVT::v16i32, MVT::v8i64 }) {
+      setOperationAction(ISD::ABS, VT, Legal);
       setOperationAction(ISD::SRL, VT, Custom);
       setOperationAction(ISD::SHL, VT, Custom);
       setOperationAction(ISD::SRA, VT, Custom);
       setOperationAction(ISD::CTPOP, VT, Custom);
       setOperationAction(ISD::CTTZ, VT, Custom);
     }
 
     // Need to promote to 64-bit even though we have 32-bit masked instructions
if (!Subtarget.useSoftFloat() && Subtarget.hasBWI()) {
     if (Subtarget.hasCDI()) {
       setOperationAction(ISD::CTLZ,            MVT::v32i16, Custom);
       setOperationAction(ISD::CTLZ,            MVT::v64i8,  Custom);
     }
 
     for (auto VT : { MVT::v64i8, MVT::v32i16 }) {
       setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
       setOperationAction(ISD::VSELECT,      VT, Legal);
+      setOperationAction(ISD::ABS,          VT, Legal);
       setOperationAction(ISD::SRL,          VT, Custom);
       setOperationAction(ISD::SHL,          VT, Custom);
       setOperationAction(ISD::SRA,          VT, Custom);
       setOperationAction(ISD::MLOAD,        VT, Legal);
       setOperationAction(ISD::MSTORE,       VT, Legal);
       setOperationAction(ISD::CTPOP,        VT, Custom);
       setOperationAction(ISD::CTTZ,         VT, Custom);
 
if (VT.getScalarType() == MVT::i1)
     return DAG.getNode(ISD::XOR, SDLoc(Op), VT,
                        Op.getOperand(0), Op.getOperand(1));
   assert(Op.getSimpleValueType().is256BitVector() &&
          Op.getSimpleValueType().isInteger() &&
          "Only handle AVX 256-bit vector integer operation");
   return Lower256IntArith(Op, DAG);
 }
 
+static SDValue LowerABS(SDValue Op, SelectionDAG &DAG) {
+  assert(Op.getSimpleValueType().is256BitVector() &&
+         Op.getSimpleValueType().isInteger() &&
+         "Only handle AVX 256-bit vector integer operation");
+  MVT VT = Op.getSimpleValueType();
+  unsigned NumElems = VT.getVectorNumElements();
+
+  SDLoc dl(Op);
+  SDValue Src = Op.getOperand(0);
+  SDValue Lo = extract128BitVector(Src, 0, DAG, dl);
+  SDValue Hi = extract128BitVector(Src, NumElems / 2, DAG, dl);
+
+  MVT EltVT = VT.getVectorElementType();
+  MVT NewVT = MVT::getVectorVT(EltVT, NumElems / 2);
+  return DAG.getNode(ISD::CONCAT_VECTORS, dl, VT,
+                     DAG.getNode(Op.getOpcode(), dl, NewVT, Lo),
+                     DAG.getNode(Op.getOpcode(), dl, NewVT, Hi));
+}
+
 static SDValue LowerMINMAX(SDValue Op, SelectionDAG &DAG) {
   assert(Op.getSimpleValueType().is256BitVector() &&
          Op.getSimpleValueType().isInteger() &&
          "Only handle AVX 256-bit vector integer operation");
   return Lower256IntArith(Op, DAG);
 }
 
 static SDValue LowerMUL(SDValue Op, const X86Subtarget &Subtarget,
SDValue X86TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   case ISD::SUBC:
   case ISD::SUBE:               return LowerADDC_ADDE_SUBC_SUBE(Op, DAG);
   case ISD::ADD:
   case ISD::SUB:                return LowerADD_SUB(Op, DAG);
   case ISD::SMAX:
   case ISD::SMIN:
   case ISD::UMAX:
   case ISD::UMIN:               return LowerMINMAX(Op, DAG);
+  case ISD::ABS:                return LowerABS(Op, DAG);
   case ISD::FSINCOS:            return LowerFSINCOS(Op, Subtarget, DAG);
   case ISD::MLOAD:              return LowerMLOAD(Op, Subtarget, DAG);
   case ISD::MSTORE:             return LowerMSTORE(Op, Subtarget, DAG);
   case ISD::MGATHER:            return LowerMGATHER(Op, Subtarget, DAG);
   case ISD::MSCATTER:           return LowerMSCATTER(Op, Subtarget, DAG);
   case ISD::GC_TRANSITION_START:
                                 return LowerGC_TRANSITION_START(Op, DAG);
   case ISD::GC_TRANSITION_END:  return LowerGC_TRANSITION_END(Op, DAG);
const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::BLENDI:             return "X86ISD::BLENDI";
   case X86ISD::SHRUNKBLEND:        return "X86ISD::SHRUNKBLEND";
   case X86ISD::ADDUS:              return "X86ISD::ADDUS";
   case X86ISD::SUBUS:              return "X86ISD::SUBUS";
   case X86ISD::HADD:               return "X86ISD::HADD";
   case X86ISD::HSUB:               return "X86ISD::HSUB";
   case X86ISD::FHADD:              return "X86ISD::FHADD";
   case X86ISD::FHSUB:              return "X86ISD::FHSUB";
-  case X86ISD::ABS:                return "X86ISD::ABS";
   case X86ISD::CONFLICT:           return "X86ISD::CONFLICT";
   case X86ISD::FMAX:               return "X86ISD::FMAX";
   case X86ISD::FMAX_RND:           return "X86ISD::FMAX_RND";
   case X86ISD::FMIN:               return "X86ISD::FMIN";
   case X86ISD::FMIN_RND:           return "X86ISD::FMIN_RND";
   case X86ISD::FMAXC:              return "X86ISD::FMAXC";
   case X86ISD::FMINC:              return "X86ISD::FMINC";
   case X86ISD::FRSQRT:             return "X86ISD::FRSQRT";

