[X86] Add STRICT versions of CVTTP2SI, CVTTP2UI, CMPM, and CMPP.

[craig.topper]

Summary

As suggested in D71841

[craig.topper]

llvm/lib/Target/X86/X86InstrAVX512.td

2532

NOTE: This line with X86cmpm_su is for masked compares that match an 'and' node. But that doesn't properly handle exceptions for strict FP so we'll need a strict_masked_cmpm node for that.

[pengfei]

llvm/lib/Target/X86/X86ISelDAGToDAG.cpp

546–549

This can be removed. Do we need handle strict cases?

[craig.topper]

Address review comment

[pengfei]

llvm/lib/Target/X86/X86ISelDAGToDAG.cpp

542

How about STRICT_FSETCC/STRICT_FSETCCS?

[craig.topper]

llvm/lib/Target/X86/X86ISelDAGToDAG.cpp

542

STRICT_FSETCC/STRICT_FSETCC don't make it to isel. They become CMPP/CMPM during lowering. ISD::SETCC only makes it here with integer vector argument type.

[pengfei]

Thanks. LGTM.

llvm/lib/Target/X86/X86ISelDAGToDAG.cpp

 
 
 // Returns true if this masked compare can be implemented legally with this
 // type.
 static bool isLegalMaskCompare(SDNode *N, const X86Subtarget *Subtarget) {
   unsigned Opcode = N->getOpcode();
   if (Opcode == X86ISD::CMPM || Opcode == ISD::SETCC ||
       Opcode == X86ISD::CMPM_SAE || Opcode == X86ISD::VFPCLASS) {
     // We can get 256-bit 8 element types here without VLX being enabled. When
     // this happens we will use 512-bit operations and the mask will not be
     // zero extended.
     EVT OpVT = N->getOperand(0).getValueType();
     // The first operand of X86ISD::CMPM is chain, so we need to get the second
     // operand.
     if (Opcode == X86ISD::CMPM)
       OpVT = N->getOperand(1).getValueType();
     if (OpVT.is256BitVector() || OpVT.is128BitVector())
       return Subtarget->hasVLX();
 
     return true;
   }
   // Scalar opcodes use 128 bit registers, but aren't subject to the VLX check.
   if (Opcode == X86ISD::VFPCLASSS || Opcode == X86ISD::FSETCCM ||
       Opcode == X86ISD::FSETCCM_SAE)
case ISD::STRICT_FP_TO_UINT: {
       // Replace vector fp_to_s/uint with their X86 specific equivalent so we
       // don't need 2 sets of patterns.
       if (!N->getSimpleValueType(0).isVector())
         break;
 
       unsigned NewOpc;
       switch (N->getOpcode()) {
       default: llvm_unreachable("Unexpected opcode!");
-      case ISD::STRICT_FP_TO_SINT:
-      case ISD::FP_TO_SINT: NewOpc = X86ISD::CVTTP2SI; break;
-      case ISD::STRICT_FP_TO_UINT:
-      case ISD::FP_TO_UINT: NewOpc = X86ISD::CVTTP2UI; break;
+      case ISD::STRICT_FP_TO_SINT: NewOpc = X86ISD::STRICT_CVTTP2SI; break;
+      case ISD::FP_TO_SINT:        NewOpc = X86ISD::CVTTP2SI;        break;
+      case ISD::STRICT_FP_TO_UINT: NewOpc = X86ISD::STRICT_CVTTP2UI; break;
+      case ISD::FP_TO_UINT:        NewOpc = X86ISD::CVTTP2UI;        break;
       }
       SDValue Res;
       if (N->isStrictFPOpcode())
         Res =
             CurDAG->getNode(NewOpc, SDLoc(N), {N->getValueType(0), MVT::Other},
                             {N->getOperand(0), N->getOperand(1)});
       else
         Res =
-            CurDAG->getNode(NewOpc, SDLoc(N), {N->getValueType(0), MVT::Other},
-                            {CurDAG->getEntryNode(), N->getOperand(0)});
+            CurDAG->getNode(NewOpc, SDLoc(N), N->getValueType(0),
+                            N->getOperand(0));
       --I;
       if (N->isStrictFPOpcode()) {
         SDValue From[] = {SDValue(N, 0), SDValue(N, 1)};
         SDValue To[] = {Res.getValue(0), Res.getValue(1)};
         CurDAG->ReplaceAllUsesOfValuesWith(From, To, 2);
       } else
         CurDAG->ReplaceAllUsesOfValueWith(SDValue(N, 0), Res);
       ++I;

llvm/lib/Target/X86/X86ISelLowering.h

enum NodeType : unsigned {
       // Shifts of mask registers.
       KSHIFTL, KSHIFTR,
 
       // Bit rotate by immediate
       VROTLI, VROTRI,
 
       // Vector packed double/float comparison.
       CMPP,
+      STRICT_CMPP,
 
       // Vector integer comparisons.
       PCMPEQ, PCMPGT,
 
       // v8i16 Horizontal minimum and position.
       PHMINPOS,
 
       MULTISHIFT,
 
       /// Vector comparison generating mask bits for fp and
       /// integer signed and unsigned data types.
       CMPM,
+      STRICT_CMPM,
       // Vector comparison with SAE for FP values
       CMPM_SAE,
 
       // Arithmetic operations with FLAGS results.
       ADD, SUB, ADC, SBB, SMUL, UMUL,
       OR, XOR, AND,
 
       // Bit field extract.
enum NodeType : unsigned {
 
       // Vector float/double to signed/unsigned integer.
       CVTP2SI, CVTP2UI, CVTP2SI_RND, CVTP2UI_RND,
       // Scalar float/double to signed/unsigned integer.
       CVTS2SI, CVTS2UI, CVTS2SI_RND, CVTS2UI_RND,
 
       // Vector float/double to signed/unsigned integer with truncation.
       CVTTP2SI, CVTTP2UI, CVTTP2SI_SAE, CVTTP2UI_SAE,
+      STRICT_CVTTP2SI, STRICT_CVTTP2UI,
       // Scalar float/double to signed/unsigned integer with truncation.
       CVTTS2SI, CVTTS2UI, CVTTS2SI_SAE, CVTTS2UI_SAE,
 
       // Vector signed/unsigned integer to float/double.
       CVTSI2P, CVTUI2P,
 
       // Masked versions of above. Used for v2f64->v4f32.
       // SRC, PASSTHRU, MASK

llvm/lib/Target/X86/X86ISelLowering.cpp

if (VT == MVT::v2i1 && SrcVT == MVT::v2f64) {
         Res = DAG.getNode(ISD::TRUNCATE, dl, TruncVT, Res);
         Res = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v2i1, Res,
                           DAG.getIntPtrConstant(0, dl));
         if (IsStrict)
           return DAG.getMergeValues({Res, Chain}, dl);
         return Res;
       }
       SDValue Res, Chain;
-      unsigned Opc = IsSigned ? X86ISD::CVTTP2SI : X86ISD::CVTTP2UI;
       if (IsStrict) {
+        unsigned Opc = IsSigned ? X86ISD::STRICT_CVTTP2SI
+                                : X86ISD::STRICT_CVTTP2UI;
         Res =
             DAG.getNode(Opc, dl, {ResVT, MVT::Other}, {Op->getOperand(0), Src});
         Chain = Res.getValue(1);
-      } else
-        Res = DAG.getNode(Opc, dl, {ResVT, MVT::Other},
-                          {DAG.getEntryNode(), Src});
+      } else {
+        unsigned Opc = IsSigned ? X86ISD::CVTTP2SI : X86ISD::CVTTP2UI;
+        Res = DAG.getNode(Opc, dl, ResVT, Src);
+      }
       Res = DAG.getNode(ISD::TRUNCATE, dl, TruncVT, Res);
       Res = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v2i1, Res,
                         DAG.getIntPtrConstant(0, dl));
       if (IsStrict)
         return DAG.getMergeValues({Res, Chain}, dl);
       return Res;
     }
 
     assert(Subtarget.hasDQI() && Subtarget.hasVLX() && "Requires AVX512DQVL!");
     if (VT == MVT::v2i64 && SrcVT  == MVT::v2f32) {
       SDValue Tmp = DAG.getNode(ISD::CONCAT_VECTORS, dl, MVT::v4f32, Src,
                                 DAG.getUNDEF(MVT::v2f32));
-      unsigned Opc = IsSigned ? X86ISD::CVTTP2SI : X86ISD::CVTTP2UI;
-      SDValue Res, Chain;
       if (IsStrict) {
-        Res = DAG.getNode(Opc, dl, {VT, MVT::Other}, {Op->getOperand(0), Tmp});
-        Chain = Res.getValue(1);
-        return DAG.getMergeValues({Res, Chain}, dl);
+        unsigned Opc = IsSigned ? X86ISD::STRICT_CVTTP2SI
+                                : X86ISD::STRICT_CVTTP2UI;
+        return DAG.getNode(Opc, dl, {VT, MVT::Other}, {Op->getOperand(0), Tmp});
       }
-      return DAG.getNode(Opc, dl, {VT, MVT::Other}, {DAG.getEntryNode(), Tmp});
+      unsigned Opc = IsSigned ? X86ISD::CVTTP2SI : X86ISD::CVTTP2UI;
+      return DAG.getNode(Opc, dl, VT, Tmp);
     }
 
     return SDValue();
   }
 
   assert(!VT.isVector());
 
   bool UseSSEReg = isScalarFPTypeInSSEReg(SrcVT);
static SDValue LowerVSETCCWithSUBUS(SDValue Op0, SDValue Op1, MVT VT,
   return DAG.getNode(X86ISD::PCMPEQ, dl, VT, Result,
                      DAG.getConstant(0, dl, VT));
 }
 
 static SDValue LowerVSETCC(SDValue Op, const X86Subtarget &Subtarget,
                            SelectionDAG &DAG) {
   bool IsStrict = Op.getOpcode() == ISD::STRICT_FSETCC ||
                   Op.getOpcode() == ISD::STRICT_FSETCCS;
-  bool IsSignaling = Op.getOpcode() == ISD::STRICT_FSETCCS;
-  SDValue Chain = IsStrict ? Op.getOperand(0) : DAG.getEntryNode();
   SDValue Op0 = Op.getOperand(IsStrict ? 1 : 0);
   SDValue Op1 = Op.getOperand(IsStrict ? 2 : 1);
   SDValue CC = Op.getOperand(IsStrict ? 3 : 2);
   MVT VT = Op->getSimpleValueType(0);
   ISD::CondCode Cond = cast<CondCodeSDNode>(CC)->get();
   bool isFP = Op1.getSimpleValueType().isFloatingPoint();
   SDLoc dl(Op);
 
   if (isFP) {
 #ifndef NDEBUG
     MVT EltVT = Op0.getSimpleValueType().getVectorElementType();
     assert(EltVT == MVT::f32 || EltVT == MVT::f64);
 #endif
 
+    bool IsSignaling = Op.getOpcode() == ISD::STRICT_FSETCCS;
+    SDValue Chain = IsStrict ? Op.getOperand(0) : SDValue();
+
     unsigned Opc;
     if (Subtarget.hasAVX512() && VT.getVectorElementType() == MVT::i1) {
       assert(VT.getVectorNumElements() <= 16);
-      Opc = X86ISD::CMPM;
+      Opc = IsStrict ? X86ISD::STRICT_CMPM : X86ISD::CMPM;
     } else {
-      Opc = X86ISD::CMPP;
+      Opc = IsStrict ? X86ISD::STRICT_CMPP : X86ISD::CMPP;
       // The SSE/AVX packed FP comparison nodes are defined with a
       // floating-point vector result that matches the operand type. This allows
       // them to work with an SSE1 target (integer vector types are not legal).
       VT = Op0.getSimpleValueType();
     }
 
     SDValue Cmp;
     bool IsAlwaysSignaling;
if (!Subtarget.hasAVX()) {
           CombineOpc = X86ISD::FOR;
         } else {
           assert(Cond == ISD::SETONE);
           CC0 = 7; // ORD
           CC1 = 4; // NEQ
           CombineOpc = X86ISD::FAND;
         }
 
-        SDValue Cmp0 = DAG.getNode(
+        SDValue Cmp0, Cmp1;
+        if (IsStrict) {
+          Cmp0 = DAG.getNode(
               Opc, dl, {VT, MVT::Other},
               {Chain, Op0, Op1, DAG.getTargetConstant(CC0, dl, MVT::i8)});
-        SDValue Cmp1 = DAG.getNode(
+          Cmp1 = DAG.getNode(
               Opc, dl, {VT, MVT::Other},
               {Chain, Op0, Op1, DAG.getTargetConstant(CC1, dl, MVT::i8)});
           Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Cmp0.getValue(1),
                               Cmp1.getValue(1));
+        } else {
+          Cmp0 = DAG.getNode(
+              Opc, dl, VT, Op0, Op1, DAG.getTargetConstant(CC0, dl, MVT::i8));
+          Cmp1 = DAG.getNode(
+              Opc, dl, VT, Op0, Op1, DAG.getTargetConstant(CC1, dl, MVT::i8));
+        }
         Cmp = DAG.getNode(CombineOpc, dl, VT, Cmp0, Cmp1);
       } else {
+        if (IsStrict) {
           Cmp = DAG.getNode(
               Opc, dl, {VT, MVT::Other},
               {Chain, Op0, Op1, DAG.getTargetConstant(SSECC, dl, MVT::i8)});
           Chain = Cmp.getValue(1);
+        } else
+          Cmp = DAG.getNode(
+              Opc, dl, VT, Op0, Op1, DAG.getTargetConstant(SSECC, dl, MVT::i8));
       }
     } else {
       // Handle all other FP comparisons here.
-      if (IsStrict)
+      if (IsStrict) {
         // Make a flip on already signaling CCs before setting bit 4 of AVX CC.
         SSECC |= (IsAlwaysSignaling ^ IsSignaling) << 4;
         Cmp = DAG.getNode(
             Opc, dl, {VT, MVT::Other},
             {Chain, Op0, Op1, DAG.getTargetConstant(SSECC, dl, MVT::i8)});
         Chain = Cmp.getValue(1);
+      } else
+        Cmp = DAG.getNode(
+            Opc, dl, VT, Op0, Op1, DAG.getTargetConstant(SSECC, dl, MVT::i8));
     }
 
     // If this is SSE/AVX CMPP, bitcast the result back to integer to match the
     // result type of SETCC. The bitcast is expected to be optimized away
     // during combining/isel.
-    if (Opc == X86ISD::CMPP)
-      Cmp = DAG.getBitcast(Op.getSimpleValueType(), Cmp);
 
     if (IsStrict)
       return DAG.getMergeValues({Cmp, Chain}, dl);
 
     return Cmp;
   }
 
   assert(!IsStrict && "Strict SETCC only handles FP operands.");
static SDValue recoverFramePointer(SelectionDAG &DAG, const Function *Fn,
   int RegNodeSize = getSEHRegistrationNodeSize(Fn);
   // RegNodeBase = EntryEBP - RegNodeSize
   // ParentFP = RegNodeBase - ParentFrameOffset
   SDValue RegNodeBase = DAG.getNode(ISD::SUB, dl, PtrVT, EntryEBP,
                                     DAG.getConstant(RegNodeSize, dl, PtrVT));
   return DAG.getNode(ISD::SUB, dl, PtrVT, RegNodeBase, ParentFrameOffset);
 }
 
-// We share some nodes between STRICT and non STRICT FP intrinsics.
-// For these nodes, we need chain them to entry token if they are not called
-// by STRICT FP intrinsics.
-static SDValue getProperNode(unsigned Opcode, const SDLoc &dl, EVT VT,
-                             ArrayRef<SDValue> Ops, SelectionDAG &DAG) {
-  switch (Opcode) {
-  default:
-    return DAG.getNode(Opcode, dl, VT, Ops);
-  case X86ISD::CVTTP2SI:
-  case X86ISD::CVTTP2UI:
-  case X86ISD::CMPP:
-  case X86ISD::CMPM:
-    break;
-  }
-
-  SmallVector<SDValue, 6> NewOps = {DAG.getEntryNode()};
-  NewOps.append(Ops.begin(), Ops.end());
-  return DAG.getNode(Opcode, dl, {VT, MVT::Other}, NewOps);
-}
-
 SDValue X86TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
                                                    SelectionDAG &DAG) const {
   // Helper to detect if the operand is CUR_DIRECTION rounding mode.
   auto isRoundModeCurDirection = [](SDValue Rnd) {
     if (auto *C = dyn_cast<ConstantSDNode>(Rnd))
       return C->getAPIntValue() == X86::STATIC_ROUNDING::CUR_DIRECTION;
 
     return false;
case INTR_TYPE_1OP: {
         unsigned RC = 0;
         if (isRoundModeSAEToX(Rnd, RC))
           return DAG.getNode(IntrWithRoundingModeOpcode, dl, Op.getValueType(),
                              Op.getOperand(1),
                              DAG.getTargetConstant(RC, dl, MVT::i32));
         if (!isRoundModeCurDirection(Rnd))
           return SDValue();
       }
-      return getProperNode(IntrData->Opc0, dl, Op.getValueType(),
-                           Op.getOperand(1), DAG);
+      return DAG.getNode(IntrData->Opc0, dl, Op.getValueType(),
+                         Op.getOperand(1));
     }
     case INTR_TYPE_1OP_SAE: {
       SDValue Sae = Op.getOperand(2);
 
       unsigned Opc;
       if (isRoundModeCurDirection(Sae))
         Opc = IntrData->Opc0;
       else if (isRoundModeSAE(Sae))
case INTR_TYPE_3OP_IMM8: {
         if (isRoundModeSAEToX(Rnd, RC))
           return DAG.getNode(IntrWithRoundingModeOpcode, dl, Op.getValueType(),
                              Src1, Src2, Src3,
                              DAG.getTargetConstant(RC, dl, MVT::i32));
         if (!isRoundModeCurDirection(Rnd))
           return SDValue();
       }
 
-      return getProperNode(IntrData->Opc0, dl, Op.getValueType(),
-                           {Src1, Src2, Src3}, DAG);
+      return DAG.getNode(IntrData->Opc0, dl, Op.getValueType(),
+                         {Src1, Src2, Src3});
     }
     case INTR_TYPE_4OP:
       return DAG.getNode(IntrData->Opc0, dl, Op.getValueType(), Op.getOperand(1),
         Op.getOperand(2), Op.getOperand(3), Op.getOperand(4));
     case INTR_TYPE_1OP_MASK: {
       SDValue Src = Op.getOperand(1);
       SDValue PassThru = Op.getOperand(2);
       SDValue Mask = Op.getOperand(3);
       // We add rounding mode to the Node when
       //   - RC Opcode is specified and
       //   - RC is not "current direction".
       unsigned IntrWithRoundingModeOpcode = IntrData->Opc1;
       if (IntrWithRoundingModeOpcode != 0) {
         SDValue Rnd = Op.getOperand(4);
         unsigned RC = 0;
         if (isRoundModeSAEToX(Rnd, RC))
           return getVectorMaskingNode(
               DAG.getNode(IntrWithRoundingModeOpcode, dl, Op.getValueType(),
                           Src, DAG.getTargetConstant(RC, dl, MVT::i32)),
               Mask, PassThru, Subtarget, DAG);
         if (!isRoundModeCurDirection(Rnd))
           return SDValue();
       }
       return getVectorMaskingNode(
-          getProperNode(IntrData->Opc0, dl, VT, Src, DAG), Mask, PassThru,
+          DAG.getNode(IntrData->Opc0, dl, VT, Src), Mask, PassThru,
           Subtarget, DAG);
     }
     case INTR_TYPE_1OP_MASK_SAE: {
       SDValue Src = Op.getOperand(1);
       SDValue PassThru = Op.getOperand(2);
       SDValue Mask = Op.getOperand(3);
       SDValue Rnd = Op.getOperand(4);
 
       unsigned Opc;
       if (isRoundModeCurDirection(Rnd))
         Opc = IntrData->Opc0;
       else if (isRoundModeSAE(Rnd))
         Opc = IntrData->Opc1;
       else
         return SDValue();
 
-      return getVectorMaskingNode(getProperNode(Opc, dl, VT, Src, DAG), Mask,
-                                  PassThru, Subtarget, DAG);
+      return getVectorMaskingNode(DAG.getNode(Opc, dl, VT, Src), Mask, PassThru,
+                                  Subtarget, DAG);
     }
     case INTR_TYPE_SCALAR_MASK: {
       SDValue Src1 = Op.getOperand(1);
       SDValue Src2 = Op.getOperand(2);
       SDValue passThru = Op.getOperand(3);
       SDValue Mask = Op.getOperand(4);
       unsigned IntrWithRoundingModeOpcode = IntrData->Opc1;
       // There are 2 kinds of intrinsics in this group:
case CMP_MASK_CC: {
         SDValue Sae = Op.getOperand(4);
         if (isRoundModeSAE(Sae))
           return DAG.getNode(IntrData->Opc1, dl, MaskVT, Op.getOperand(1),
                              Op.getOperand(2), CC, Sae);
         if (!isRoundModeCurDirection(Sae))
           return SDValue();
       }
       //default rounding mode
-      return getProperNode(IntrData->Opc0, dl, MaskVT,
-                           {Op.getOperand(1), Op.getOperand(2), CC}, DAG);
+      return DAG.getNode(IntrData->Opc0, dl, MaskVT,
+                         {Op.getOperand(1), Op.getOperand(2), CC});
     }
     case CMP_MASK_SCALAR_CC: {
       SDValue Src1 = Op.getOperand(1);
       SDValue Src2 = Op.getOperand(2);
       SDValue CC = Op.getOperand(3);
       SDValue Mask = Op.getOperand(4);
 
       SDValue Cmp;
if (IntrData) {
     case CVTPD2DQ_MASK:
     case CVTQQ2PS_MASK:
     case TRUNCATE_TO_REG: {
       SDValue Src = Op.getOperand(1);
       SDValue PassThru = Op.getOperand(2);
       SDValue Mask = Op.getOperand(3);
 
       if (isAllOnesConstant(Mask))
-        return getProperNode(IntrData->Opc0, dl, Op.getValueType(), Src, DAG);
+        return DAG.getNode(IntrData->Opc0, dl, Op.getValueType(), Src);
 
       MVT SrcVT = Src.getSimpleValueType();
       MVT MaskVT = MVT::getVectorVT(MVT::i1, SrcVT.getVectorNumElements());
       Mask = getMaskNode(Mask, MaskVT, Subtarget, DAG, dl);
-      return getProperNode(IntrData->Opc1, dl, Op.getValueType(),
-                           {Src, PassThru, Mask}, DAG);
+      return DAG.getNode(IntrData->Opc1, dl, Op.getValueType(),
+                         {Src, PassThru, Mask});
     }
     case CVTPS2PH_MASK: {
       SDValue Src = Op.getOperand(1);
       SDValue Rnd = Op.getOperand(2);
       SDValue PassThru = Op.getOperand(3);
       SDValue Mask = Op.getOperand(4);
 
       if (isAllOnesConstant(Mask))
if (VT == MVT::v2i32) {
       if (Src.getValueType() == MVT::v2f64) {
         if (!IsSigned && !Subtarget.hasVLX()) {
           // If we have VLX we can emit a target specific FP_TO_UINT node,
           // otherwise we can defer to the generic legalizer which will widen
           // the input as well. This will be further widened during op
           // legalization to v8i32<-v8f64.
           return;
         }
-        unsigned Opc = IsSigned ? X86ISD::CVTTP2SI : X86ISD::CVTTP2UI;
         SDValue Res;
         SDValue Chain;
         if (IsStrict) {
+          unsigned Opc = IsSigned ? X86ISD::STRICT_CVTTP2SI
+                                  : X86ISD::STRICT_CVTTP2UI;
           Res = DAG.getNode(Opc, dl, {MVT::v4i32, MVT::Other},
                             {N->getOperand(0), Src});
           Chain = Res.getValue(1);
-        } else
-          Res = DAG.getNode(Opc, dl, {MVT::v4i32, MVT::Other},
-                            {DAG.getEntryNode(), Src});
+        } else {
+          unsigned Opc = IsSigned ? X86ISD::CVTTP2SI : X86ISD::CVTTP2UI;
+          Res = DAG.getNode(Opc, dl, MVT::v4i32, Src);
+        }
         Results.push_back(Res);
         if (IsStrict)
           Results.push_back(Chain);
         return;
       }
 
       // The FP_TO_INTHelper below only handles f32/f64/f80 scalar inputs,
       // so early out here.
const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::CALL:               return "X86ISD::CALL";
   case X86ISD::BT:                 return "X86ISD::BT";
   case X86ISD::CMP:                return "X86ISD::CMP";
   case X86ISD::STRICT_FCMP:        return "X86ISD::STRICT_FCMP";
   case X86ISD::STRICT_FCMPS:       return "X86ISD::STRICT_FCMPS";
   case X86ISD::COMI:               return "X86ISD::COMI";
   case X86ISD::UCOMI:              return "X86ISD::UCOMI";
   case X86ISD::CMPM:               return "X86ISD::CMPM";
+  case X86ISD::STRICT_CMPM:        return "X86ISD::STRICT_CMPM";
   case X86ISD::CMPM_SAE:           return "X86ISD::CMPM_SAE";
   case X86ISD::SETCC:              return "X86ISD::SETCC";
   case X86ISD::SETCC_CARRY:        return "X86ISD::SETCC_CARRY";
   case X86ISD::FSETCC:             return "X86ISD::FSETCC";
   case X86ISD::FSETCCM:            return "X86ISD::FSETCCM";
   case X86ISD::FSETCCM_SAE:        return "X86ISD::FSETCCM_SAE";
   case X86ISD::CMOV:               return "X86ISD::CMOV";
   case X86ISD::BRCOND:             return "X86ISD::BRCOND";
const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::VSRAI:              return "X86ISD::VSRAI";
   case X86ISD::VSHLV:              return "X86ISD::VSHLV";
   case X86ISD::VSRLV:              return "X86ISD::VSRLV";
   case X86ISD::VSRAV:              return "X86ISD::VSRAV";
   case X86ISD::VROTLI:             return "X86ISD::VROTLI";
   case X86ISD::VROTRI:             return "X86ISD::VROTRI";
   case X86ISD::VPPERM:             return "X86ISD::VPPERM";
   case X86ISD::CMPP:               return "X86ISD::CMPP";
+  case X86ISD::STRICT_CMPP:        return "X86ISD::STRICT_CMPP";
   case X86ISD::PCMPEQ:             return "X86ISD::PCMPEQ";
   case X86ISD::PCMPGT:             return "X86ISD::PCMPGT";
   case X86ISD::PHMINPOS:           return "X86ISD::PHMINPOS";
   case X86ISD::ADD:                return "X86ISD::ADD";
   case X86ISD::SUB:                return "X86ISD::SUB";
   case X86ISD::ADC:                return "X86ISD::ADC";
   case X86ISD::SBB:                return "X86ISD::SBB";
   case X86ISD::SMUL:               return "X86ISD::SMUL";
const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::SCALEFS:            return "X86ISD::SCALEFS";
   case X86ISD::SCALEFS_RND:        return "X86ISD::SCALEFS_RND";
   case X86ISD::AVG:                return "X86ISD::AVG";
   case X86ISD::MULHRS:             return "X86ISD::MULHRS";
   case X86ISD::SINT_TO_FP_RND:     return "X86ISD::SINT_TO_FP_RND";
   case X86ISD::UINT_TO_FP_RND:     return "X86ISD::UINT_TO_FP_RND";
   case X86ISD::CVTTP2SI:           return "X86ISD::CVTTP2SI";
   case X86ISD::CVTTP2UI:           return "X86ISD::CVTTP2UI";
+  case X86ISD::STRICT_CVTTP2SI:    return "X86ISD::STRICT_CVTTP2SI";
+  case X86ISD::STRICT_CVTTP2UI:    return "X86ISD::STRICT_CVTTP2UI";
   case X86ISD::MCVTTP2SI:          return "X86ISD::MCVTTP2SI";
   case X86ISD::MCVTTP2UI:          return "X86ISD::MCVTTP2UI";
   case X86ISD::CVTTP2SI_SAE:       return "X86ISD::CVTTP2SI_SAE";
   case X86ISD::CVTTP2UI_SAE:       return "X86ISD::CVTTP2UI_SAE";
   case X86ISD::CVTTS2SI:           return "X86ISD::CVTTS2SI";
   case X86ISD::CVTTS2UI:           return "X86ISD::CVTTS2UI";
   case X86ISD::CVTTS2SI_SAE:       return "X86ISD::CVTTS2SI_SAE";
   case X86ISD::CVTTS2UI_SAE:       return "X86ISD::CVTTS2UI_SAE";
if (N->getOpcode() == X86ISD::VZEXT_MOVL &&
       (N->getOperand(0).getOperand(0).getValueType() == MVT::v4f32 ||
        N->getOperand(0).getOperand(0).getValueType() == MVT::v4i32)) {
     SDValue In = N->getOperand(0).getOperand(0);
     switch (In.getOpcode()) {
     default:
       break;
     case X86ISD::CVTP2SI:   case X86ISD::CVTP2UI:
     case X86ISD::MCVTP2SI:  case X86ISD::MCVTP2UI:
+    case X86ISD::CVTTP2SI:  case X86ISD::CVTTP2UI:
     case X86ISD::MCVTTP2SI: case X86ISD::MCVTTP2UI:
     case X86ISD::CVTSI2P:   case X86ISD::CVTUI2P:
     case X86ISD::MCVTSI2P:  case X86ISD::MCVTUI2P:
     case X86ISD::VFPROUND:  case X86ISD::VMFPROUND:
       if (In.getOperand(0).getValueType() == MVT::v2f64 ||
           In.getOperand(0).getValueType() == MVT::v2i64)
         return N->getOperand(0); // return the bitcast
       break;
-    case X86ISD::CVTTP2SI:
-    case X86ISD::CVTTP2UI:
+    case X86ISD::STRICT_CVTTP2SI:
+    case X86ISD::STRICT_CVTTP2UI:
       if (In.getOperand(1).getValueType() == MVT::v2f64 ||
           In.getOperand(1).getValueType() == MVT::v2i64)
         return N->getOperand(0);
       break;
     }
   }
 
   // Pull subvector inserts into undef through VZEXT_MOVL by making it an
if (VT.getVectorNumElements() < InVT.getVectorNumElements() &&
     }
   }
 
   return SDValue();
 }
 
 static SDValue combineCVTP2I_CVTTP2I(SDNode *N, SelectionDAG &DAG,
                                      TargetLowering::DAGCombinerInfo &DCI) {
+  // FIXME: Handle strict fp nodes.
   EVT VT = N->getValueType(0);
 
   // Convert a full vector load into vzload when not all bits are needed.
-  SDValue In;
-  if (N->getOpcode() == X86ISD::CVTTP2SI || N->getOpcode() == X86ISD::CVTTP2UI)
-    In = N->getOperand(1);
-  else
-    In = N->getOperand(0);
+  SDValue In = N->getOperand(0);
   MVT InVT = In.getSimpleValueType();
   if (VT.getVectorNumElements() < InVT.getVectorNumElements() &&
       ISD::isNormalLoad(In.getNode()) && In.hasOneUse()) {
     assert(InVT.is128BitVector() && "Expected 128-bit input vector");
     LoadSDNode *LN = cast<LoadSDNode>(In);
     // Unless the load is volatile or atomic.
     if (LN->isSimple()) {
       SDLoc dl(N);
       unsigned NumBits = InVT.getScalarSizeInBits() * VT.getVectorNumElements();
       MVT MemVT = MVT::getFloatingPointVT(NumBits);
       MVT LoadVT = MVT::getVectorVT(MemVT, 128 / NumBits);
       SDVTList Tys = DAG.getVTList(LoadVT, MVT::Other);
       SDValue Ops[] = { LN->getChain(), LN->getBasePtr() };
       SDValue VZLoad =
           DAG.getMemIntrinsicNode(X86ISD::VZEXT_LOAD, dl, Tys, Ops, MemVT,
                                   LN->getPointerInfo(),
                                   LN->getAlignment(),
                                   LN->getMemOperand()->getFlags());
-      SDValue Convert = getProperNode(N->getOpcode(), dl, VT,
-                                      DAG.getBitcast(InVT, VZLoad), DAG);
-      if (Convert->getOpcode() == X86ISD::CVTTP2SI ||
-          Convert->getOpcode() == X86ISD::CVTTP2UI)
-        DCI.CombineTo(N, Convert.getValue(0), Convert.getValue(1));
-      else
-        DCI.CombineTo(N, Convert);
+      SDValue Convert = DAG.getNode(N->getOpcode(), dl, VT,
+                                    DAG.getBitcast(InVT, VZLoad));
       DAG.ReplaceAllUsesOfValueWith(SDValue(LN, 1), VZLoad.getValue(1));
       return SDValue(N, 0);
     }
   }
 
   return SDValue();
 }
 

llvm/lib/Target/X86/X86InstrAVX512.td

 
 multiclass avx512_vcmp_common<X86FoldableSchedWrite sched, X86VectorVTInfo _,
                               string Name> {
 let Uses = [MXCSR], mayRaiseFPException = 1 in {
   defm  rri  : AVX512_maskable_cmp<0xC2, MRMSrcReg, _,
                    (outs _.KRC:$dst), (ins _.RC:$src1, _.RC:$src2,u8imm:$cc),
                    "vcmp"#_.Suffix,
                    "$cc, $src2, $src1", "$src1, $src2, $cc",
-                   (X86cmpm (_.VT _.RC:$src1), (_.VT _.RC:$src2), timm:$cc),
+                   (X86any_cmpm (_.VT _.RC:$src1), (_.VT _.RC:$src2), timm:$cc),
                    (X86cmpm_su (_.VT _.RC:$src1), (_.VT _.RC:$src2), timm:$cc),
                    1>, Sched<[sched]>;
 
   defm  rmi  : AVX512_maskable_cmp<0xC2, MRMSrcMem, _,
                 (outs _.KRC:$dst),(ins _.RC:$src1, _.MemOp:$src2, u8imm:$cc),
                 "vcmp"#_.Suffix,
                 "$cc, $src2, $src1", "$src1, $src2, $cc",
-                (X86cmpm (_.VT _.RC:$src1), (_.VT (_.LdFrag addr:$src2)),
+                (X86any_cmpm (_.VT _.RC:$src1), (_.VT (_.LdFrag addr:$src2)),
                              timm:$cc),
                 (X86cmpm_su (_.VT _.RC:$src1), (_.VT (_.LdFrag addr:$src2)),
                             timm:$cc)>,
                 Sched<[sched.Folded, sched.ReadAfterFold]>;
 
   defm  rmbi : AVX512_maskable_cmp<0xC2, MRMSrcMem, _,
                 (outs _.KRC:$dst),
                 (ins _.RC:$src1, _.ScalarMemOp:$src2, u8imm:$cc),
                 "vcmp"#_.Suffix,
                 "$cc, ${src2}"#_.BroadcastStr#", $src1",
                 "$src1, ${src2}"#_.BroadcastStr#", $cc",
-                (X86cmpm (_.VT _.RC:$src1),
+                (X86any_cmpm (_.VT _.RC:$src1),
                              (_.VT (_.BroadcastLdFrag addr:$src2)),
                              timm:$cc),
                 (X86cmpm_su (_.VT _.RC:$src1),
                             (_.VT (_.BroadcastLdFrag addr:$src2)),
                             timm:$cc)>,
                 EVEX_B, Sched<[sched.Folded, sched.ReadAfterFold]>;
   }
 