lib/Target/X86/X86InstrAVX512.td

multiclass avx512_unary_rm_vl_all<bits<8> opc_b, bits<8> opc_w,
                                   bits<8> opc_d, bits<8> opc_q,
                                   string OpcodeStr, SDNode OpNode> {
   defm NAME : avx512_unary_rm_vl_dq<opc_d, opc_q, OpcodeStr, OpNode,
                                     HasAVX512>,
               avx512_unary_rm_vl_bw<opc_b, opc_w, OpcodeStr, OpNode,
                                     HasBWI>;
 }
 
-defm VPABS : avx512_unary_rm_vl_all<0x1C, 0x1D, 0x1E, 0x1F, "vpabs", X86Abs>;
-
-def avx512_v16i1sextv16i8 : PatLeaf<(v16i8 (X86pcmpgt (bc_v16i8 (v4i32 immAllZerosV)),
-                                                      VR128X:$src))>;
-def avx512_v8i1sextv8i16 : PatLeaf<(v8i16 (X86vsrai VR128X:$src, (i8 15)))>;
-def avx512_v4i1sextv4i32  : PatLeaf<(v4i32 (X86vsrai VR128X:$src, (i8 31)))>;
-def avx512_v32i1sextv32i8 : PatLeaf<(v32i8 (X86pcmpgt (bc_v32i8 (v8i32 immAllZerosV)),
-                                                      VR256X:$src))>;
-def avx512_v16i1sextv16i16: PatLeaf<(v16i16 (X86vsrai VR256X:$src, (i8 15)))>;
-def avx512_v8i1sextv8i32  : PatLeaf<(v8i32 (X86vsrai VR256X:$src, (i8 31)))>;
-
-let Predicates = [HasBWI, HasVLX] in {
-  def : Pat<(xor
-            (bc_v2i64 (avx512_v16i1sextv16i8)),
-            (bc_v2i64 (add (v16i8 VR128X:$src), (avx512_v16i1sextv16i8)))),
-            (VPABSBZ128rr VR128X:$src)>;
-  def : Pat<(xor
-            (bc_v2i64 (avx512_v8i1sextv8i16)),
-            (bc_v2i64 (add (v8i16 VR128X:$src), (avx512_v8i1sextv8i16)))),
-            (VPABSWZ128rr VR128X:$src)>;
-  def : Pat<(xor
-            (bc_v4i64 (avx512_v32i1sextv32i8)),
-            (bc_v4i64 (add (v32i8 VR256X:$src), (avx512_v32i1sextv32i8)))),
-            (VPABSBZ256rr VR256X:$src)>;
-  def : Pat<(xor
-            (bc_v4i64 (avx512_v16i1sextv16i16)),
-            (bc_v4i64 (add (v16i16 VR256X:$src), (avx512_v16i1sextv16i16)))),
-            (VPABSWZ256rr VR256X:$src)>;
-}
-let Predicates = [HasAVX512, HasVLX] in {
-  def : Pat<(xor
-            (bc_v2i64 (avx512_v4i1sextv4i32)),
-            (bc_v2i64 (add (v4i32 VR128X:$src), (avx512_v4i1sextv4i32)))),
-            (VPABSDZ128rr VR128X:$src)>;
-  def : Pat<(xor
-            (bc_v4i64 (avx512_v8i1sextv8i32)),
-            (bc_v4i64 (add (v8i32 VR256X:$src), (avx512_v8i1sextv8i32)))),
-            (VPABSDZ256rr VR256X:$src)>;
-}
-
-let Predicates = [HasAVX512] in {
-def : Pat<(xor
-          (bc_v8i64 (v16i1sextv16i32)),
-          (bc_v8i64 (add (v16i32 VR512:$src), (v16i1sextv16i32)))),
-          (VPABSDZrr VR512:$src)>;
-def : Pat<(xor
-          (bc_v8i64 (v8i1sextv8i64)),
-          (bc_v8i64 (add (v8i64 VR512:$src), (v8i1sextv8i64)))),
-          (VPABSQZrr VR512:$src)>;
-}
-let Predicates = [HasBWI] in {
-def : Pat<(xor
-          (bc_v8i64 (v64i1sextv64i8)),
-          (bc_v8i64 (add (v64i8 VR512:$src), (v64i1sextv64i8)))),
-          (VPABSBZrr VR512:$src)>;
-def : Pat<(xor
-          (bc_v8i64 (v32i1sextv32i16)),
-          (bc_v8i64 (add (v32i16 VR512:$src), (v32i1sextv32i16)))),
-          (VPABSWZrr VR512:$src)>;
-}
+defm VPABS : avx512_unary_rm_vl_all<0x1C, 0x1D, 0x1E, 0x1F, "vpabs", abs>;
 