   // Patterns for selecting with loads in other operand.
-  def : Pat<(X86cmpm (_.LdFrag addr:$src2), (_.VT _.RC:$src1),
+  def : Pat<(X86any_cmpm (_.LdFrag addr:$src2), (_.VT _.RC:$src1),
                          timm:$cc),
             (!cast<Instruction>(Name#_.ZSuffix#"rmi") _.RC:$src1, addr:$src2,
                                                       (X86cmpm_imm_commute timm:$cc))>;
 
   def : Pat<(and _.KRCWM:$mask, (X86cmpm_su (_.LdFrag addr:$src2),
                                             (_.VT _.RC:$src1),
                                             timm:$cc)),
             (!cast<Instruction>(Name#_.ZSuffix#"rmik") _.KRCWM:$mask,
                                                        _.RC:$src1, addr:$src2,
                                                        (X86cmpm_imm_commute timm:$cc))>;
 
-  def : Pat<(X86cmpm (_.BroadcastLdFrag addr:$src2),
+  def : Pat<(X86any_cmpm (_.BroadcastLdFrag addr:$src2),
                          (_.VT _.RC:$src1), timm:$cc),
             (!cast<Instruction>(Name#_.ZSuffix#"rmbi") _.RC:$src1, addr:$src2,
                                                        (X86cmpm_imm_commute timm:$cc))>;
 
   def : Pat<(and _.KRCWM:$mask, (X86cmpm_su (_.BroadcastLdFrag addr:$src2),
                                             (_.VT _.RC:$src1),
                                             timm:$cc)),
             (!cast<Instruction>(Name#_.ZSuffix#"rmbik") _.KRCWM:$mask,
                                                         _.RC:$src1, addr:$src2,
def : Pat<(Narrow.KVT (and Narrow.KRC:$mask,
            (Wide.VT (INSERT_SUBREG (IMPLICIT_DEF), Narrow.RC:$src1, Narrow.SubRegIdx)),
            addr:$src2, (CommFrag_su.OperandTransform $cc)), Narrow.KRC)>;
 }
 
 // Same as above, but for fp types which don't use PatFrags.
 multiclass axv512_cmp_packed_cc_no_vlx_lowering<string InstStr,
                                                 X86VectorVTInfo Narrow,
                                                 X86VectorVTInfo Wide> {
-def : Pat<(Narrow.KVT (X86cmpm (Narrow.VT Narrow.RC:$src1),
+def : Pat<(Narrow.KVT (X86any_cmpm (Narrow.VT Narrow.RC:$src1),
                                    (Narrow.VT Narrow.RC:$src2), timm:$cc)),
           (COPY_TO_REGCLASS
            (!cast<Instruction>(InstStr#"Zrri")
             (Wide.VT (INSERT_SUBREG (IMPLICIT_DEF), Narrow.RC:$src1, Narrow.SubRegIdx)),
             (Wide.VT (INSERT_SUBREG (IMPLICIT_DEF), Narrow.RC:$src2, Narrow.SubRegIdx)),
             timm:$cc), Narrow.KRC)>;
 
 def : Pat<(Narrow.KVT (and Narrow.KRC:$mask,
                            (X86cmpm_su (Narrow.VT Narrow.RC:$src1),
                                        (Narrow.VT Narrow.RC:$src2), timm:$cc))),
           (COPY_TO_REGCLASS (!cast<Instruction>(InstStr#"Zrrik")
            (COPY_TO_REGCLASS Narrow.KRC:$mask, Wide.KRC),
            (Wide.VT (INSERT_SUBREG (IMPLICIT_DEF), Narrow.RC:$src1, Narrow.SubRegIdx)),
            (Wide.VT (INSERT_SUBREG (IMPLICIT_DEF), Narrow.RC:$src2, Narrow.SubRegIdx)),
            timm:$cc), Narrow.KRC)>;
 
 // Broadcast load.
-def : Pat<(Narrow.KVT (X86cmpm (Narrow.VT Narrow.RC:$src1),
+def : Pat<(Narrow.KVT (X86any_cmpm (Narrow.VT Narrow.RC:$src1),
                                    (Narrow.VT (Narrow.BroadcastLdFrag addr:$src2)), timm:$cc)),
           (COPY_TO_REGCLASS
            (!cast<Instruction>(InstStr#"Zrmbi")
             (Wide.VT (INSERT_SUBREG (IMPLICIT_DEF), Narrow.RC:$src1, Narrow.SubRegIdx)),
             addr:$src2, timm:$cc), Narrow.KRC)>;
 
 def : Pat<(Narrow.KVT (and Narrow.KRC:$mask,
                            (X86cmpm_su (Narrow.VT Narrow.RC:$src1),
                                        (Narrow.VT (Narrow.BroadcastLdFrag addr:$src2)), timm:$cc))),
           (COPY_TO_REGCLASS (!cast<Instruction>(InstStr#"Zrmbik")
            (COPY_TO_REGCLASS Narrow.KRC:$mask, Wide.KRC),
            (Wide.VT (INSERT_SUBREG (IMPLICIT_DEF), Narrow.RC:$src1, Narrow.SubRegIdx)),
            addr:$src2, timm:$cc), Narrow.KRC)>;
 
 // Commuted with broadcast load.
-def : Pat<(Narrow.KVT (X86cmpm (Narrow.VT (Narrow.BroadcastLdFrag addr:$src2)),
+def : Pat<(Narrow.KVT (X86any_cmpm (Narrow.VT (Narrow.BroadcastLdFrag addr:$src2)),
                                    (Narrow.VT Narrow.RC:$src1), timm:$cc)),
           (COPY_TO_REGCLASS
            (!cast<Instruction>(InstStr#"Zrmbi")
             (Wide.VT (INSERT_SUBREG (IMPLICIT_DEF), Narrow.RC:$src1, Narrow.SubRegIdx)),
             addr:$src2, (X86cmpm_imm_commute timm:$cc)), Narrow.KRC)>;
 
 def : Pat<(Narrow.KVT (and Narrow.KRC:$mask,
                            (X86cmpm_su (Narrow.VT (Narrow.BroadcastLdFrag addr:$src2)),
                                        (Narrow.VT Narrow.RC:$src1), timm:$cc))),

 
 defm VCVTDQ2PD : avx512_cvtdq2pd<0xE6, "vcvtdq2pd", any_sint_to_fp, X86VSintToFP,
                                  SchedWriteCvtDQ2PD>, XS, EVEX_CD8<32, CD8VH>;
 
 defm VCVTDQ2PS : avx512_cvtdq2ps<0x5B, "vcvtdq2ps", any_sint_to_fp,
                                 X86VSintToFpRnd, SchedWriteCvtDQ2PS>,
                                 PS, EVEX_CD8<32, CD8VF>;
 
-defm VCVTTPS2DQ : avx512_cvttps2dq<0x5B, "vcvttps2dq", X86cvttp2si,
+defm VCVTTPS2DQ : avx512_cvttps2dq<0x5B, "vcvttps2dq", X86any_cvttp2si,
                                 X86cvttp2siSAE, SchedWriteCvtPS2DQ>,
                                 XS, EVEX_CD8<32, CD8VF>;
 
-defm VCVTTPD2DQ : avx512_cvttpd2dq<0xE6, "vcvttpd2dq", X86cvttp2si,
+defm VCVTTPD2DQ : avx512_cvttpd2dq<0xE6, "vcvttpd2dq", X86any_cvttp2si,
                                  X86cvttp2siSAE, SchedWriteCvtPD2DQ>,
                                  PD, VEX_W, EVEX_CD8<64, CD8VF>;
 
-defm VCVTTPS2UDQ : avx512_cvttps2dq<0x78, "vcvttps2udq", X86cvttp2ui,
+defm VCVTTPS2UDQ : avx512_cvttps2dq<0x78, "vcvttps2udq", X86any_cvttp2ui,
                                  X86cvttp2uiSAE, SchedWriteCvtPS2DQ>, PS,
                                  EVEX_CD8<32, CD8VF>;
 
-defm VCVTTPD2UDQ : avx512_cvttpd2dq<0x78, "vcvttpd2udq", X86cvttp2ui,
+defm VCVTTPD2UDQ : avx512_cvttpd2dq<0x78, "vcvttpd2udq", X86any_cvttp2ui,
                                  X86cvttp2uiSAE, SchedWriteCvtPD2DQ>,
                                  PS, VEX_W, EVEX_CD8<64, CD8VF>;
 
 defm VCVTUDQ2PD : avx512_cvtdq2pd<0x7A, "vcvtudq2pd", any_uint_to_fp,
                                   X86VUintToFP, SchedWriteCvtDQ2PD>, XS,
                                   EVEX_CD8<32, CD8VH>;
 
 defm VCVTUDQ2PS : avx512_cvtdq2ps<0x7A, "vcvtudq2ps", any_uint_to_fp,

 defm VCVTPD2UQQ : avx512_cvtpd2qq<0x79, "vcvtpd2uqq", X86cvtp2UInt,
                                  X86cvtp2UIntRnd, SchedWriteCvtPD2DQ>, VEX_W,
                                  PD, EVEX_CD8<64, CD8VF>;
 
 defm VCVTPS2UQQ : avx512_cvtps2qq<0x79, "vcvtps2uqq", X86cvtp2UInt,
                                  X86cvtp2UIntRnd, SchedWriteCvtPS2DQ>, PD,
                                  EVEX_CD8<32, CD8VH>;
 
-defm VCVTTPD2QQ : avx512_cvttpd2qq<0x7A, "vcvttpd2qq", X86cvttp2si,
+defm VCVTTPD2QQ : avx512_cvttpd2qq<0x7A, "vcvttpd2qq", X86any_cvttp2si,
                                  X86cvttp2siSAE, SchedWriteCvtPD2DQ>, VEX_W,
                                  PD, EVEX_CD8<64, CD8VF>;
 
-defm VCVTTPS2QQ : avx512_cvttps2qq<0x7A, "vcvttps2qq", X86cvttp2si,
+defm VCVTTPS2QQ : avx512_cvttps2qq<0x7A, "vcvttps2qq", X86any_cvttp2si,
                                  X86cvttp2siSAE, SchedWriteCvtPS2DQ>, PD,
                                  EVEX_CD8<32, CD8VH>;
 
-defm VCVTTPD2UQQ : avx512_cvttpd2qq<0x78, "vcvttpd2uqq", X86cvttp2ui,
+defm VCVTTPD2UQQ : avx512_cvttpd2qq<0x78, "vcvttpd2uqq", X86any_cvttp2ui,
                                  X86cvttp2uiSAE, SchedWriteCvtPD2DQ>, VEX_W,
                                  PD, EVEX_CD8<64, CD8VF>;
 
-defm VCVTTPS2UQQ : avx512_cvttps2qq<0x78, "vcvttps2uqq", X86cvttp2ui,
+defm VCVTTPS2UQQ : avx512_cvttps2qq<0x78, "vcvttps2uqq", X86any_cvttp2ui,
                                  X86cvttp2uiSAE, SchedWriteCvtPS2DQ>, PD,
                                  EVEX_CD8<32, CD8VH>;
 
 defm VCVTQQ2PD : avx512_cvtqq2pd<0xE6, "vcvtqq2pd", any_sint_to_fp,
                             X86VSintToFpRnd, SchedWriteCvtDQ2PD>, VEX_W, XS,
                             EVEX_CD8<64, CD8VF>;
 
 defm VCVTUQQ2PD : avx512_cvtqq2pd<0x7A, "vcvtuqq2pd", any_uint_to_fp,
def : Pat<(X86mcvtp2Int (v2f64 (X86VBroadcastld64 addr:$src)),
                           (v4i32 VR128X:$src0), VK2WM:$mask),
             (VCVTPD2DQZ128rmbk VR128X:$src0, VK2WM:$mask, addr:$src)>;
   def : Pat<(X86mcvtp2Int (v2f64 (X86VBroadcastld64 addr:$src)),
                           v4i32x_info.ImmAllZerosV, VK2WM:$mask),
             (VCVTPD2DQZ128rmbkz VK2WM:$mask, addr:$src)>;
 
   // Special patterns to allow use of X86mcvttp2si for masking. Instruction
   // patterns have been disabled with null_frag.
-  def : Pat<(v4i32 (X86cvttp2si (v2f64 VR128X:$src))),
+  def : Pat<(v4i32 (X86any_cvttp2si (v2f64 VR128X:$src))),
             (VCVTTPD2DQZ128rr VR128X:$src)>;
   def : Pat<(X86mcvttp2si (v2f64 VR128X:$src), (v4i32 VR128X:$src0),
                           VK2WM:$mask),
             (VCVTTPD2DQZ128rrk VR128X:$src0, VK2WM:$mask, VR128X:$src)>;
   def : Pat<(X86mcvttp2si (v2f64 VR128X:$src), v4i32x_info.ImmAllZerosV,
                           VK2WM:$mask),
             (VCVTTPD2DQZ128rrkz VK2WM:$mask, VR128X:$src)>;
 
-  def : Pat<(v4i32 (X86cvttp2si (loadv2f64 addr:$src))),
+  def : Pat<(v4i32 (X86any_cvttp2si (loadv2f64 addr:$src))),
             (VCVTTPD2DQZ128rm addr:$src)>;
   def : Pat<(X86mcvttp2si (loadv2f64 addr:$src), (v4i32 VR128X:$src0),
                           VK2WM:$mask),
             (VCVTTPD2DQZ128rmk VR128X:$src0, VK2WM:$mask, addr:$src)>;
   def : Pat<(X86mcvttp2si (loadv2f64 addr:$src), v4i32x_info.ImmAllZerosV,
                           VK2WM:$mask),
             (VCVTTPD2DQZ128rmkz VK2WM:$mask, addr:$src)>;
 
-  def : Pat<(v4i32 (X86cvttp2si (v2f64 (X86VBroadcastld64 addr:$src)))),
+  def : Pat<(v4i32 (X86any_cvttp2si (v2f64 (X86VBroadcastld64 addr:$src)))),
             (VCVTTPD2DQZ128rmb addr:$src)>;
   def : Pat<(X86mcvttp2si (v2f64 (X86VBroadcastld64 addr:$src)),
                           (v4i32 VR128X:$src0), VK2WM:$mask),
             (VCVTTPD2DQZ128rmbk VR128X:$src0, VK2WM:$mask, addr:$src)>;
   def : Pat<(X86mcvttp2si (v2f64 (X86VBroadcastld64 addr:$src)),
                           v4i32x_info.ImmAllZerosV, VK2WM:$mask),
             (VCVTTPD2DQZ128rmbkz VK2WM:$mask, addr:$src)>;
 
def : Pat<(X86mcvtp2UInt (v2f64 (X86VBroadcastld64 addr:$src)),
                            (v4i32 VR128X:$src0), VK2WM:$mask),
             (VCVTPD2UDQZ128rmbk VR128X:$src0, VK2WM:$mask, addr:$src)>;
   def : Pat<(X86mcvtp2UInt (v2f64 (X86VBroadcastld64 addr:$src)),
                            v4i32x_info.ImmAllZerosV, VK2WM:$mask),
             (VCVTPD2UDQZ128rmbkz VK2WM:$mask, addr:$src)>;
 
   // Special patterns to allow use of X86mcvtp2UInt for masking. Instruction
   // patterns have been disabled with null_frag.
-  def : Pat<(v4i32 (X86cvttp2ui (v2f64 VR128X:$src))),
+  def : Pat<(v4i32 (X86any_cvttp2ui (v2f64 VR128X:$src))),
             (VCVTTPD2UDQZ128rr VR128X:$src)>;
   def : Pat<(X86mcvttp2ui (v2f64 VR128X:$src), (v4i32 VR128X:$src0),
                           VK2WM:$mask),
             (VCVTTPD2UDQZ128rrk VR128X:$src0, VK2WM:$mask, VR128X:$src)>;
   def : Pat<(X86mcvttp2ui (v2f64 VR128X:$src), v4i32x_info.ImmAllZerosV,
                           VK2WM:$mask),
             (VCVTTPD2UDQZ128rrkz VK2WM:$mask, VR128X:$src)>;
 