 multiclass avx512_ctlz<bits<8> opc, string OpcodeStr, Predicate prd>{
 
   defm NAME :          avx512_unary_rm_vl_dq<opc, opc, OpcodeStr, ctlz, prd>;
 }
 
 defm VPLZCNT    : avx512_ctlz<0x44, "vplzcnt", HasCDI>;
 defm VPCONFLICT : avx512_unary_rm_vl_dq<0xC4, 0xC4, "vpconflict", X86Conflict, HasCDI>;

lib/Target/X86/X86InstrFragmentsSIMD.td

def SDTFma : SDTypeProfile<1, 3, [SDTCisSameAs<0,1>,
                            SDTCisSameAs<1,2>, SDTCisSameAs<1,3>]>;
 def SDTFmaRound : SDTypeProfile<1, 4, [SDTCisSameAs<0,1>,
                            SDTCisSameAs<1,2>, SDTCisSameAs<1,3>,
                            SDTCisVT<4, i32>]>;
 
 def X86PAlignr : SDNode<"X86ISD::PALIGNR", SDTShuff3OpI>;
 def X86VAlign  : SDNode<"X86ISD::VALIGN", SDTShuff3OpI>;
 
-def X86Abs      : SDNode<"X86ISD::ABS", SDTIntUnaryOp>;
 def X86Conflict : SDNode<"X86ISD::CONFLICT", SDTIntUnaryOp>;
 
 def X86PShufd  : SDNode<"X86ISD::PSHUFD", SDTShuff2OpI>;
 def X86PShufhw : SDNode<"X86ISD::PSHUFHW", SDTShuff2OpI>;
 def X86PShuflw : SDNode<"X86ISD::PSHUFLW", SDTShuff2OpI>;
 
 def X86Shufp   : SDNode<"X86ISD::SHUFP", SDTShuff3OpI>;
 def X86Shuf128 : SDNode<"X86ISD::SHUF128", SDTShuff3OpI>;

lib/Target/X86/X86InstrSSE.td

multiclass SS3I_unop_rm_y<bits<8> opc, string OpcodeStr, ValueType vt,
   def Yrm : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
                   (ins i256mem:$src),
                   !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
                   [(set VR256:$dst,
                     (vt (OpNode (bitconvert (loadv4i64 addr:$src)))))]>,
                   Sched<[WriteVecALULd]>;
 }
 
-// Helper fragments to match sext vXi1 to vXiY.
-def v16i1sextv16i8 : PatLeaf<(v16i8 (X86pcmpgt (bc_v16i8 (v4i32 immAllZerosV)),
-                                               VR128:$src))>;
-def v8i1sextv8i16  : PatLeaf<(v8i16 (X86vsrai VR128:$src, (i8 15)))>;
-def v4i1sextv4i32  : PatLeaf<(v4i32 (X86vsrai VR128:$src, (i8 31)))>;
-def v32i1sextv32i8 : PatLeaf<(v32i8 (X86pcmpgt (bc_v32i8 (v8i32 immAllZerosV)),
-                                               VR256:$src))>;
-def v16i1sextv16i16: PatLeaf<(v16i16 (X86vsrai VR256:$src, (i8 15)))>;
-def v8i1sextv8i32  : PatLeaf<(v8i32 (X86vsrai VR256:$src, (i8 31)))>;
-
 let Predicates = [HasAVX, NoVLX_Or_NoBWI] in {
-  defm VPABSB  : SS3I_unop_rm<0x1C, "vpabsb", v16i8, X86Abs, loadv2i64>, VEX;
-  defm VPABSW  : SS3I_unop_rm<0x1D, "vpabsw", v8i16, X86Abs, loadv2i64>, VEX;
+  defm VPABSB  : SS3I_unop_rm<0x1C, "vpabsb", v16i8, abs, loadv2i64>, VEX;
+  defm VPABSW  : SS3I_unop_rm<0x1D, "vpabsw", v8i16, abs, loadv2i64>, VEX;
 }
 let Predicates = [HasAVX, NoVLX] in {
-  defm VPABSD  : SS3I_unop_rm<0x1E, "vpabsd", v4i32, X86Abs, loadv2i64>, VEX;
-}
-
-let Predicates = [HasAVX, NoVLX_Or_NoBWI] in {
-  def : Pat<(xor
-            (bc_v2i64 (v16i1sextv16i8)),
-            (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
-            (VPABSBrr VR128:$src)>;
-  def : Pat<(xor
-            (bc_v2i64 (v8i1sextv8i16)),
-            (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
-            (VPABSWrr VR128:$src)>;
-}
-let Predicates = [HasAVX, NoVLX] in {
-  def : Pat<(xor
-            (bc_v2i64 (v4i1sextv4i32)),
-            (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
-            (VPABSDrr VR128:$src)>;
-}
-
-let Predicates = [HasAVX2, NoVLX_Or_NoBWI] in {
-  defm VPABSB  : SS3I_unop_rm_y<0x1C, "vpabsb", v32i8, X86Abs>, VEX, VEX_L;
-  defm VPABSW  : SS3I_unop_rm_y<0x1D, "vpabsw", v16i16, X86Abs>, VEX, VEX_L;
-}
-let Predicates = [HasAVX2, NoVLX] in {
-  defm VPABSD  : SS3I_unop_rm_y<0x1E, "vpabsd", v8i32, X86Abs>, VEX, VEX_L;
+  defm VPABSD  : SS3I_unop_rm<0x1E, "vpabsd", v4i32, abs, loadv2i64>, VEX;
 }
 