-  def : Pat<(v4i32 (X86cvttp2ui (loadv2f64 addr:$src))),
+  def : Pat<(v4i32 (X86any_cvttp2ui (loadv2f64 addr:$src))),
             (VCVTTPD2UDQZ128rm addr:$src)>;
   def : Pat<(X86mcvttp2ui (loadv2f64 addr:$src), (v4i32 VR128X:$src0),
                           VK2WM:$mask),
             (VCVTTPD2UDQZ128rmk VR128X:$src0, VK2WM:$mask, addr:$src)>;
   def : Pat<(X86mcvttp2ui (loadv2f64 addr:$src), v4i32x_info.ImmAllZerosV,
                           VK2WM:$mask),
             (VCVTTPD2UDQZ128rmkz VK2WM:$mask, addr:$src)>;
 
-  def : Pat<(v4i32 (X86cvttp2ui (v2f64 (X86VBroadcastld64 addr:$src)))),
+  def : Pat<(v4i32 (X86any_cvttp2ui (v2f64 (X86VBroadcastld64 addr:$src)))),
             (VCVTTPD2UDQZ128rmb addr:$src)>;
   def : Pat<(X86mcvttp2ui (v2f64 (X86VBroadcastld64 addr:$src)),
                           (v4i32 VR128X:$src0), VK2WM:$mask),
             (VCVTTPD2UDQZ128rmbk VR128X:$src0, VK2WM:$mask, addr:$src)>;
   def : Pat<(X86mcvttp2ui (v2f64 (X86VBroadcastld64 addr:$src)),
                           v4i32x_info.ImmAllZerosV, VK2WM:$mask),
             (VCVTTPD2UDQZ128rmbkz VK2WM:$mask, addr:$src)>;
 }
def : Pat<(v2i64 (vselect VK2WM:$mask,
                             (X86cvtp2UInt (bc_v4f32 (v2f64 (X86vzload64 addr:$src)))),
                             VR128X:$src0)),
             (VCVTPS2UQQZ128rmk VR128X:$src0, VK2WM:$mask, addr:$src)>;
   def : Pat<(v2i64 (vselect VK2WM:$mask,
                             (X86cvtp2UInt (bc_v4f32 (v2f64 (X86vzload64 addr:$src)))),
                             v2i64x_info.ImmAllZerosV)),
             (VCVTPS2UQQZ128rmkz VK2WM:$mask, addr:$src)>;
 
-  def : Pat<(v2i64 (X86cvttp2si (bc_v4f32 (v2f64 (X86vzload64 addr:$src))))),
+  def : Pat<(v2i64 (X86any_cvttp2si (bc_v4f32 (v2f64 (X86vzload64 addr:$src))))),
             (VCVTTPS2QQZ128rm addr:$src)>;
   def : Pat<(v2i64 (vselect VK2WM:$mask,
                             (X86cvttp2si (bc_v4f32 (v2f64 (X86vzload64 addr:$src)))),
                             VR128X:$src0)),
             (VCVTTPS2QQZ128rmk VR128X:$src0, VK2WM:$mask, addr:$src)>;
   def : Pat<(v2i64 (vselect VK2WM:$mask,
                             (X86cvttp2si (bc_v4f32 (v2f64 (X86vzload64 addr:$src)))),
                             v2i64x_info.ImmAllZerosV)),
             (VCVTTPS2QQZ128rmkz VK2WM:$mask, addr:$src)>;
 
-  def : Pat<(v2i64 (X86cvttp2ui (bc_v4f32 (v2f64 (X86vzload64 addr:$src))))),
+  def : Pat<(v2i64 (X86any_cvttp2ui (bc_v4f32 (v2f64 (X86vzload64 addr:$src))))),
             (VCVTTPS2UQQZ128rm addr:$src)>;
   def : Pat<(v2i64 (vselect VK2WM:$mask,
                             (X86cvttp2ui (bc_v4f32 (v2f64 (X86vzload64 addr:$src)))),
                             VR128X:$src0)),
             (VCVTTPS2UQQZ128rmk VR128X:$src0, VK2WM:$mask, addr:$src)>;
   def : Pat<(v2i64 (vselect VK2WM:$mask,
                             (X86cvttp2ui (bc_v4f32 (v2f64 (X86vzload64 addr:$src)))),
                             v2i64x_info.ImmAllZerosV)),
             (VCVTTPS2UQQZ128rmkz VK2WM:$mask, addr:$src)>;
 }
 
 let Predicates = [HasAVX512, NoVLX] in {
-def : Pat<(v8i32 (X86cvttp2ui (v8f32 VR256X:$src1))),
+def : Pat<(v8i32 (X86any_cvttp2ui (v8f32 VR256X:$src1))),
           (EXTRACT_SUBREG (v16i32 (VCVTTPS2UDQZrr
            (v16f32 (INSERT_SUBREG (IMPLICIT_DEF),
                                   VR256X:$src1, sub_ymm)))), sub_ymm)>;
 
-def : Pat<(v4i32 (X86cvttp2ui (v4f32 VR128X:$src1))),
+def : Pat<(v4i32 (X86any_cvttp2ui (v4f32 VR128X:$src1))),
           (EXTRACT_SUBREG (v16i32 (VCVTTPS2UDQZrr
            (v16f32 (INSERT_SUBREG (IMPLICIT_DEF),
                                   VR128X:$src1, sub_xmm)))), sub_xmm)>;
 
-def : Pat<(v4i32 (X86cvttp2ui (v4f64 VR256X:$src1))),
+def : Pat<(v4i32 (X86any_cvttp2ui (v4f64 VR256X:$src1))),
           (EXTRACT_SUBREG (v8i32 (VCVTTPD2UDQZrr
            (v8f64 (INSERT_SUBREG (IMPLICIT_DEF),
                                  VR256X:$src1, sub_ymm)))), sub_xmm)>;
 
 def : Pat<(v8f32 (any_uint_to_fp (v8i32 VR256X:$src1))),
           (EXTRACT_SUBREG (v16f32 (VCVTUDQ2PSZrr
            (v16i32 (INSERT_SUBREG (IMPLICIT_DEF),
                                   VR256X:$src1, sub_ymm)))), sub_ymm)>;
def : Pat<(X86VMUintToFP (v2i64 (X86VBroadcastld64 addr:$src)),
                            (v4f32 VR128X:$src0), VK2WM:$mask),
             (VCVTUQQ2PSZ128rmbk VR128X:$src0, VK2WM:$mask, addr:$src)>;
   def : Pat<(X86VMUintToFP (v2i64 (X86VBroadcastld64 addr:$src)),
                            v4f32x_info.ImmAllZerosV, VK2WM:$mask),
             (VCVTUQQ2PSZ128rmbkz VK2WM:$mask, addr:$src)>;
 }
 
 let Predicates = [HasDQI, NoVLX] in {
-def : Pat<(v2i64 (X86cvttp2si (v2f64 VR128X:$src1))),
+def : Pat<(v2i64 (X86any_cvttp2si (v2f64 VR128X:$src1))),
           (EXTRACT_SUBREG (v8i64 (VCVTTPD2QQZrr
            (v8f64 (INSERT_SUBREG (IMPLICIT_DEF),
                                   VR128X:$src1, sub_xmm)))), sub_xmm)>;
 
-def : Pat<(v4i64 (X86cvttp2si (v4f32 VR128X:$src1))),
+def : Pat<(v4i64 (X86any_cvttp2si (v4f32 VR128X:$src1))),
           (EXTRACT_SUBREG (v8i64 (VCVTTPS2QQZrr
            (v8f32 (INSERT_SUBREG (IMPLICIT_DEF),
                                   VR128X:$src1, sub_xmm)))), sub_ymm)>;
 
-def : Pat<(v4i64 (X86cvttp2si (v4f64 VR256X:$src1))),
+def : Pat<(v4i64 (X86any_cvttp2si (v4f64 VR256X:$src1))),
           (EXTRACT_SUBREG (v8i64 (VCVTTPD2QQZrr
            (v8f64 (INSERT_SUBREG (IMPLICIT_DEF),
                                   VR256X:$src1, sub_ymm)))), sub_ymm)>;
 
-def : Pat<(v2i64 (X86cvttp2ui (v2f64 VR128X:$src1))),
+def : Pat<(v2i64 (X86any_cvttp2ui (v2f64 VR128X:$src1))),
           (EXTRACT_SUBREG (v8i64 (VCVTTPD2UQQZrr
            (v8f64 (INSERT_SUBREG (IMPLICIT_DEF),
                                   VR128X:$src1, sub_xmm)))), sub_xmm)>;
 
-def : Pat<(v4i64 (X86cvttp2ui (v4f32 VR128X:$src1))),
+def : Pat<(v4i64 (X86any_cvttp2ui (v4f32 VR128X:$src1))),
           (EXTRACT_SUBREG (v8i64 (VCVTTPS2UQQZrr
            (v8f32 (INSERT_SUBREG (IMPLICIT_DEF),
                                   VR128X:$src1, sub_xmm)))), sub_ymm)>;
 
-def : Pat<(v4i64 (X86cvttp2ui (v4f64 VR256X:$src1))),
+def : Pat<(v4i64 (X86any_cvttp2ui (v4f64 VR256X:$src1))),
           (EXTRACT_SUBREG (v8i64 (VCVTTPD2UQQZrr
            (v8f64 (INSERT_SUBREG (IMPLICIT_DEF),
                                   VR256X:$src1, sub_ymm)))), sub_ymm)>;
 
 def : Pat<(v4f32 (any_sint_to_fp (v4i64 VR256X:$src1))),
           (EXTRACT_SUBREG (v8f32 (VCVTQQ2PSZrr
            (v8i64 (INSERT_SUBREG (IMPLICIT_DEF),
                                   VR256X:$src1, sub_ymm)))), sub_xmm)>;

llvm/lib/Target/X86/X86InstrFragmentsSIMD.td

def X86vmfpround: SDNode<"X86ISD::VMFPROUND",
                                               SDTCVecEltisVT<3, i1>,
                                               SDTCisSameNumEltsAs<1, 3>]>>;
 
 def X86vshiftimm : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
                                         SDTCisVT<2, i8>, SDTCisInt<0>]>;
 
 def X86vshldq  : SDNode<"X86ISD::VSHLDQ",    X86vshiftimm>;
 def X86vshrdq  : SDNode<"X86ISD::VSRLDQ",    X86vshiftimm>;
-def X86cmpp    : SDNode<"X86ISD::CMPP",      SDTX86VFCMP, [SDNPHasChain]>;
 def X86pcmpeq  : SDNode<"X86ISD::PCMPEQ", SDTIntBinOp, [SDNPCommutative]>;
 def X86pcmpgt  : SDNode<"X86ISD::PCMPGT", SDTIntBinOp>;
 
+def X86cmpp    : SDNode<"X86ISD::CMPP",      SDTX86VFCMP>;
+def X86strict_cmpp : SDNode<"X86ISD::STRICT_CMPP", SDTX86VFCMP, [SDNPHasChain]>;
+def X86any_cmpp    : PatFrags<(ops node:$src1, node:$src2, node:$src3),
+                               [(X86strict_cmpp node:$src1, node:$src2, node:$src3),
+                                (X86cmpp node:$src1, node:$src2, node:$src3)]>;
+
 def X86CmpMaskCC :
       SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCVecEltisVT<0, i1>,
                        SDTCisVec<1>, SDTCisSameAs<2, 1>,
                        SDTCisSameNumEltsAs<0, 1>, SDTCisVT<3, i8>]>;
 def X86CmpMaskCCScalar :
       SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisFP<1>, SDTCisSameAs<1, 2>,
                            SDTCisVT<3, i8>]>;
 
-def X86cmpm     : SDNode<"X86ISD::CMPM",     X86CmpMaskCC, [SDNPHasChain]>;
+def X86cmpm     : SDNode<"X86ISD::CMPM",     X86CmpMaskCC>;
+def X86strict_cmpm : SDNode<"X86ISD::STRICT_CMPM", X86CmpMaskCC, [SDNPHasChain]>;
+def X86any_cmpm    : PatFrags<(ops node:$src1, node:$src2, node:$src3),
+                               [(X86strict_cmpm node:$src1, node:$src2, node:$src3),
+                                (X86cmpm node:$src1, node:$src2, node:$src3)]>;
 def X86cmpmSAE  : SDNode<"X86ISD::CMPM_SAE", X86CmpMaskCC>;
 def X86cmpms    : SDNode<"X86ISD::FSETCCM",   X86CmpMaskCCScalar>;
 def X86cmpmsSAE : SDNode<"X86ISD::FSETCCM_SAE",   X86CmpMaskCCScalar>;
 
 def X86phminpos: SDNode<"X86ISD::PHMINPOS", 
                  SDTypeProfile<1, 1, [SDTCisVT<0, v8i16>, SDTCisVT<1, v8i16>]>>;
 
 def X86vshiftuniform : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,

 
 // cvt fp-to-int staff
 def X86cvtp2IntRnd      : SDNode<"X86ISD::CVTP2SI_RND",  SDTFloatToIntRnd>;
 def X86cvtp2UIntRnd     : SDNode<"X86ISD::CVTP2UI_RND",  SDTFloatToIntRnd>;
 
 // Vector without rounding mode
 
 // cvtt fp-to-int staff
-def X86cvttp2si      : SDNode<"X86ISD::CVTTP2SI",  SDTFloatToInt, [SDNPHasChain]>;
-def X86cvttp2ui      : SDNode<"X86ISD::CVTTP2UI",  SDTFloatToInt, [SDNPHasChain]>;
+def X86cvttp2si      : SDNode<"X86ISD::CVTTP2SI",  SDTFloatToInt>;
+def X86cvttp2ui      : SDNode<"X86ISD::CVTTP2UI",  SDTFloatToInt>;
+def X86strict_cvttp2si : SDNode<"X86ISD::STRICT_CVTTP2SI",  SDTFloatToInt, [SDNPHasChain]>;
+def X86strict_cvttp2ui : SDNode<"X86ISD::STRICT_CVTTP2UI",  SDTFloatToInt, [SDNPHasChain]>;
+def X86any_cvttp2si : PatFrags<(ops node:$src),
+                               [(X86strict_cvttp2si node:$src),
+                                (X86cvttp2si node:$src)]>;
+def X86any_cvttp2ui : PatFrags<(ops node:$src),
+                               [(X86strict_cvttp2ui node:$src),
+                                (X86cvttp2ui node:$src)]>;
 
 def X86VSintToFP      : SDNode<"X86ISD::CVTSI2P",  SDTVintToFP>;
 def X86VUintToFP      : SDNode<"X86ISD::CVTUI2P",  SDTVintToFP>;
 
 // cvt int-to-fp staff
 def X86cvtp2Int      : SDNode<"X86ISD::CVTP2SI",  SDTFloatToInt>;
 def X86cvtp2UInt     : SDNode<"X86ISD::CVTP2UI",  SDTFloatToInt>;
 

llvm/lib/Target/X86/X86InstrSSE.td

 
 // Convert with truncation packed single/double fp to doubleword
 // SSE2 packed instructions with XS prefix
 let Uses = [MXCSR], mayRaiseFPException = 1 in {
 let Predicates = [HasAVX, NoVLX] in {
 def VCVTTPS2DQrr : VS2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                          "cvttps2dq\t{$src, $dst|$dst, $src}",
                          [(set VR128:$dst,
-                           (v4i32 (X86cvttp2si (v4f32 VR128:$src))))]>,
+                           (v4i32 (X86any_cvttp2si (v4f32 VR128:$src))))]>,
                          VEX, Sched<[WriteCvtPS2I]>, VEX_WIG;
 def VCVTTPS2DQrm : VS2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                          "cvttps2dq\t{$src, $dst|$dst, $src}",
                          [(set VR128:$dst,
-                           (v4i32 (X86cvttp2si (loadv4f32 addr:$src))))]>,
+                           (v4i32 (X86any_cvttp2si (loadv4f32 addr:$src))))]>,
                          VEX, Sched<[WriteCvtPS2ILd]>, VEX_WIG;
 def VCVTTPS2DQYrr : VS2SI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
                           "cvttps2dq\t{$src, $dst|$dst, $src}",
                           [(set VR256:$dst,
-                            (v8i32 (X86cvttp2si (v8f32 VR256:$src))))]>,
+                            (v8i32 (X86any_cvttp2si (v8f32 VR256:$src))))]>,
                           VEX, VEX_L, Sched<[WriteCvtPS2IY]>, VEX_WIG;
 def VCVTTPS2DQYrm : VS2SI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
                           "cvttps2dq\t{$src, $dst|$dst, $src}",
                           [(set VR256:$dst,
-                            (v8i32 (X86cvttp2si (loadv8f32 addr:$src))))]>,
+                            (v8i32 (X86any_cvttp2si (loadv8f32 addr:$src))))]>,
                           VEX, VEX_L,
                           Sched<[WriteCvtPS2IYLd]>, VEX_WIG;
 }
 
 def CVTTPS2DQrr : S2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                        "cvttps2dq\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst,
-                         (v4i32 (X86cvttp2si (v4f32 VR128:$src))))]>,
+                         (v4i32 (X86any_cvttp2si (v4f32 VR128:$src))))]>,
                        Sched<[WriteCvtPS2I]>;
 def CVTTPS2DQrm : S2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                        "cvttps2dq\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst,
-                         (v4i32 (X86cvttp2si (memopv4f32 addr:$src))))]>,
+                         (v4i32 (X86any_cvttp2si (memopv4f32 addr:$src))))]>,
                        Sched<[WriteCvtPS2ILd]>;
 }
 