 let Predicates = [HasAVX2, NoVLX_Or_NoBWI] in {
-  def : Pat<(xor
-            (bc_v4i64 (v32i1sextv32i8)),
-            (bc_v4i64 (add (v32i8 VR256:$src), (v32i1sextv32i8)))),
-            (VPABSBYrr VR256:$src)>;
-  def : Pat<(xor
-            (bc_v4i64 (v16i1sextv16i16)),
-            (bc_v4i64 (add (v16i16 VR256:$src), (v16i1sextv16i16)))),
-            (VPABSWYrr VR256:$src)>;
+  defm VPABSB  : SS3I_unop_rm_y<0x1C, "vpabsb", v32i8, abs>, VEX, VEX_L;
+  defm VPABSW  : SS3I_unop_rm_y<0x1D, "vpabsw", v16i16, abs>, VEX, VEX_L;
 }
 let Predicates = [HasAVX2, NoVLX] in {
-  def : Pat<(xor
-            (bc_v4i64 (v8i1sextv8i32)),
-            (bc_v4i64 (add (v8i32 VR256:$src), (v8i1sextv8i32)))),
-            (VPABSDYrr VR256:$src)>;
+  defm VPABSD  : SS3I_unop_rm_y<0x1E, "vpabsd", v8i32, abs>, VEX, VEX_L;
 }
 
-defm PABSB : SS3I_unop_rm<0x1C, "pabsb", v16i8, X86Abs, memopv2i64>;
-defm PABSW : SS3I_unop_rm<0x1D, "pabsw", v8i16, X86Abs, memopv2i64>;
-defm PABSD : SS3I_unop_rm<0x1E, "pabsd", v4i32, X86Abs, memopv2i64>;
-
-let Predicates = [UseSSSE3] in {
-  def : Pat<(xor
-            (bc_v2i64 (v16i1sextv16i8)),
-            (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
-            (PABSBrr VR128:$src)>;
-  def : Pat<(xor
-            (bc_v2i64 (v8i1sextv8i16)),
-            (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
-            (PABSWrr VR128:$src)>;
-  def : Pat<(xor
-            (bc_v2i64 (v4i1sextv4i32)),
-            (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
-            (PABSDrr VR128:$src)>;
-}
+defm PABSB : SS3I_unop_rm<0x1C, "pabsb", v16i8, abs, memopv2i64>;
+defm PABSW : SS3I_unop_rm<0x1D, "pabsw", v8i16, abs, memopv2i64>;
+defm PABSD : SS3I_unop_rm<0x1E, "pabsd", v4i32, abs, memopv2i64>;
 
 //===---------------------------------------------------------------------===//
 // SSSE3 - Packed Binary Operator Instructions
 //===---------------------------------------------------------------------===//
 
 let Sched = WriteVecALU in {
 def SSE_PHADDSUBD : OpndItins<
   IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM

lib/Target/X86/X86IntrinsicsInfo.h

static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx_sqrt_ps_256,   INTR_TYPE_1OP, ISD::FSQRT, 0),
   X86_INTRINSIC_DATA(avx_vperm2f128_pd_256, INTR_TYPE_3OP, X86ISD::VPERM2X128, 0),
   X86_INTRINSIC_DATA(avx_vperm2f128_ps_256, INTR_TYPE_3OP, X86ISD::VPERM2X128, 0),
   X86_INTRINSIC_DATA(avx_vperm2f128_si_256, INTR_TYPE_3OP, X86ISD::VPERM2X128, 0),
   X86_INTRINSIC_DATA(avx_vpermilvar_pd,     INTR_TYPE_2OP, X86ISD::VPERMILPV, 0),
   X86_INTRINSIC_DATA(avx_vpermilvar_pd_256, INTR_TYPE_2OP, X86ISD::VPERMILPV, 0),
   X86_INTRINSIC_DATA(avx_vpermilvar_ps,     INTR_TYPE_2OP, X86ISD::VPERMILPV, 0),
   X86_INTRINSIC_DATA(avx_vpermilvar_ps_256, INTR_TYPE_2OP, X86ISD::VPERMILPV, 0),
-  X86_INTRINSIC_DATA(avx2_pabs_b, INTR_TYPE_1OP, X86ISD::ABS, 0),
-  X86_INTRINSIC_DATA(avx2_pabs_d, INTR_TYPE_1OP, X86ISD::ABS, 0),
-  X86_INTRINSIC_DATA(avx2_pabs_w, INTR_TYPE_1OP, X86ISD::ABS, 0),
+  X86_INTRINSIC_DATA(avx2_pabs_b, INTR_TYPE_1OP, ISD::ABS, 0),
+  X86_INTRINSIC_DATA(avx2_pabs_d, INTR_TYPE_1OP, ISD::ABS, 0),
+  X86_INTRINSIC_DATA(avx2_pabs_w, INTR_TYPE_1OP, ISD::ABS, 0),
   X86_INTRINSIC_DATA(avx2_packssdw, INTR_TYPE_2OP, X86ISD::PACKSS, 0),
   X86_INTRINSIC_DATA(avx2_packsswb, INTR_TYPE_2OP, X86ISD::PACKSS, 0),
   X86_INTRINSIC_DATA(avx2_packusdw, INTR_TYPE_2OP, X86ISD::PACKUS, 0),
   X86_INTRINSIC_DATA(avx2_packuswb, INTR_TYPE_2OP, X86ISD::PACKUS, 0),
   X86_INTRINSIC_DATA(avx2_padds_b, INTR_TYPE_2OP, X86ISD::ADDS, 0),
   X86_INTRINSIC_DATA(avx2_padds_w, INTR_TYPE_2OP, X86ISD::ADDS, 0),
   X86_INTRINSIC_DATA(avx2_paddus_b, INTR_TYPE_2OP, X86ISD::ADDUS, 0),
   X86_INTRINSIC_DATA(avx2_paddus_w, INTR_TYPE_2OP, X86ISD::ADDUS, 0),
static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx512_mask_mul_pd_512, INTR_TYPE_2OP_MASK, ISD::FMUL,
                      X86ISD::FMUL_RND),
   X86_INTRINSIC_DATA(avx512_mask_mul_ps_512, INTR_TYPE_2OP_MASK, ISD::FMUL,
                      X86ISD::FMUL_RND),
   X86_INTRINSIC_DATA(avx512_mask_mul_sd_round, INTR_TYPE_SCALAR_MASK_RM,
                      X86ISD::FMUL_RND, 0),
   X86_INTRINSIC_DATA(avx512_mask_mul_ss_round, INTR_TYPE_SCALAR_MASK_RM,
                      X86ISD::FMUL_RND, 0),
-  X86_INTRINSIC_DATA(avx512_mask_pabs_b_128, INTR_TYPE_1OP_MASK, X86ISD::ABS, 0),
-  X86_INTRINSIC_DATA(avx512_mask_pabs_b_256, INTR_TYPE_1OP_MASK, X86ISD::ABS, 0),
-  X86_INTRINSIC_DATA(avx512_mask_pabs_b_512, INTR_TYPE_1OP_MASK, X86ISD::ABS, 0),
-  X86_INTRINSIC_DATA(avx512_mask_pabs_d_128, INTR_TYPE_1OP_MASK, X86ISD::ABS, 0),
-  X86_INTRINSIC_DATA(avx512_mask_pabs_d_256, INTR_TYPE_1OP_MASK, X86ISD::ABS, 0),
-  X86_INTRINSIC_DATA(avx512_mask_pabs_d_512, INTR_TYPE_1OP_MASK, X86ISD::ABS, 0),
-  X86_INTRINSIC_DATA(avx512_mask_pabs_q_128, INTR_TYPE_1OP_MASK, X86ISD::ABS, 0),
-  X86_INTRINSIC_DATA(avx512_mask_pabs_q_256, INTR_TYPE_1OP_MASK, X86ISD::ABS, 0),
-  X86_INTRINSIC_DATA(avx512_mask_pabs_q_512, INTR_TYPE_1OP_MASK, X86ISD::ABS, 0),
-  X86_INTRINSIC_DATA(avx512_mask_pabs_w_128, INTR_TYPE_1OP_MASK, X86ISD::ABS, 0),
-  X86_INTRINSIC_DATA(avx512_mask_pabs_w_256, INTR_TYPE_1OP_MASK, X86ISD::ABS, 0),
-  X86_INTRINSIC_DATA(avx512_mask_pabs_w_512, INTR_TYPE_1OP_MASK, X86ISD::ABS, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pabs_b_128, INTR_TYPE_1OP_MASK, ISD::ABS, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pabs_b_256, INTR_TYPE_1OP_MASK, ISD::ABS, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pabs_b_512, INTR_TYPE_1OP_MASK, ISD::ABS, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pabs_d_128, INTR_TYPE_1OP_MASK, ISD::ABS, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pabs_d_256, INTR_TYPE_1OP_MASK, ISD::ABS, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pabs_d_512, INTR_TYPE_1OP_MASK, ISD::ABS, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pabs_q_128, INTR_TYPE_1OP_MASK, ISD::ABS, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pabs_q_256, INTR_TYPE_1OP_MASK, ISD::ABS, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pabs_q_512, INTR_TYPE_1OP_MASK, ISD::ABS, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pabs_w_128, INTR_TYPE_1OP_MASK, ISD::ABS, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pabs_w_256, INTR_TYPE_1OP_MASK, ISD::ABS, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pabs_w_512, INTR_TYPE_1OP_MASK, ISD::ABS, 0),
   X86_INTRINSIC_DATA(avx512_mask_packssdw_128, INTR_TYPE_2OP_MASK, X86ISD::PACKSS, 0),
   X86_INTRINSIC_DATA(avx512_mask_packssdw_256, INTR_TYPE_2OP_MASK, X86ISD::PACKSS, 0),
   X86_INTRINSIC_DATA(avx512_mask_packssdw_512, INTR_TYPE_2OP_MASK, X86ISD::PACKSS, 0),
   X86_INTRINSIC_DATA(avx512_mask_packsswb_128, INTR_TYPE_2OP_MASK, X86ISD::PACKSS, 0),
   X86_INTRINSIC_DATA(avx512_mask_packsswb_256, INTR_TYPE_2OP_MASK, X86ISD::PACKSS, 0),
   X86_INTRINSIC_DATA(avx512_mask_packsswb_512, INTR_TYPE_2OP_MASK, X86ISD::PACKSS, 0),
   X86_INTRINSIC_DATA(avx512_mask_packusdw_128, INTR_TYPE_2OP_MASK, X86ISD::PACKUS, 0),
   X86_INTRINSIC_DATA(avx512_mask_packusdw_256, INTR_TYPE_2OP_MASK, X86ISD::PACKUS, 0),

   X86_INTRINSIC_DATA(sse3_hadd_ps,      INTR_TYPE_2OP, X86ISD::FHADD, 0),
   X86_INTRINSIC_DATA(sse3_hsub_pd,      INTR_TYPE_2OP, X86ISD::FHSUB, 0),
   X86_INTRINSIC_DATA(sse3_hsub_ps,      INTR_TYPE_2OP, X86ISD::FHSUB, 0),
   X86_INTRINSIC_DATA(sse41_insertps,    INTR_TYPE_3OP, X86ISD::INSERTPS, 0),
   X86_INTRINSIC_DATA(sse41_packusdw,    INTR_TYPE_2OP, X86ISD::PACKUS, 0),
   X86_INTRINSIC_DATA(sse41_pmuldq,      INTR_TYPE_2OP, X86ISD::PMULDQ, 0),
   X86_INTRINSIC_DATA(sse4a_extrqi,      INTR_TYPE_3OP, X86ISD::EXTRQI, 0),
   X86_INTRINSIC_DATA(sse4a_insertqi,    INTR_TYPE_4OP, X86ISD::INSERTQI, 0),
-  X86_INTRINSIC_DATA(ssse3_pabs_b_128,  INTR_TYPE_1OP, X86ISD::ABS, 0),
-  X86_INTRINSIC_DATA(ssse3_pabs_d_128,  INTR_TYPE_1OP, X86ISD::ABS, 0),
-  X86_INTRINSIC_DATA(ssse3_pabs_w_128,  INTR_TYPE_1OP, X86ISD::ABS, 0),
+  X86_INTRINSIC_DATA(ssse3_pabs_b_128,  INTR_TYPE_1OP, ISD::ABS, 0),
+  X86_INTRINSIC_DATA(ssse3_pabs_d_128,  INTR_TYPE_1OP, ISD::ABS, 0),
+  X86_INTRINSIC_DATA(ssse3_pabs_w_128,  INTR_TYPE_1OP, ISD::ABS, 0),
   X86_INTRINSIC_DATA(ssse3_phadd_d_128, INTR_TYPE_2OP, X86ISD::HADD, 0),
   X86_INTRINSIC_DATA(ssse3_phadd_w_128, INTR_TYPE_2OP, X86ISD::HADD, 0),
   X86_INTRINSIC_DATA(ssse3_phsub_d_128, INTR_TYPE_2OP, X86ISD::HSUB, 0),
   X86_INTRINSIC_DATA(ssse3_phsub_w_128, INTR_TYPE_2OP, X86ISD::HSUB, 0),
   X86_INTRINSIC_DATA(ssse3_pmadd_ub_sw_128, INTR_TYPE_2OP, X86ISD::VPMADDUBSW, 0),
   X86_INTRINSIC_DATA(ssse3_pmul_hr_sw_128, INTR_TYPE_2OP, X86ISD::MULHRS, 0),
   X86_INTRINSIC_DATA(ssse3_pshuf_b_128, INTR_TYPE_2OP, X86ISD::PSHUFB, 0),
   X86_INTRINSIC_DATA(xop_vpcomb,        INTR_TYPE_3OP, X86ISD::VPCOM, 0),

test/CodeGen/X86/combine-abs.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s
 
-; FIXME: Various missed opportunities to simplify integer absolute instructions.
-
 ; fold (abs c1) -> c2
 define <4 x i32> @combine_v4i32_abs_constant() {
 ; CHECK-LABEL: combine_v4i32_abs_constant:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    vpabsd {{.*}}(%rip), %xmm0
+; CHECK-NEXT:    vmovaps {{.*#+}} xmm0 = [0,1,3,2147483648]
 ; CHECK-NEXT:    retq
   %1 = call <4 x i32> @llvm.x86.ssse3.pabs.d.128(<4 x i32> <i32 0, i32 -1, i32 3, i32 -2147483648>)
   ret <4 x i32> %1
 }
 
 define <16 x i16> @combine_v16i16_abs_constant() {
 ; CHECK-LABEL: combine_v16i16_abs_constant:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    vpabsw {{.*}}(%rip), %ymm0
+; CHECK-NEXT:    vmovaps {{.*#+}} ymm0 = [0,1,1,3,3,7,7,255,255,4096,4096,32767,32767,32768,32768,0]
 ; CHECK-NEXT:    retq
   %1 = call <16 x i16> @llvm.x86.avx2.pabs.w(<16 x i16> <i16 0, i16 1, i16 -1, i16 3, i16 -3, i16 7, i16 -7, i16 255, i16 -255, i16 4096, i16 -4096, i16 32767, i16 -32767, i16 -32768, i16 32768, i16 65536>)
   ret <16 x i16> %1
 }
 