 // The assembler can recognize rr 256-bit instructions by seeing a ymm
 // register, but the same isn't true when using memory operands instead.
 // Provide other assembly rr and rm forms to address this explicitly.
 let Predicates = [HasAVX, NoVLX], Uses = [MXCSR], mayRaiseFPException = 1 in {
 // XMM only
 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                         "cvttpd2dq\t{$src, $dst|$dst, $src}",
                         [(set VR128:$dst,
-                          (v4i32 (X86cvttp2si (v2f64 VR128:$src))))]>,
+                          (v4i32 (X86any_cvttp2si (v2f64 VR128:$src))))]>,
                         VEX, Sched<[WriteCvtPD2I]>, VEX_WIG;
 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                         "cvttpd2dq{x}\t{$src, $dst|$dst, $src}",
                         [(set VR128:$dst,
-                          (v4i32 (X86cvttp2si (loadv2f64 addr:$src))))]>,
+                          (v4i32 (X86any_cvttp2si (loadv2f64 addr:$src))))]>,
                         VEX, Sched<[WriteCvtPD2ILd]>, VEX_WIG;
 
 // YMM only
 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
                          "cvttpd2dq\t{$src, $dst|$dst, $src}",
                          [(set VR128:$dst,
-                           (v4i32 (X86cvttp2si (v4f64 VR256:$src))))]>,
+                           (v4i32 (X86any_cvttp2si (v4f64 VR256:$src))))]>,
                          VEX, VEX_L, Sched<[WriteCvtPD2IY]>, VEX_WIG;
 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
                          "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
                          [(set VR128:$dst,
-                           (v4i32 (X86cvttp2si (loadv4f64 addr:$src))))]>,
+                           (v4i32 (X86any_cvttp2si (loadv4f64 addr:$src))))]>,
                          VEX, VEX_L, Sched<[WriteCvtPD2IYLd]>, VEX_WIG;
 } // Predicates = [HasAVX, NoVLX]
 
 def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
                 (VCVTTPD2DQrr VR128:$dst, VR128:$src), 0, "att">;
 def : InstAlias<"vcvttpd2dqy\t{$src, $dst|$dst, $src}",
                 (VCVTTPD2DQYrr VR128:$dst, VR256:$src), 0, "att">;
 
 let Predicates = [HasAVX, NoVLX] in {
   def : Pat<(v4i32 (any_fp_to_sint (v4f64 VR256:$src))),
             (VCVTTPD2DQYrr VR256:$src)>;
   def : Pat<(v4i32 (any_fp_to_sint (loadv4f64 addr:$src))),
             (VCVTTPD2DQYrm addr:$src)>;
 }
 
 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                       "cvttpd2dq\t{$src, $dst|$dst, $src}",
                       [(set VR128:$dst,
-                        (v4i32 (X86cvttp2si (v2f64 VR128:$src))))]>,
+                        (v4i32 (X86any_cvttp2si (v2f64 VR128:$src))))]>,
                       Sched<[WriteCvtPD2I]>, SIMD_EXC;
 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
                       "cvttpd2dq\t{$src, $dst|$dst, $src}",
                       [(set VR128:$dst,
-                        (v4i32 (X86cvttp2si (memopv2f64 addr:$src))))]>,
+                        (v4i32 (X86any_cvttp2si (memopv2f64 addr:$src))))]>,
                       Sched<[WriteCvtPD2ILd]>, SIMD_EXC;
 
 // Convert packed single to packed double
 let Predicates = [HasAVX, NoVLX], Uses = [MXCSR], mayRaiseFPException = 1 in {
                   // SSE2 instructions without OpSize prefix
 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                     "vcvtps2pd\t{$src, $dst|$dst, $src}",
                     [(set VR128:$dst, (v2f64 (X86vfpext (v4f32 VR128:$src))))]>,

 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
                             ValueType VT, string asm,
                             X86FoldableSchedWrite sched,
                             Domain d, PatFrag ld_frag> {
 let Uses = [MXCSR], mayRaiseFPException = 1 in {
   let isCommutable = 1 in
   def rri : PIi8<0xC2, MRMSrcReg,
              (outs RC:$dst), (ins RC:$src1, RC:$src2, u8imm:$cc), asm,
-             [(set RC:$dst, (VT (X86cmpp RC:$src1, RC:$src2, timm:$cc)))], d>,
+             [(set RC:$dst, (VT (X86any_cmpp RC:$src1, RC:$src2, timm:$cc)))], d>,
             Sched<[sched]>;
   def rmi : PIi8<0xC2, MRMSrcMem,
              (outs RC:$dst), (ins RC:$src1, x86memop:$src2, u8imm:$cc), asm,
              [(set RC:$dst,
-               (VT (X86cmpp RC:$src1, (ld_frag addr:$src2), timm:$cc)))], d>,
+               (VT (X86any_cmpp RC:$src1, (ld_frag addr:$src2), timm:$cc)))], d>,
             Sched<[sched.Folded, sched.ReadAfterFold]>;
 }
 }
 
 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, v4f32,
                "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
                SchedWriteFCmpSizes.PS.XMM, SSEPackedSingle, loadv4f32>, PS, VEX_4V, VEX_WIG;
 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, v2f64,

 
 def CommutableCMPCC : PatLeaf<(timm), [{
   uint64_t Imm = N->getZExtValue() & 0x7;
   return (Imm == 0x00 || Imm == 0x03 || Imm == 0x04 || Imm == 0x07);
 }]>;
 
 // Patterns to select compares with loads in first operand.
 let Predicates = [HasAVX] in {
-  def : Pat<(v4f64 (X86cmpp (loadv4f64 addr:$src2), VR256:$src1,
+  def : Pat<(v4f64 (X86any_cmpp (loadv4f64 addr:$src2), VR256:$src1,
                                 CommutableCMPCC:$cc)),
             (VCMPPDYrmi VR256:$src1, addr:$src2, timm:$cc)>;
 
-  def : Pat<(v8f32 (X86cmpp (loadv8f32 addr:$src2), VR256:$src1,
+  def : Pat<(v8f32 (X86any_cmpp (loadv8f32 addr:$src2), VR256:$src1,
                                 CommutableCMPCC:$cc)),
             (VCMPPSYrmi VR256:$src1, addr:$src2, timm:$cc)>;
 
-  def : Pat<(v2f64 (X86cmpp (loadv2f64 addr:$src2), VR128:$src1,
+  def : Pat<(v2f64 (X86any_cmpp (loadv2f64 addr:$src2), VR128:$src1,
                                 CommutableCMPCC:$cc)),
             (VCMPPDrmi VR128:$src1, addr:$src2, timm:$cc)>;
 
-  def : Pat<(v4f32 (X86cmpp (loadv4f32 addr:$src2), VR128:$src1,
+  def : Pat<(v4f32 (X86any_cmpp (loadv4f32 addr:$src2), VR128:$src1,
                                 CommutableCMPCC:$cc)),
             (VCMPPSrmi VR128:$src1, addr:$src2, timm:$cc)>;
 
   def : Pat<(f64 (X86cmps (loadf64 addr:$src2), FR64:$src1,
                           CommutableCMPCC:$cc)),
             (VCMPSDrm FR64:$src1, addr:$src2, timm:$cc)>;
 
   def : Pat<(f32 (X86cmps (loadf32 addr:$src2), FR32:$src1,
                           CommutableCMPCC:$cc)),
             (VCMPSSrm FR32:$src1, addr:$src2, timm:$cc)>;
 }
 
 let Predicates = [UseSSE2] in {
-  def : Pat<(v2f64 (X86cmpp (memopv2f64 addr:$src2), VR128:$src1,
+  def : Pat<(v2f64 (X86any_cmpp (memopv2f64 addr:$src2), VR128:$src1,
                                 CommutableCMPCC:$cc)),
             (CMPPDrmi VR128:$src1, addr:$src2, timm:$cc)>;
 
   def : Pat<(f64 (X86cmps (loadf64 addr:$src2), FR64:$src1,
                           CommutableCMPCC:$cc)),
             (CMPSDrm FR64:$src1, addr:$src2, timm:$cc)>;
 }
 
 let Predicates = [UseSSE1] in {
-  def : Pat<(v4f32 (X86cmpp (memopv4f32 addr:$src2), VR128:$src1,
+  def : Pat<(v4f32 (X86any_cmpp (memopv4f32 addr:$src2), VR128:$src1,
                                 CommutableCMPCC:$cc)),
             (CMPPSrmi VR128:$src1, addr:$src2, timm:$cc)>;
 
   def : Pat<(f32 (X86cmps (loadf32 addr:$src2), FR32:$src1,
                           CommutableCMPCC:$cc)),
             (CMPSSrm FR32:$src1, addr:$src2, timm:$cc)>;
 }
 
 //===----------------------------------------------------------------------===//

llvm/test/CodeGen/X86/avx512-intrinsics.ll

; X86-NEXT:    retl
   ret <16 x i32> %res2
 }
 
 declare <8 x i32> @llvm.x86.avx512.mask.cvttpd2dq.512(<8 x double>, <8 x i32>, i8, i32)
 
 define <8 x i32>@test_int_x86_avx512_mask_cvtt_pd2dq_512(<8 x double> %x0, <8 x i32> %x1, i8 %x2) {
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_pd2dq_512:
 ; X64:       # %bb.0:
-; X64-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vcvttpd2dq %zmm0, %ymm2
-; X64-NEXT:    vmovdqa32 %zmm2, %zmm1 {%k1}
+; X64-NEXT:    vcvttpd2dq %zmm0, %ymm1 {%k1}
 ; X64-NEXT:    vcvttpd2dq {sae}, %zmm0, %ymm0
 ; X64-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
 ; X64-NEXT:    retq
 ;
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_pd2dq_512:
 ; X86:       # %bb.0:
-; X86-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vcvttpd2dq %zmm0, %ymm2
-; X86-NEXT:    vmovdqa32 %zmm2, %zmm1 {%k1}
+; X86-NEXT:    vcvttpd2dq %zmm0, %ymm1 {%k1}
 ; X86-NEXT:    vcvttpd2dq {sae}, %zmm0, %ymm0
 ; X86-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
 ; X86-NEXT:    retl
   %res = call <8 x i32> @llvm.x86.avx512.mask.cvttpd2dq.512(<8 x double> %x0, <8 x i32> %x1, i8 %x2, i32 4)
   %res1 = call <8 x i32> @llvm.x86.avx512.mask.cvttpd2dq.512(<8 x double> %x0, <8 x i32> %x1, i8 -1, i32 8)
   %res2 = add <8 x i32> %res, %res1
   ret <8 x i32> %res2
 }
; X86-NEXT:    retl
   ret <16 x float> %res2
 }
 
 declare <8 x i32> @llvm.x86.avx512.mask.cvttpd2udq.512(<8 x double>, <8 x i32>, i8, i32)
 
 define <8 x i32>@test_int_x86_avx512_mask_cvtt_pd2udq_512(<8 x double> %x0, <8 x i32> %x1, i8 %x2) {
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_pd2udq_512:
 ; X64:       # %bb.0:
-; X64-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vcvttpd2udq %zmm0, %ymm2
-; X64-NEXT:    vmovdqa32 %zmm2, %zmm1 {%k1}
+; X64-NEXT:    vcvttpd2udq %zmm0, %ymm1 {%k1}
 ; X64-NEXT:    vcvttpd2udq {sae}, %zmm0, %ymm0
 ; X64-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
 ; X64-NEXT:    retq
 ;
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_pd2udq_512:
 ; X86:       # %bb.0:
-; X86-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vcvttpd2udq %zmm0, %ymm2
-; X86-NEXT:    vmovdqa32 %zmm2, %zmm1 {%k1}
+; X86-NEXT:    vcvttpd2udq %zmm0, %ymm1 {%k1}
 ; X86-NEXT:    vcvttpd2udq {sae}, %zmm0, %ymm0
 ; X86-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
 ; X86-NEXT:    retl
   %res = call <8 x i32> @llvm.x86.avx512.mask.cvttpd2udq.512(<8 x double> %x0, <8 x i32> %x1, i8 %x2, i32 4)
   %res1 = call <8 x i32> @llvm.x86.avx512.mask.cvttpd2udq.512(<8 x double> %x0, <8 x i32> %x1, i8 -1, i32 8)
   %res2 = add <8 x i32> %res, %res1
   ret <8 x i32> %res2
 }
 
 declare <16 x i32> @llvm.x86.avx512.mask.cvttps2dq.512(<16 x float>, <16 x i32>, i16, i32)
 
 define <16 x i32>@test_int_x86_avx512_mask_cvtt_ps2dq_512(<16 x float> %x0, <16 x i32> %x1, i16 %x2) {
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2dq_512:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vcvttps2dq %zmm0, %zmm2
-; X64-NEXT:    vmovdqa32 %zmm2, %zmm1 {%k1}
+; X64-NEXT:    vcvttps2dq %zmm0, %zmm1 {%k1}
 ; X64-NEXT:    vcvttps2dq {sae}, %zmm0, %zmm0
 ; X64-NEXT:    vpaddd %zmm0, %zmm1, %zmm0
 ; X64-NEXT:    retq
 ;
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2dq_512:
 ; X86:       # %bb.0:
 ; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
-; X86-NEXT:    vcvttps2dq %zmm0, %zmm2
-; X86-NEXT:    vmovdqa32 %zmm2, %zmm1 {%k1}
+; X86-NEXT:    vcvttps2dq %zmm0, %zmm1 {%k1}
 ; X86-NEXT:    vcvttps2dq {sae}, %zmm0, %zmm0
 ; X86-NEXT:    vpaddd %zmm0, %zmm1, %zmm0
 ; X86-NEXT:    retl
   %res = call <16 x i32> @llvm.x86.avx512.mask.cvttps2dq.512(<16 x float> %x0, <16 x i32> %x1, i16 %x2, i32 4)
   %res1 = call <16 x i32> @llvm.x86.avx512.mask.cvttps2dq.512(<16 x float> %x0, <16 x i32> %x1, i16 -1, i32 8)
   %res2 = add <16 x i32> %res, %res1
   ret <16 x i32> %res2
 }
 
 declare <16 x i32> @llvm.x86.avx512.mask.cvttps2udq.512(<16 x float>, <16 x i32>, i16, i32)
 
 define <16 x i32>@test_int_x86_avx512_mask_cvtt_ps2udq_512(<16 x float> %x0, <16 x i32> %x1, i16 %x2) {
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2udq_512:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vcvttps2udq %zmm0, %zmm2
-; X64-NEXT:    vmovdqa32 %zmm2, %zmm1 {%k1}
+; X64-NEXT:    vcvttps2udq %zmm0, %zmm1 {%k1}
 ; X64-NEXT:    vcvttps2udq {sae}, %zmm0, %zmm0
 ; X64-NEXT:    vpaddd %zmm0, %zmm1, %zmm0
 ; X64-NEXT:    retq
 ;
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2udq_512:
 ; X86:       # %bb.0:
 ; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
-; X86-NEXT:    vcvttps2udq %zmm0, %zmm2
-; X86-NEXT:    vmovdqa32 %zmm2, %zmm1 {%k1}
+; X86-NEXT:    vcvttps2udq %zmm0, %zmm1 {%k1}
 ; X86-NEXT:    vcvttps2udq {sae}, %zmm0, %zmm0
 ; X86-NEXT:    vpaddd %zmm0, %zmm1, %zmm0
 ; X86-NEXT:    retl
   %res = call <16 x i32> @llvm.x86.avx512.mask.cvttps2udq.512(<16 x float> %x0, <16 x i32> %x1, i16 %x2, i32 4)
   %res1 = call <16 x i32> @llvm.x86.avx512.mask.cvttps2udq.512(<16 x float> %x0, <16 x i32> %x1, i16 -1, i32 8)
   %res2 = add <16 x i32> %res, %res1
   ret <16 x i32> %res2
 }

llvm/test/CodeGen/X86/avx512dq-intrinsics.ll

 }
 
 declare <8 x i64> @llvm.x86.avx512.mask.cvttpd2qq.512(<8 x double>, <8 x i64>, i8, i32)
 
 define <8 x i64>@test_int_x86_avx512_mask_cvtt_pd2qq_512(<8 x double> %x0, <8 x i64> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_pd2qq_512:
 ; X86:       # %bb.0:
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vcvttpd2qq %zmm0, %zmm2 # encoding: [0x62,0xf1,0xfd,0x48,0x7a,0xd0]
-; X86-NEXT:    vmovdqa64 %zmm2, %zmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x49,0x6f,0xca]
+; X86-NEXT:    vcvttpd2qq %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x49,0x7a,0xc8]
 ; X86-NEXT:    vcvttpd2qq {sae}, %zmm0, %zmm0 # encoding: [0x62,0xf1,0xfd,0x18,0x7a,0xc0]
 ; X86-NEXT:    vpaddq %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0xf5,0x48,0xd4,0xc0]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_pd2qq_512:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
-; X64-NEXT:    vcvttpd2qq %zmm0, %zmm2 # encoding: [0x62,0xf1,0xfd,0x48,0x7a,0xd0]
-; X64-NEXT:    vmovdqa64 %zmm2, %zmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x49,0x6f,0xca]
+; X64-NEXT:    vcvttpd2qq %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x49,0x7a,0xc8]
 ; X64-NEXT:    vcvttpd2qq {sae}, %zmm0, %zmm0 # encoding: [0x62,0xf1,0xfd,0x18,0x7a,0xc0]
 ; X64-NEXT:    vpaddq %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0xf5,0x48,0xd4,0xc0]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <8 x i64> @llvm.x86.avx512.mask.cvttpd2qq.512(<8 x double> %x0, <8 x i64> %x1, i8 %x2, i32 4)
   %res1 = call <8 x i64> @llvm.x86.avx512.mask.cvttpd2qq.512(<8 x double> %x0, <8 x i64> %x1, i8 -1, i32 8)
   %res2 = add <8 x i64> %res, %res1
   ret <8 x i64> %res2
 }
 
 declare <8 x i64> @llvm.x86.avx512.mask.cvttpd2uqq.512(<8 x double>, <8 x i64>, i8, i32)
 
 define <8 x i64>@test_int_x86_avx512_mask_cvtt_pd2uqq_512(<8 x double> %x0, <8 x i64> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_pd2uqq_512:
 ; X86:       # %bb.0:
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vcvttpd2uqq %zmm0, %zmm2 # encoding: [0x62,0xf1,0xfd,0x48,0x78,0xd0]
-; X86-NEXT:    vmovdqa64 %zmm2, %zmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x49,0x6f,0xca]
+; X86-NEXT:    vcvttpd2uqq %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x49,0x78,0xc8]
 ; X86-NEXT:    vcvttpd2uqq {sae}, %zmm0, %zmm0 # encoding: [0x62,0xf1,0xfd,0x18,0x78,0xc0]
 ; X86-NEXT:    vpaddq %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0xf5,0x48,0xd4,0xc0]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_pd2uqq_512:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
-; X64-NEXT:    vcvttpd2uqq %zmm0, %zmm2 # encoding: [0x62,0xf1,0xfd,0x48,0x78,0xd0]
-; X64-NEXT:    vmovdqa64 %zmm2, %zmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x49,0x6f,0xca]
+; X64-NEXT:    vcvttpd2uqq %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x49,0x78,0xc8]
 ; X64-NEXT:    vcvttpd2uqq {sae}, %zmm0, %zmm0 # encoding: [0x62,0xf1,0xfd,0x18,0x78,0xc0]
 ; X64-NEXT:    vpaddq %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0xf5,0x48,0xd4,0xc0]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <8 x i64> @llvm.x86.avx512.mask.cvttpd2uqq.512(<8 x double> %x0, <8 x i64> %x1, i8 %x2, i32 4)
   %res1 = call <8 x i64> @llvm.x86.avx512.mask.cvttpd2uqq.512(<8 x double> %x0, <8 x i64> %x1, i8 -1, i32 8)
   %res2 = add <8 x i64> %res, %res1
   ret <8 x i64> %res2
 }
 
 declare <8 x i64> @llvm.x86.avx512.mask.cvttps2qq.512(<8 x float>, <8 x i64>, i8, i32)
 
 define <8 x i64>@test_int_x86_avx512_mask_cvtt_ps2qq_512(<8 x float> %x0, <8 x i64> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2qq_512:
 ; X86:       # %bb.0:
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vcvttps2qq %ymm0, %zmm2 # encoding: [0x62,0xf1,0x7d,0x48,0x7a,0xd0]
-; X86-NEXT:    vmovdqa64 %zmm2, %zmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x49,0x6f,0xca]
+; X86-NEXT:    vcvttps2qq %ymm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x7a,0xc8]
 ; X86-NEXT:    vcvttps2qq {sae}, %ymm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x18,0x7a,0xc0]
 ; X86-NEXT:    vpaddq %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0xf5,0x48,0xd4,0xc0]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2qq_512:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
-; X64-NEXT:    vcvttps2qq %ymm0, %zmm2 # encoding: [0x62,0xf1,0x7d,0x48,0x7a,0xd0]
-; X64-NEXT:    vmovdqa64 %zmm2, %zmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x49,0x6f,0xca]
+; X64-NEXT:    vcvttps2qq %ymm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x7a,0xc8]
 ; X64-NEXT:    vcvttps2qq {sae}, %ymm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x18,0x7a,0xc0]
 ; X64-NEXT:    vpaddq %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0xf5,0x48,0xd4,0xc0]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <8 x i64> @llvm.x86.avx512.mask.cvttps2qq.512(<8 x float> %x0, <8 x i64> %x1, i8 %x2, i32 4)
   %res1 = call <8 x i64> @llvm.x86.avx512.mask.cvttps2qq.512(<8 x float> %x0, <8 x i64> %x1, i8 -1, i32 8)
   %res2 = add <8 x i64> %res, %res1
   ret <8 x i64> %res2
 }
 
 declare <8 x i64> @llvm.x86.avx512.mask.cvttps2uqq.512(<8 x float>, <8 x i64>, i8, i32)
 
 define <8 x i64>@test_int_x86_avx512_mask_cvtt_ps2uqq_512(<8 x float> %x0, <8 x i64> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2uqq_512:
 ; X86:       # %bb.0:
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vcvttps2uqq %ymm0, %zmm2 # encoding: [0x62,0xf1,0x7d,0x48,0x78,0xd0]
-; X86-NEXT:    vmovdqa64 %zmm2, %zmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x49,0x6f,0xca]
+; X86-NEXT:    vcvttps2uqq %ymm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x78,0xc8]
 ; X86-NEXT:    vcvttps2uqq {sae}, %ymm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x18,0x78,0xc0]
 ; X86-NEXT:    vpaddq %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0xf5,0x48,0xd4,0xc0]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2uqq_512:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
-; X64-NEXT:    vcvttps2uqq %ymm0, %zmm2 # encoding: [0x62,0xf1,0x7d,0x48,0x78,0xd0]
-; X64-NEXT:    vmovdqa64 %zmm2, %zmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x49,0x6f,0xca]
+; X64-NEXT:    vcvttps2uqq %ymm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x78,0xc8]
 ; X64-NEXT:    vcvttps2uqq {sae}, %ymm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x18,0x78,0xc0]
 ; X64-NEXT:    vpaddq %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0xf5,0x48,0xd4,0xc0]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <8 x i64> @llvm.x86.avx512.mask.cvttps2uqq.512(<8 x float> %x0, <8 x i64> %x1, i8 %x2, i32 4)
   %res1 = call <8 x i64> @llvm.x86.avx512.mask.cvttps2uqq.512(<8 x float> %x0, <8 x i64> %x1, i8 -1, i32 8)
   %res2 = add <8 x i64> %res, %res1
   ret <8 x i64> %res2
 }

llvm/test/CodeGen/X86/avx512dqvl-intrinsics.ll

; X64-NEXT:    retq # encoding: [0xc3]
   ret <4 x float> %res2
 }
 
 declare <2 x i64> @llvm.x86.avx512.mask.cvttpd2qq.128(<2 x double>, <2 x i64>, i8)
 
 define <2 x i64>@test_int_x86_avx512_mask_cvtt_pd2qq_128(<2 x double> %x0, <2 x i64> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_pd2qq_128:
 ; X86:       # %bb.0:
-; X86-NEXT:    vcvttpd2qq %xmm0, %xmm0 # encoding: [0x62,0xf1,0xfd,0x08,0x7a,0xc0]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc8]
+; X86-NEXT:    vcvttpd2qq %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x7a,0xc8]
+; X86-NEXT:    vcvttpd2qq %xmm0, %xmm0 # encoding: [0x62,0xf1,0xfd,0x08,0x7a,0xc0]
 ; X86-NEXT:    vpaddq %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xd4,0xc0]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_pd2qq_128:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vcvttpd2qq %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x7a,0xc8]
 ; X64-NEXT:    vcvttpd2qq %xmm0, %xmm0 # encoding: [0x62,0xf1,0xfd,0x08,0x7a,0xc0]
-; X64-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc8]
 ; X64-NEXT:    vpaddq %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xd4,0xc0]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttpd2qq.128(<2 x double> %x0, <2 x i64> %x1, i8 %x2)
   %res1 = call <2 x i64> @llvm.x86.avx512.mask.cvttpd2qq.128(<2 x double> %x0, <2 x i64> %x1, i8 -1)
   %res2 = add <2 x i64> %res, %res1
   ret <2 x i64> %res2
 }
 
 declare <4 x i64> @llvm.x86.avx512.mask.cvttpd2qq.256(<4 x double>, <4 x i64>, i8)
 
 define <4 x i64>@test_int_x86_avx512_mask_cvtt_pd2qq_256(<4 x double> %x0, <4 x i64> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_pd2qq_256:
 ; X86:       # %bb.0:
-; X86-NEXT:    vcvttpd2qq %ymm0, %ymm0 # encoding: [0x62,0xf1,0xfd,0x28,0x7a,0xc0]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vmovdqa64 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x29,0x6f,0xc8]
+; X86-NEXT:    vcvttpd2qq %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x29,0x7a,0xc8]
+; X86-NEXT:    vcvttpd2qq %ymm0, %ymm0 # encoding: [0x62,0xf1,0xfd,0x28,0x7a,0xc0]
 ; X86-NEXT:    vpaddq %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xd4,0xc0]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_pd2qq_256:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vcvttpd2qq %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x29,0x7a,0xc8]
 ; X64-NEXT:    vcvttpd2qq %ymm0, %ymm0 # encoding: [0x62,0xf1,0xfd,0x28,0x7a,0xc0]
-; X64-NEXT:    vmovdqa64 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x29,0x6f,0xc8]
 ; X64-NEXT:    vpaddq %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xd4,0xc0]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <4 x i64> @llvm.x86.avx512.mask.cvttpd2qq.256(<4 x double> %x0, <4 x i64> %x1, i8 %x2)
   %res1 = call <4 x i64> @llvm.x86.avx512.mask.cvttpd2qq.256(<4 x double> %x0, <4 x i64> %x1, i8 -1)
   %res2 = add <4 x i64> %res, %res1
   ret <4 x i64> %res2
 }
 
 declare <2 x i64> @llvm.x86.avx512.mask.cvttpd2uqq.128(<2 x double>, <2 x i64>, i8)
 
 define <2 x i64>@test_int_x86_avx512_mask_cvtt_pd2uqq_128(<2 x double> %x0, <2 x i64> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_pd2uqq_128:
 ; X86:       # %bb.0:
-; X86-NEXT:    vcvttpd2uqq %xmm0, %xmm0 # encoding: [0x62,0xf1,0xfd,0x08,0x78,0xc0]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc8]
+; X86-NEXT:    vcvttpd2uqq %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x78,0xc8]
+; X86-NEXT:    vcvttpd2uqq %xmm0, %xmm0 # encoding: [0x62,0xf1,0xfd,0x08,0x78,0xc0]
 ; X86-NEXT:    vpaddq %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xd4,0xc0]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_pd2uqq_128:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vcvttpd2uqq %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x78,0xc8]
 ; X64-NEXT:    vcvttpd2uqq %xmm0, %xmm0 # encoding: [0x62,0xf1,0xfd,0x08,0x78,0xc0]
-; X64-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc8]
 ; X64-NEXT:    vpaddq %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xd4,0xc0]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttpd2uqq.128(<2 x double> %x0, <2 x i64> %x1, i8 %x2)
   %res1 = call <2 x i64> @llvm.x86.avx512.mask.cvttpd2uqq.128(<2 x double> %x0, <2 x i64> %x1, i8 -1)
   %res2 = add <2 x i64> %res, %res1
   ret <2 x i64> %res2
 }
 
 declare <4 x i64> @llvm.x86.avx512.mask.cvttpd2uqq.256(<4 x double>, <4 x i64>, i8)
 
 define <4 x i64>@test_int_x86_avx512_mask_cvtt_pd2uqq_256(<4 x double> %x0, <4 x i64> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_pd2uqq_256:
 ; X86:       # %bb.0:
-; X86-NEXT:    vcvttpd2uqq %ymm0, %ymm0 # encoding: [0x62,0xf1,0xfd,0x28,0x78,0xc0]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vmovdqa64 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x29,0x6f,0xc8]
+; X86-NEXT:    vcvttpd2uqq %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x29,0x78,0xc8]
+; X86-NEXT:    vcvttpd2uqq %ymm0, %ymm0 # encoding: [0x62,0xf1,0xfd,0x28,0x78,0xc0]
 ; X86-NEXT:    vpaddq %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xd4,0xc0]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_pd2uqq_256:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vcvttpd2uqq %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x29,0x78,0xc8]
 ; X64-NEXT:    vcvttpd2uqq %ymm0, %ymm0 # encoding: [0x62,0xf1,0xfd,0x28,0x78,0xc0]
-; X64-NEXT:    vmovdqa64 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x29,0x6f,0xc8]
 ; X64-NEXT:    vpaddq %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xd4,0xc0]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <4 x i64> @llvm.x86.avx512.mask.cvttpd2uqq.256(<4 x double> %x0, <4 x i64> %x1, i8 %x2)
   %res1 = call <4 x i64> @llvm.x86.avx512.mask.cvttpd2uqq.256(<4 x double> %x0, <4 x i64> %x1, i8 -1)
   %res2 = add <4 x i64> %res, %res1
   ret <4 x i64> %res2
 }
 
 declare <2 x i64> @llvm.x86.avx512.mask.cvttps2qq.128(<4 x float>, <2 x i64>, i8)
 
 define <2 x i64>@test_int_x86_avx512_mask_cvtt_ps2qq_128(<4 x float> %x0, <2 x i64> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2qq_128:
 ; X86:       # %bb.0:
-; X86-NEXT:    vcvttps2qq %xmm0, %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x7a,0xc0]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc8]
+; X86-NEXT:    vcvttps2qq %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x7a,0xc8]
+; X86-NEXT:    vcvttps2qq %xmm0, %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x7a,0xc0]
 ; X86-NEXT:    vpaddq %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xd4,0xc0]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2qq_128:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vcvttps2qq %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x7a,0xc8]
 ; X64-NEXT:    vcvttps2qq %xmm0, %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x7a,0xc0]
-; X64-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc8]
 ; X64-NEXT:    vpaddq %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xd4,0xc0]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2qq.128(<4 x float> %x0, <2 x i64> %x1, i8 %x2)
   %res1 = call <2 x i64> @llvm.x86.avx512.mask.cvttps2qq.128(<4 x float> %x0, <2 x i64> %x1, i8 -1)
   %res2 = add <2 x i64> %res, %res1
   ret <2 x i64> %res2
 }
 
; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2qq.128(<4 x float> %x0b, <2 x i64> undef, i8 -1)
   ret <2 x i64> %res
 }
 
 define <2 x i64> @test_int_x86_avx512_mask_cvtt_ps2qq_128_load(<2 x float>* %p, <2 x i64> %passthru, i8 %mask) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2qq_128_load:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    vcvttps2qq (%eax), %xmm1 # encoding: [0x62,0xf1,0x7d,0x08,0x7a,0x08]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc1]
+; X86-NEXT:    vcvttps2qq (%eax), %xmm0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x7a,0x00]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2qq_128_load:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
-; X64-NEXT:    vcvttps2qq (%rdi), %xmm1 # encoding: [0x62,0xf1,0x7d,0x08,0x7a,0x0f]
-; X64-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc1]
+; X64-NEXT:    vcvttps2qq (%rdi), %xmm0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x7a,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %x0 = load <2 x float>, <2 x float>* %p
   %x0b = shufflevector <2 x float> %x0, <2 x float> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2qq.128(<4 x float> %x0b, <2 x i64> %passthru, i8 %mask)
   ret <2 x i64> %res
 }
 
 define <2 x i64> @test_int_x86_avx512_maskz_cvtt_ps2qq_128_load(<2 x float>* %p, i8 %mask) {
 ; X86-LABEL: test_int_x86_avx512_maskz_cvtt_ps2qq_128_load:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    vcvttps2qq (%eax), %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x7a,0x00]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xfd,0x89,0x6f,0xc0]
+; X86-NEXT:    vcvttps2qq (%eax), %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0x7a,0x00]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_maskz_cvtt_ps2qq_128_load:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
-; X64-NEXT:    vcvttps2qq (%rdi), %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x7a,0x07]
-; X64-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xfd,0x89,0x6f,0xc0]
+; X64-NEXT:    vcvttps2qq (%rdi), %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0x7a,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %x0 = load <2 x float>, <2 x float>* %p
   %x0b = shufflevector <2 x float> %x0, <2 x float> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2qq.128(<4 x float> %x0b, <2 x i64> zeroinitializer, i8 %mask)
   ret <2 x i64> %res
 }
 