 ; fold (abs (abs x)) -> (abs x)
 define <8 x i16> @combine_v8i16_abs_abs(<8 x i16> %a) {
 ; CHECK-LABEL: combine_v8i16_abs_abs:
 ; CHECK:       # BB#0:
 ; CHECK-NEXT:    vpabsw %xmm0, %xmm0
-; CHECK-NEXT:    vpabsw %xmm0, %xmm0
 ; CHECK-NEXT:    retq
   %a1 = call <8 x i16> @llvm.x86.ssse3.pabs.w.128(<8 x i16> %a)
   %n2 = sub <8 x i16> zeroinitializer, %a1
   %c2 = icmp slt <8 x i16> %a1, zeroinitializer
   %a2 = select <8 x i1> %c2, <8 x i16> %n2, <8 x i16> %a1
   ret <8 x i16> %a2
 }
 
 define <32 x i8> @combine_v32i8_abs_abs(<32 x i8> %a) {
 ; CHECK-LABEL: combine_v32i8_abs_abs:
 ; CHECK:       # BB#0:
 ; CHECK-NEXT:    vpabsb %ymm0, %ymm0
-; CHECK-NEXT:    vpabsb %ymm0, %ymm0
 ; CHECK-NEXT:    retq
   %n1 = sub <32 x i8> zeroinitializer, %a
   %b1 = icmp slt <32 x i8> %a, zeroinitializer
   %a1 = select <32 x i1> %b1, <32 x i8> %n1, <32 x i8> %a
   %a2 = call <32 x i8> @llvm.x86.avx2.pabs.b(<32 x i8> %a1)
   ret <32 x i8> %a2
 }
 

test/CodeGen/X86/viabs.ll

 ; SSSE3-LABEL: test_abs_gt_v8i32:
 ; SSSE3:       # BB#0:
 ; SSSE3-NEXT:    pabsd %xmm0, %xmm0
 ; SSSE3-NEXT:    pabsd %xmm1, %xmm1
 ; SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: test_abs_gt_v8i32:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vpsrad $31, %xmm1, %xmm2
-; AVX1-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrad $31, %xmm0, %xmm3
-; AVX1-NEXT:    vpaddd %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm1
-; AVX1-NEXT:    vxorps %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vpabsd %xmm0, %xmm1
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vpabsd %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_abs_gt_v8i32:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpabsd %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_abs_gt_v8i32:

 ; SSSE3-LABEL: test_abs_ge_v8i32:
 ; SSSE3:       # BB#0:
 ; SSSE3-NEXT:    pabsd %xmm0, %xmm0
 ; SSSE3-NEXT:    pabsd %xmm1, %xmm1
 ; SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: test_abs_ge_v8i32:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vpsrad $31, %xmm1, %xmm2
-; AVX1-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrad $31, %xmm0, %xmm3
-; AVX1-NEXT:    vpaddd %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm1
-; AVX1-NEXT:    vxorps %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vpabsd %xmm0, %xmm1
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vpabsd %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_abs_ge_v8i32:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpabsd %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_abs_ge_v8i32:

 ; SSSE3-LABEL: test_abs_gt_v16i16:
 ; SSSE3:       # BB#0:
 ; SSSE3-NEXT:    pabsw %xmm0, %xmm0
 ; SSSE3-NEXT:    pabsw %xmm1, %xmm1
 ; SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: test_abs_gt_v16i16:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vpsraw $15, %xmm1, %xmm2
-; AVX1-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpsraw $15, %xmm0, %xmm3
-; AVX1-NEXT:    vpaddw %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm1
-; AVX1-NEXT:    vxorps %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vpabsw %xmm0, %xmm1
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vpabsw %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_abs_gt_v16i16:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpabsw %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_abs_gt_v16i16:

 ; SSSE3-LABEL: test_abs_lt_v32i8:
 ; SSSE3:       # BB#0:
 ; SSSE3-NEXT:    pabsb %xmm0, %xmm0
 ; SSSE3-NEXT:    pabsb %xmm1, %xmm1
 ; SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: test_abs_lt_v32i8:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT:    vpcmpgtb %xmm1, %xmm2, %xmm3
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm4
-; AVX1-NEXT:    vpaddb %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX1-NEXT:    vxorps %ymm4, %ymm0, %ymm0
+; AVX1-NEXT:    vpabsb %xmm0, %xmm1
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vpabsb %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_abs_lt_v32i8:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpabsb %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_abs_lt_v32i8:

 ; SSSE3-LABEL: test_abs_le_v8i32:
 ; SSSE3:       # BB#0:
 ; SSSE3-NEXT:    pabsd %xmm0, %xmm0
 ; SSSE3-NEXT:    pabsd %xmm1, %xmm1
 ; SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: test_abs_le_v8i32:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vpsrad $31, %xmm1, %xmm2
-; AVX1-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrad $31, %xmm0, %xmm3
-; AVX1-NEXT:    vpaddd %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm1
-; AVX1-NEXT:    vxorps %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vpabsd %xmm0, %xmm1
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vpabsd %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_abs_le_v8i32:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpabsd %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_abs_le_v8i32:

 ; SSSE3-NEXT:    pabsd %xmm0, %xmm0
 ; SSSE3-NEXT:    pabsd %xmm1, %xmm1
 ; SSSE3-NEXT:    pabsd %xmm2, %xmm2
 ; SSSE3-NEXT:    pabsd %xmm3, %xmm3
 ; SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: test_abs_le_16i32:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpsrad $31, %xmm2, %xmm3
-; AVX1-NEXT:    vpaddd %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpsrad $31, %xmm0, %xmm4
-; AVX1-NEXT:    vpaddd %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm2
-; AVX1-NEXT:    vxorps %ymm2, %ymm0, %ymm0
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vpsrad $31, %xmm2, %xmm3
-; AVX1-NEXT:    vpaddd %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpsrad $31, %xmm1, %xmm4
-; AVX1-NEXT:    vpaddd %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm2
-; AVX1-NEXT:    vxorps %ymm2, %ymm1, %ymm1
+; AVX1-NEXT:    vpabsd %xmm0, %xmm2
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vpabsd %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
+; AVX1-NEXT:    vpabsd %xmm1, %xmm2
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vpabsd %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_abs_le_16i32:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpabsd %ymm0, %ymm0
 ; AVX2-NEXT:    vpabsd %ymm1, %ymm1
 ; AVX2-NEXT:    retq
 ;

 ; AVX2-NEXT:    vpsrad $31, %xmm0, %xmm1
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; AVX2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_abs_ge_v2i64:
 ; AVX512:       # BB#0:
-; AVX512-NEXT:    vpsraq $63, %xmm0, %xmm1
-; AVX512-NEXT:    vpaddq %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpabsq %xmm0, %xmm0
 ; AVX512-NEXT:    retq
   %tmp1neg = sub <2 x i64> zeroinitializer, %a
   %b = icmp sge <2 x i64> %a, zeroinitializer
   %abs = select <2 x i1> %b, <2 x i64> %a, <2 x i64> %tmp1neg
   ret <2 x i64> %abs
 }
 
 define <4 x i64> @test_abs_gt_v4i64(<4 x i64> %a) nounwind {

 ; AVX2-NEXT:    vpsrad $31, %ymm0, %ymm1
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,1,3,3,5,5,7,7]
 ; AVX2-NEXT:    vpaddq %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_abs_gt_v4i64:
 ; AVX512:       # BB#0:
-; AVX512-NEXT:    vpsraq $63, %ymm0, %ymm1
-; AVX512-NEXT:    vpaddq %ymm1, %ymm0, %ymm0
-; AVX512-NEXT:    vpxor %ymm1, %ymm0, %ymm0
+; AVX512-NEXT:    vpabsq %ymm0, %ymm0
 ; AVX512-NEXT:    retq
   %tmp1neg = sub <4 x i64> zeroinitializer, %a
   %b = icmp sgt <4 x i64> %a, <i64 -1, i64 -1, i64 -1, i64 -1>
   %abs = select <4 x i1> %b, <4 x i64> %a, <4 x i64> %tmp1neg
   ret <4 x i64> %abs
 }
 
 define <8 x i64> @test_abs_le_v8i64(<8 x i64> %a) nounwind {

 ; SSSE3-NEXT:    pabsb %xmm0, %xmm0
 ; SSSE3-NEXT:    pabsb %xmm1, %xmm1
 ; SSSE3-NEXT:    pabsb %xmm2, %xmm2
 ; SSSE3-NEXT:    pabsb %xmm3, %xmm3
 ; SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: test_abs_lt_v64i8:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpgtb %xmm2, %xmm3, %xmm4
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm3, %xmm5
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm5, %ymm6
-; AVX1-NEXT:    vpaddb %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vpaddb %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX1-NEXT:    vxorps %ymm6, %ymm0, %ymm0
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vpcmpgtb %xmm2, %xmm3, %xmm4
-; AVX1-NEXT:    vpcmpgtb %xmm1, %xmm3, %xmm3
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm3, %ymm5
-; AVX1-NEXT:    vpaddb %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vpaddb %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX1-NEXT:    vxorps %ymm5, %ymm1, %ymm1
+; AVX1-NEXT:    vpabsb %xmm0, %xmm2
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vpabsb %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
+; AVX1-NEXT:    vpabsb %xmm1, %xmm2
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vpabsb %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_abs_lt_v64i8:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpabsb %ymm0, %ymm0
 ; AVX2-NEXT:    vpabsb %ymm1, %ymm1
 ; AVX2-NEXT:    retq
 ;

 ; SSSE3-NEXT:    pabsw %xmm0, %xmm0
 ; SSSE3-NEXT:    pabsw %xmm1, %xmm1
 ; SSSE3-NEXT:    pabsw %xmm2, %xmm2
 ; SSSE3-NEXT:    pabsw %xmm3, %xmm3
 ; SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: test_abs_gt_v32i16:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpsraw $15, %xmm2, %xmm3
-; AVX1-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpsraw $15, %xmm0, %xmm4
-; AVX1-NEXT:    vpaddw %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm2
-; AVX1-NEXT:    vxorps %ymm2, %ymm0, %ymm0
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vpsraw $15, %xmm2, %xmm3
-; AVX1-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpsraw $15, %xmm1, %xmm4
-; AVX1-NEXT:    vpaddw %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm2
-; AVX1-NEXT:    vxorps %ymm2, %ymm1, %ymm1
+; AVX1-NEXT:    vpabsw %xmm0, %xmm2
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vpabsw %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
+; AVX1-NEXT:    vpabsw %xmm1, %xmm2
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vpabsw %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_abs_gt_v32i16:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpabsw %ymm0, %ymm0
 ; AVX2-NEXT:    vpabsw %ymm1, %ymm1
 ; AVX2-NEXT:    retq
 ;