 
; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2qq.128(<4 x float> %x0b, <2 x i64> undef, i8 -1)
   ret <2 x i64> %res
 }
 
 define <2 x i64> @test_int_x86_avx512_mask_cvtt_ps2qq_128_load_2(<2 x float>* %p, <2 x i64> %passthru, i8 %mask) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2qq_128_load_2:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    vcvttps2qq (%eax), %xmm1 # encoding: [0x62,0xf1,0x7d,0x08,0x7a,0x08]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc1]
+; X86-NEXT:    vcvttps2qq (%eax), %xmm0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x7a,0x00]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2qq_128_load_2:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
-; X64-NEXT:    vcvttps2qq (%rdi), %xmm1 # encoding: [0x62,0xf1,0x7d,0x08,0x7a,0x0f]
-; X64-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc1]
+; X64-NEXT:    vcvttps2qq (%rdi), %xmm0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x7a,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %x0 = load <2 x float>, <2 x float>* %p
   %x0b = shufflevector <2 x float> %x0, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2qq.128(<4 x float> %x0b, <2 x i64> %passthru, i8 %mask)
   ret <2 x i64> %res
 }
 
 define <2 x i64> @test_int_x86_avx512_maskz_cvtt_ps2qq_128_load_2(<2 x float>* %p, i8 %mask) {
 ; X86-LABEL: test_int_x86_avx512_maskz_cvtt_ps2qq_128_load_2:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    vcvttps2qq (%eax), %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x7a,0x00]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xfd,0x89,0x6f,0xc0]
+; X86-NEXT:    vcvttps2qq (%eax), %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0x7a,0x00]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_maskz_cvtt_ps2qq_128_load_2:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
-; X64-NEXT:    vcvttps2qq (%rdi), %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x7a,0x07]
-; X64-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xfd,0x89,0x6f,0xc0]
+; X64-NEXT:    vcvttps2qq (%rdi), %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0x7a,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %x0 = load <2 x float>, <2 x float>* %p
   %x0b = shufflevector <2 x float> %x0, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2qq.128(<4 x float> %x0b, <2 x i64> zeroinitializer, i8 %mask)
   ret <2 x i64> %res
 }
 
 define <2 x i64> @test_int_x86_avx512_cvtt_ps2qq_128_load_3(<4 x float>* %p) {
; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2qq.128(<4 x float> %x0, <2 x i64> undef, i8 -1)
   ret <2 x i64> %res
 }
 
 define <2 x i64> @test_int_x86_avx512_mask_cvtt_ps2qq_128_load_3(<4 x float>* %p, <2 x i64> %passthru, i8 %mask) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2qq_128_load_3:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    vcvttps2qq (%eax), %xmm1 # encoding: [0x62,0xf1,0x7d,0x08,0x7a,0x08]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc1]
+; X86-NEXT:    vcvttps2qq (%eax), %xmm0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x7a,0x00]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2qq_128_load_3:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
-; X64-NEXT:    vcvttps2qq (%rdi), %xmm1 # encoding: [0x62,0xf1,0x7d,0x08,0x7a,0x0f]
-; X64-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc1]
+; X64-NEXT:    vcvttps2qq (%rdi), %xmm0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x7a,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %x0 = load <4 x float>, <4 x float>* %p
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2qq.128(<4 x float> %x0, <2 x i64> %passthru, i8 %mask)
   ret <2 x i64> %res
 }
 
 define <2 x i64> @test_int_x86_avx512_maskz_cvtt_ps2qq_128_load_3(<4 x float>* %p, i8 %mask) {
 ; X86-LABEL: test_int_x86_avx512_maskz_cvtt_ps2qq_128_load_3:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    vcvttps2qq (%eax), %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x7a,0x00]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xfd,0x89,0x6f,0xc0]
+; X86-NEXT:    vcvttps2qq (%eax), %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0x7a,0x00]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_maskz_cvtt_ps2qq_128_load_3:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
-; X64-NEXT:    vcvttps2qq (%rdi), %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x7a,0x07]
-; X64-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xfd,0x89,0x6f,0xc0]
+; X64-NEXT:    vcvttps2qq (%rdi), %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0x7a,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %x0 = load <4 x float>, <4 x float>* %p
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2qq.128(<4 x float> %x0, <2 x i64> zeroinitializer, i8 %mask)
   ret <2 x i64> %res
 }
 
 declare <4 x i64> @llvm.x86.avx512.mask.cvttps2qq.256(<4 x float>, <4 x i64>, i8)
 
 define <4 x i64>@test_int_x86_avx512_mask_cvtt_ps2qq_256(<4 x float> %x0, <4 x i64> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2qq_256:
 ; X86:       # %bb.0:
-; X86-NEXT:    vcvttps2qq %xmm0, %ymm0 # encoding: [0x62,0xf1,0x7d,0x28,0x7a,0xc0]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vmovdqa64 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x29,0x6f,0xc8]
+; X86-NEXT:    vcvttps2qq %xmm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0x7a,0xc8]
+; X86-NEXT:    vcvttps2qq %xmm0, %ymm0 # encoding: [0x62,0xf1,0x7d,0x28,0x7a,0xc0]
 ; X86-NEXT:    vpaddq %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xd4,0xc0]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2qq_256:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vcvttps2qq %xmm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0x7a,0xc8]
 ; X64-NEXT:    vcvttps2qq %xmm0, %ymm0 # encoding: [0x62,0xf1,0x7d,0x28,0x7a,0xc0]
-; X64-NEXT:    vmovdqa64 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x29,0x6f,0xc8]
 ; X64-NEXT:    vpaddq %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xd4,0xc0]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <4 x i64> @llvm.x86.avx512.mask.cvttps2qq.256(<4 x float> %x0, <4 x i64> %x1, i8 %x2)
   %res1 = call <4 x i64> @llvm.x86.avx512.mask.cvttps2qq.256(<4 x float> %x0, <4 x i64> %x1, i8 -1)
   %res2 = add <4 x i64> %res, %res1
   ret <4 x i64> %res2
 }
 
; X64-NEXT:    retq # encoding: [0xc3]
   ret <4 x float> %res2
 }
 
 declare <2 x i64> @llvm.x86.avx512.mask.cvttps2uqq.128(<4 x float>, <2 x i64>, i8)
 
 define <2 x i64>@test_int_x86_avx512_mask_cvtt_ps2uqq_128(<4 x float> %x0, <2 x i64> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2uqq_128:
 ; X86:       # %bb.0:
-; X86-NEXT:    vcvttps2uqq %xmm0, %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x78,0xc0]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc8]
+; X86-NEXT:    vcvttps2uqq %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x78,0xc8]
+; X86-NEXT:    vcvttps2uqq %xmm0, %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x78,0xc0]
 ; X86-NEXT:    vpaddq %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xd4,0xc0]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2uqq_128:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vcvttps2uqq %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x78,0xc8]
 ; X64-NEXT:    vcvttps2uqq %xmm0, %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x78,0xc0]
-; X64-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc8]
 ; X64-NEXT:    vpaddq %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xd4,0xc0]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2uqq.128(<4 x float> %x0, <2 x i64> %x1, i8 %x2)
   %res1 = call <2 x i64> @llvm.x86.avx512.mask.cvttps2uqq.128(<4 x float> %x0, <2 x i64> %x1, i8 -1)
   %res2 = add <2 x i64> %res, %res1
   ret <2 x i64> %res2
 }
 
; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2uqq.128(<4 x float> %x0b, <2 x i64> undef, i8 -1)
   ret <2 x i64> %res
 }
 
 define <2 x i64> @test_int_x86_avx512_mask_cvtt_ps2uqq_128_load(<2 x float>* %p, <2 x i64> %passthru, i8 %mask) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2uqq_128_load:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    vcvttps2uqq (%eax), %xmm1 # encoding: [0x62,0xf1,0x7d,0x08,0x78,0x08]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc1]
+; X86-NEXT:    vcvttps2uqq (%eax), %xmm0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x78,0x00]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2uqq_128_load:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
-; X64-NEXT:    vcvttps2uqq (%rdi), %xmm1 # encoding: [0x62,0xf1,0x7d,0x08,0x78,0x0f]
-; X64-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc1]
+; X64-NEXT:    vcvttps2uqq (%rdi), %xmm0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x78,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %x0 = load <2 x float>, <2 x float>* %p
   %x0b = shufflevector <2 x float> %x0, <2 x float> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2uqq.128(<4 x float> %x0b, <2 x i64> %passthru, i8 %mask)
   ret <2 x i64> %res
 }
 
 define <2 x i64> @test_int_x86_avx512_maskz_cvtt_ps2uqq_128_load(<2 x float>* %p, i8 %mask) {
 ; X86-LABEL: test_int_x86_avx512_maskz_cvtt_ps2uqq_128_load:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    vcvttps2uqq (%eax), %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x78,0x00]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xfd,0x89,0x6f,0xc0]
+; X86-NEXT:    vcvttps2uqq (%eax), %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0x78,0x00]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_maskz_cvtt_ps2uqq_128_load:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
-; X64-NEXT:    vcvttps2uqq (%rdi), %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x78,0x07]
-; X64-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xfd,0x89,0x6f,0xc0]
+; X64-NEXT:    vcvttps2uqq (%rdi), %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0x78,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %x0 = load <2 x float>, <2 x float>* %p
   %x0b = shufflevector <2 x float> %x0, <2 x float> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2uqq.128(<4 x float> %x0b, <2 x i64> zeroinitializer, i8 %mask)
   ret <2 x i64> %res
 }
 
 
; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2uqq.128(<4 x float> %x0b, <2 x i64> undef, i8 -1)
   ret <2 x i64> %res
 }
 
 define <2 x i64> @test_int_x86_avx512_mask_cvtt_ps2uqq_128_load_2(<2 x float>* %p, <2 x i64> %passthru, i8 %mask) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2uqq_128_load_2:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    vcvttps2uqq (%eax), %xmm1 # encoding: [0x62,0xf1,0x7d,0x08,0x78,0x08]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc1]
+; X86-NEXT:    vcvttps2uqq (%eax), %xmm0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x78,0x00]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2uqq_128_load_2:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
-; X64-NEXT:    vcvttps2uqq (%rdi), %xmm1 # encoding: [0x62,0xf1,0x7d,0x08,0x78,0x0f]
-; X64-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc1]
+; X64-NEXT:    vcvttps2uqq (%rdi), %xmm0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x78,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %x0 = load <2 x float>, <2 x float>* %p
   %x0b = shufflevector <2 x float> %x0, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2uqq.128(<4 x float> %x0b, <2 x i64> %passthru, i8 %mask)
   ret <2 x i64> %res
 }
 
 define <2 x i64> @test_int_x86_avx512_maskz_cvtt_ps2uqq_128_load_2(<2 x float>* %p, i8 %mask) {
 ; X86-LABEL: test_int_x86_avx512_maskz_cvtt_ps2uqq_128_load_2:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    vcvttps2uqq (%eax), %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x78,0x00]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xfd,0x89,0x6f,0xc0]
+; X86-NEXT:    vcvttps2uqq (%eax), %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0x78,0x00]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_maskz_cvtt_ps2uqq_128_load_2:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
-; X64-NEXT:    vcvttps2uqq (%rdi), %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x78,0x07]
-; X64-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xfd,0x89,0x6f,0xc0]
+; X64-NEXT:    vcvttps2uqq (%rdi), %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0x78,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %x0 = load <2 x float>, <2 x float>* %p
   %x0b = shufflevector <2 x float> %x0, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2uqq.128(<4 x float> %x0b, <2 x i64> zeroinitializer, i8 %mask)
   ret <2 x i64> %res
 }
 
 define <2 x i64> @test_int_x86_avx512_cvtt_ps2uqq_128_load_3(<4 x float>* %p) {
; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2uqq.128(<4 x float> %x0, <2 x i64> undef, i8 -1)
   ret <2 x i64> %res
 }
 
 define <2 x i64> @test_int_x86_avx512_mask_cvtt_ps2uqq_128_load_3(<4 x float>* %p, <2 x i64> %passthru, i8 %mask) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2uqq_128_load_3:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    vcvttps2uqq (%eax), %xmm1 # encoding: [0x62,0xf1,0x7d,0x08,0x78,0x08]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc1]
+; X86-NEXT:    vcvttps2uqq (%eax), %xmm0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x78,0x00]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2uqq_128_load_3:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
-; X64-NEXT:    vcvttps2uqq (%rdi), %xmm1 # encoding: [0x62,0xf1,0x7d,0x08,0x78,0x0f]
-; X64-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xfd,0x09,0x6f,0xc1]
+; X64-NEXT:    vcvttps2uqq (%rdi), %xmm0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x78,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %x0 = load <4 x float>, <4 x float>* %p
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2uqq.128(<4 x float> %x0, <2 x i64> %passthru, i8 %mask)
   ret <2 x i64> %res
 }
 
 define <2 x i64> @test_int_x86_avx512_maskz_cvtt_ps2uqq_128_load_3(<4 x float>* %p, i8 %mask) {
 ; X86-LABEL: test_int_x86_avx512_maskz_cvtt_ps2uqq_128_load_3:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    vcvttps2uqq (%eax), %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x78,0x00]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xfd,0x89,0x6f,0xc0]
+; X86-NEXT:    vcvttps2uqq (%eax), %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0x78,0x00]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_maskz_cvtt_ps2uqq_128_load_3:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
-; X64-NEXT:    vcvttps2uqq (%rdi), %xmm0 # encoding: [0x62,0xf1,0x7d,0x08,0x78,0x07]
-; X64-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xfd,0x89,0x6f,0xc0]
+; X64-NEXT:    vcvttps2uqq (%rdi), %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0x78,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %x0 = load <4 x float>, <4 x float>* %p
   %res = call <2 x i64> @llvm.x86.avx512.mask.cvttps2uqq.128(<4 x float> %x0, <2 x i64> zeroinitializer, i8 %mask)
   ret <2 x i64> %res
 }
 
 declare <4 x i64> @llvm.x86.avx512.mask.cvttps2uqq.256(<4 x float>, <4 x i64>, i8)
 
 define <4 x i64>@test_int_x86_avx512_mask_cvtt_ps2uqq_256(<4 x float> %x0, <4 x i64> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2uqq_256:
 ; X86:       # %bb.0:
-; X86-NEXT:    vcvttps2uqq %xmm0, %ymm0 # encoding: [0x62,0xf1,0x7d,0x28,0x78,0xc0]
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vmovdqa64 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x29,0x6f,0xc8]
+; X86-NEXT:    vcvttps2uqq %xmm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0x78,0xc8]
+; X86-NEXT:    vcvttps2uqq %xmm0, %ymm0 # encoding: [0x62,0xf1,0x7d,0x28,0x78,0xc0]
 ; X86-NEXT:    vpaddq %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xd4,0xc0]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2uqq_256:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vcvttps2uqq %xmm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0x78,0xc8]
 ; X64-NEXT:    vcvttps2uqq %xmm0, %ymm0 # encoding: [0x62,0xf1,0x7d,0x28,0x78,0xc0]
-; X64-NEXT:    vmovdqa64 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x29,0x6f,0xc8]
 ; X64-NEXT:    vpaddq %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xd4,0xc0]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <4 x i64> @llvm.x86.avx512.mask.cvttps2uqq.256(<4 x float> %x0, <4 x i64> %x1, i8 %x2)
   %res1 = call <4 x i64> @llvm.x86.avx512.mask.cvttps2uqq.256(<4 x float> %x0, <4 x i64> %x1, i8 -1)
   %res2 = add <4 x i64> %res, %res1
   ret <4 x i64> %res2
 }
 

llvm/test/CodeGen/X86/avx512vl-intrinsics-fast-isel.ll

entry:
   %1 = bitcast <4 x i32> %0 to <2 x i64>
   ret <2 x i64> %1
 }
 
 define <2 x i64> @test_mm256_mask_cvttpd_epi32(<2 x i64> %__W, i8 zeroext %__U, <4 x double> %__A) {
 ; X86-LABEL: test_mm256_mask_cvttpd_epi32:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vcvttpd2dq %ymm1, %xmm1
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vmovdqa32 %xmm1, %xmm0 {%k1}
+; X86-NEXT:    vcvttpd2dq %ymm1, %xmm0 {%k1}
 ; X86-NEXT:    vzeroupper
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm256_mask_cvttpd_epi32:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vcvttpd2dq %ymm1, %xmm1
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vmovdqa32 %xmm1, %xmm0 {%k1}
+; X64-NEXT:    vcvttpd2dq %ymm1, %xmm0 {%k1}
 ; X64-NEXT:    vzeroupper
 ; X64-NEXT:    retq
 entry:
   %0 = tail call <4 x i32> @llvm.x86.avx.cvtt.pd2dq.256(<4 x double> %__A) #8
   %1 = bitcast <2 x i64> %__W to <4 x i32>
   %2 = bitcast i8 %__U to <8 x i1>
   %extract.i = shufflevector <8 x i1> %2, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %3 = select <4 x i1> %extract.i, <4 x i32> %0, <4 x i32> %1
   %4 = bitcast <4 x i32> %3 to <2 x i64>
   ret <2 x i64> %4
 }
 
 define <2 x i64> @test_mm256_maskz_cvttpd_epi32(i8 zeroext %__U, <4 x double> %__A) {
 ; X86-LABEL: test_mm256_maskz_cvttpd_epi32:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vcvttpd2dq %ymm0, %xmm0
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
+; X86-NEXT:    vcvttpd2dq %ymm0, %xmm0 {%k1} {z}
 ; X86-NEXT:    vzeroupper
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm256_maskz_cvttpd_epi32:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vcvttpd2dq %ymm0, %xmm0
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
+; X64-NEXT:    vcvttpd2dq %ymm0, %xmm0 {%k1} {z}
 ; X64-NEXT:    vzeroupper
 ; X64-NEXT:    retq
 entry:
   %0 = tail call <4 x i32> @llvm.x86.avx.cvtt.pd2dq.256(<4 x double> %__A) #8
   %1 = bitcast i8 %__U to <8 x i1>
   %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %2 = select <4 x i1> %extract.i, <4 x i32> %0, <4 x i32> zeroinitializer
   %3 = bitcast <4 x i32> %2 to <2 x i64>
entry:
   ret <2 x i64> %1
 }
 
 define <2 x i64> @test_mm256_mask_cvttpd_epu32(<2 x i64> %__W, i8 zeroext %__U, <4 x double> %__A) {
 ; X86-LABEL: test_mm256_mask_cvttpd_epu32:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vcvttpd2udq %ymm1, %xmm1
-; X86-NEXT:    vmovdqa32 %xmm1, %xmm0 {%k1}
+; X86-NEXT:    vcvttpd2udq %ymm1, %xmm0 {%k1}
 ; X86-NEXT:    vzeroupper
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm256_mask_cvttpd_epu32:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vcvttpd2udq %ymm1, %xmm1
-; X64-NEXT:    vmovdqa32 %xmm1, %xmm0 {%k1}
+; X64-NEXT:    vcvttpd2udq %ymm1, %xmm0 {%k1}
 ; X64-NEXT:    vzeroupper
 ; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__W to <4 x i32>
   %1 = tail call <4 x i32> @llvm.x86.avx512.mask.cvttpd2udq.256(<4 x double> %__A, <4 x i32> %0, i8 %__U) #8
   %2 = bitcast <4 x i32> %1 to <2 x i64>
   ret <2 x i64> %2
 }
 
 define <2 x i64> @test_mm256_maskz_cvttpd_epu32(i8 zeroext %__U, <4 x double> %__A) {
 ; X86-LABEL: test_mm256_maskz_cvttpd_epu32:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vcvttpd2udq %ymm0, %xmm0
-; X86-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
+; X86-NEXT:    vcvttpd2udq %ymm0, %xmm0 {%k1} {z}
 ; X86-NEXT:    vzeroupper
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm256_maskz_cvttpd_epu32:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vcvttpd2udq %ymm0, %xmm0
-; X64-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
+; X64-NEXT:    vcvttpd2udq %ymm0, %xmm0 {%k1} {z}
 ; X64-NEXT:    vzeroupper
 ; X64-NEXT:    retq
 entry:
   %0 = tail call <4 x i32> @llvm.x86.avx512.mask.cvttpd2udq.256(<4 x double> %__A, <4 x i32> zeroinitializer, i8 %__U) #8
   %1 = bitcast <4 x i32> %0 to <2 x i64>
   ret <2 x i64> %1
 }
 
 define <2 x i64> @test_mm_mask_cvttps_epi32(<2 x i64> %__W, i8 zeroext %__U, <4 x float> %__A) {
 ; X86-LABEL: test_mm_mask_cvttps_epi32:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vcvttps2dq %xmm1, %xmm1
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vmovdqa32 %xmm1, %xmm0 {%k1}
+; X86-NEXT:    vcvttps2dq %xmm1, %xmm0 {%k1}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_mask_cvttps_epi32:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vcvttps2dq %xmm1, %xmm1
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vmovdqa32 %xmm1, %xmm0 {%k1}
+; X64-NEXT:    vcvttps2dq %xmm1, %xmm0 {%k1}
 ; X64-NEXT:    retq
 entry:
   %0 = tail call <4 x i32> @llvm.x86.sse2.cvttps2dq(<4 x float> %__A) #8
   %1 = bitcast <2 x i64> %__W to <4 x i32>
   %2 = bitcast i8 %__U to <8 x i1>
   %extract.i = shufflevector <8 x i1> %2, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %3 = select <4 x i1> %extract.i, <4 x i32> %0, <4 x i32> %1
   %4 = bitcast <4 x i32> %3 to <2 x i64>
   ret <2 x i64> %4
 }
 
 define <2 x i64> @test_mm_maskz_cvttps_epi32(i8 zeroext %__U, <4 x float> %__A) {
 ; X86-LABEL: test_mm_maskz_cvttps_epi32:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vcvttps2dq %xmm0, %xmm0
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
+; X86-NEXT:    vcvttps2dq %xmm0, %xmm0 {%k1} {z}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_maskz_cvttps_epi32:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vcvttps2dq %xmm0, %xmm0
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
+; X64-NEXT:    vcvttps2dq %xmm0, %xmm0 {%k1} {z}
 ; X64-NEXT:    retq
 entry:
   %0 = tail call <4 x i32> @llvm.x86.sse2.cvttps2dq(<4 x float> %__A) #8
   %1 = bitcast i8 %__U to <8 x i1>
   %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %2 = select <4 x i1> %extract.i, <4 x i32> %0, <4 x i32> zeroinitializer
   %3 = bitcast <4 x i32> %2 to <2 x i64>
   ret <2 x i64> %3
 }
 
 define <4 x i64> @test_mm256_mask_cvttps_epi32(<4 x i64> %__W, i8 zeroext %__U, <8 x float> %__A) {
 ; X86-LABEL: test_mm256_mask_cvttps_epi32:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vcvttps2dq %ymm1, %ymm1
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vmovdqa32 %ymm1, %ymm0 {%k1}
+; X86-NEXT:    vcvttps2dq %ymm1, %ymm0 {%k1}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm256_mask_cvttps_epi32:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vcvttps2dq %ymm1, %ymm1
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vmovdqa32 %ymm1, %ymm0 {%k1}
+; X64-NEXT:    vcvttps2dq %ymm1, %ymm0 {%k1}
 ; X64-NEXT:    retq
 entry:
   %0 = tail call <8 x i32> @llvm.x86.avx.cvtt.ps2dq.256(<8 x float> %__A) #8
   %1 = bitcast <4 x i64> %__W to <8 x i32>
   %2 = bitcast i8 %__U to <8 x i1>
   %3 = select <8 x i1> %2, <8 x i32> %0, <8 x i32> %1
   %4 = bitcast <8 x i32> %3 to <4 x i64>
   ret <4 x i64> %4
 }
 
 define <4 x i64> @test_mm256_maskz_cvttps_epi32(i8 zeroext %__U, <8 x float> %__A) {
 ; X86-LABEL: test_mm256_maskz_cvttps_epi32:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
+; X86-NEXT:    vcvttps2dq %ymm0, %ymm0 {%k1} {z}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm256_maskz_cvttps_epi32:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
+; X64-NEXT:    vcvttps2dq %ymm0, %ymm0 {%k1} {z}
 ; X64-NEXT:    retq
 entry:
   %0 = tail call <8 x i32> @llvm.x86.avx.cvtt.ps2dq.256(<8 x float> %__A) #8
   %1 = bitcast i8 %__U to <8 x i1>
   %2 = select <8 x i1> %1, <8 x i32> %0, <8 x i32> zeroinitializer
   %3 = bitcast <8 x i32> %2 to <4 x i64>
   ret <4 x i64> %3
 }
entry:
   ret <2 x i64> %1
 }
 
 define <2 x i64> @test_mm_mask_cvttps_epu32(<2 x i64> %__W, i8 zeroext %__U, <4 x float> %__A) {
 ; X86-LABEL: test_mm_mask_cvttps_epu32:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vcvttps2udq %xmm1, %xmm1
-; X86-NEXT:    vmovdqa32 %xmm1, %xmm0 {%k1}
+; X86-NEXT:    vcvttps2udq %xmm1, %xmm0 {%k1}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_mask_cvttps_epu32:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vcvttps2udq %xmm1, %xmm1
-; X64-NEXT:    vmovdqa32 %xmm1, %xmm0 {%k1}
+; X64-NEXT:    vcvttps2udq %xmm1, %xmm0 {%k1}
 ; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__W to <4 x i32>
   %1 = tail call <4 x i32> @llvm.x86.avx512.mask.cvttps2udq.128(<4 x float> %__A, <4 x i32> %0, i8 %__U) #8
   %2 = bitcast <4 x i32> %1 to <2 x i64>
   ret <2 x i64> %2
 }
 
 define <2 x i64> @test_mm_maskz_cvttps_epu32(i8 zeroext %__U, <4 x float> %__A) {
 ; X86-LABEL: test_mm_maskz_cvttps_epu32:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vcvttps2udq %xmm0, %xmm0
-; X86-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
+; X86-NEXT:    vcvttps2udq %xmm0, %xmm0 {%k1} {z}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_maskz_cvttps_epu32:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vcvttps2udq %xmm0, %xmm0
-; X64-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
+; X64-NEXT:    vcvttps2udq %xmm0, %xmm0 {%k1} {z}
 ; X64-NEXT:    retq
 entry:
   %0 = tail call <4 x i32> @llvm.x86.avx512.mask.cvttps2udq.128(<4 x float> %__A, <4 x i32> zeroinitializer, i8 %__U) #8
   %1 = bitcast <4 x i32> %0 to <2 x i64>
   ret <2 x i64> %1
 }
 
 define <4 x i64> @test_mm256_cvttps_epu32(<8 x float> %__A) {
 ; CHECK-LABEL: test_mm256_cvttps_epu32:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vcvttps2udq %ymm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %0 = tail call <8 x i32> @llvm.x86.avx512.mask.cvttps2udq.256(<8 x float> %__A, <8 x i32> zeroinitializer, i8 -1) #8
   %1 = bitcast <8 x i32> %0 to <4 x i64>
   ret <4 x i64> %1
 }
 
 define <4 x i64> @test_mm256_mask_cvttps_epu32(<4 x i64> %__W, i8 zeroext %__U, <8 x float> %__A) {
 ; X86-LABEL: test_mm256_mask_cvttps_epu32:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vcvttps2udq %ymm1, %ymm1
-; X86-NEXT:    vmovdqa32 %ymm1, %ymm0 {%k1}
+; X86-NEXT:    vcvttps2udq %ymm1, %ymm0 {%k1}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm256_mask_cvttps_epu32:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vcvttps2udq %ymm1, %ymm1
-; X64-NEXT:    vmovdqa32 %ymm1, %ymm0 {%k1}
+; X64-NEXT:    vcvttps2udq %ymm1, %ymm0 {%k1}
 ; X64-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__W to <8 x i32>
   %1 = tail call <8 x i32> @llvm.x86.avx512.mask.cvttps2udq.256(<8 x float> %__A, <8 x i32> %0, i8 %__U) #8
   %2 = bitcast <8 x i32> %1 to <4 x i64>
   ret <4 x i64> %2
 }
 
 define <4 x i64> @test_mm256_maskz_cvttps_epu32(i8 zeroext %__U, <8 x float> %__A) {
 ; X86-LABEL: test_mm256_maskz_cvttps_epu32:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vcvttps2udq %ymm0, %ymm0
-; X86-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
+; X86-NEXT:    vcvttps2udq %ymm0, %ymm0 {%k1} {z}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm256_maskz_cvttps_epu32:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vcvttps2udq %ymm0, %ymm0
-; X64-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
+; X64-NEXT:    vcvttps2udq %ymm0, %ymm0 {%k1} {z}
 ; X64-NEXT:    retq
 entry:
   %0 = tail call <8 x i32> @llvm.x86.avx512.mask.cvttps2udq.256(<8 x float> %__A, <8 x i32> zeroinitializer, i8 %__U) #8
   %1 = bitcast <8 x i32> %0 to <4 x i64>
   ret <4 x i64> %1
 }
 
 define <2 x double> @test_mm_cvtepu32_pd(<2 x i64> %__A) local_unnamed_addr #0 {

llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll

; X64-NEXT:    retq # encoding: [0xc3]
   ret <2 x double> %res2
 }
 
 declare <4 x i32> @llvm.x86.avx512.mask.cvttpd2dq.256(<4 x double>, <4 x i32>, i8)
 
 define <4 x i32>@test_int_x86_avx512_mask_cvtt_pd2dq_256(<4 x double> %x0, <4 x i32> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_pd2dq_256:
 ; X86:       # %bb.0:
-; X86-NEXT:    vcvttpd2dq %ymm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0xe6,0xc0]
+; X86-NEXT:    vcvttpd2dq %ymm0, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0xe6,0xd0]
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
 ; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
-; X86-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x6f,0xc8]
-; X86-NEXT:    vpaddd %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xfe,0xc0]
+; X86-NEXT:    vcvttpd2dq %ymm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x29,0xe6,0xc8]
+; X86-NEXT:    vpaddd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xfe,0xc2]
 ; X86-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_pd2dq_256:
 ; X64:       # %bb.0:
-; X64-NEXT:    vcvttpd2dq %ymm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0xe6,0xc0]
+; X64-NEXT:    vcvttpd2dq %ymm0, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0xe6,0xd0]
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
-; X64-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x6f,0xc8]
-; X64-NEXT:    vpaddd %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xfe,0xc0]
+; X64-NEXT:    vcvttpd2dq %ymm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfd,0x29,0xe6,0xc8]
+; X64-NEXT:    vpaddd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xfe,0xc2]
 ; X64-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <4 x i32> @llvm.x86.avx512.mask.cvttpd2dq.256(<4 x double> %x0, <4 x i32> %x1, i8 %x2)
   %res1 = call <4 x i32> @llvm.x86.avx512.mask.cvttpd2dq.256(<4 x double> %x0, <4 x i32> %x1, i8 -1)
   %res2 = add <4 x i32> %res, %res1
   ret <4 x i32> %res2
 }
 
 declare <4 x i32> @llvm.x86.avx512.mask.cvttps2dq.128(<4 x float>, <4 x i32>, i8)
 
 define <4 x i32>@test_int_x86_avx512_mask_cvtt_ps2dq_128(<4 x float> %x0, <4 x i32> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2dq_128:
 ; X86:       # %bb.0:
-; X86-NEXT:    vcvttps2dq %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x5b,0xc0]
+; X86-NEXT:    vcvttps2dq %xmm0, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x5b,0xd0]
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
 ; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
-; X86-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x6f,0xc8]
-; X86-NEXT:    vpaddd %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xfe,0xc0]
+; X86-NEXT:    vcvttps2dq %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7e,0x09,0x5b,0xc8]
+; X86-NEXT:    vpaddd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xfe,0xc2]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2dq_128:
 ; X64:       # %bb.0:
-; X64-NEXT:    vcvttps2dq %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x5b,0xc0]
+; X64-NEXT:    vcvttps2dq %xmm0, %xmm2 # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x5b,0xd0]
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
-; X64-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x6f,0xc8]
-; X64-NEXT:    vpaddd %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xfe,0xc0]
+; X64-NEXT:    vcvttps2dq %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7e,0x09,0x5b,0xc8]
+; X64-NEXT:    vpaddd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xfe,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <4 x i32> @llvm.x86.avx512.mask.cvttps2dq.128(<4 x float> %x0, <4 x i32> %x1, i8 %x2)
   %res1 = call <4 x i32> @llvm.x86.avx512.mask.cvttps2dq.128(<4 x float> %x0, <4 x i32> %x1, i8 -1)
   %res2 = add <4 x i32> %res, %res1
   ret <4 x i32> %res2
 }
 
 declare <8 x i32> @llvm.x86.avx512.mask.cvttps2dq.256(<8 x float>, <8 x i32>, i8)
 
 define <8 x i32>@test_int_x86_avx512_mask_cvtt_ps2dq_256(<8 x float> %x0, <8 x i32> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2dq_256:
 ; X86:       # %bb.0:
-; X86-NEXT:    vcvttps2dq %ymm0, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfe,0x5b,0xc0]
+; X86-NEXT:    vcvttps2dq %ymm0, %ymm2 # EVEX TO VEX Compression encoding: [0xc5,0xfe,0x5b,0xd0]
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
 ; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
-; X86-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0x6f,0xc8]
-; X86-NEXT:    vpaddd %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xfe,0xc0]
+; X86-NEXT:    vcvttps2dq %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7e,0x29,0x5b,0xc8]
+; X86-NEXT:    vpaddd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xfe,0xc2]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2dq_256:
 ; X64:       # %bb.0:
-; X64-NEXT:    vcvttps2dq %ymm0, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfe,0x5b,0xc0]
+; X64-NEXT:    vcvttps2dq %ymm0, %ymm2 # EVEX TO VEX Compression encoding: [0xc5,0xfe,0x5b,0xd0]
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
-; X64-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0x6f,0xc8]
-; X64-NEXT:    vpaddd %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xfe,0xc0]
+; X64-NEXT:    vcvttps2dq %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7e,0x29,0x5b,0xc8]
+; X64-NEXT:    vpaddd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xfe,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <8 x i32> @llvm.x86.avx512.mask.cvttps2dq.256(<8 x float> %x0, <8 x i32> %x1, i8 %x2)
   %res1 = call <8 x i32> @llvm.x86.avx512.mask.cvttps2dq.256(<8 x float> %x0, <8 x i32> %x1, i8 -1)
   %res2 = add <8 x i32> %res, %res1
   ret <8 x i32> %res2
 }
 
 declare <8 x float> @llvm.x86.avx512.mask.permvar.sf.256(<8 x float>, <8 x i32>, <8 x float>, i8)

llvm/test/CodeGen/X86/avx512vl-intrinsics.ll

 
 declare <4 x i32> @llvm.x86.avx512.mask.cvttpd2udq.256(<4 x double>, <4 x i32>, i8)
 
 define <4 x i32>@test_int_x86_avx512_mask_cvtt_pd2udq_256(<4 x double> %x0, <4 x i32> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_pd2udq_256:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
 ; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vcvttpd2udq %ymm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfc,0x29,0x78,0xc8]
 ; X86-NEXT:    vcvttpd2udq %ymm0, %xmm0 # encoding: [0x62,0xf1,0xfc,0x28,0x78,0xc0]
-; X86-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x6f,0xc8]
 ; X86-NEXT:    vpaddd %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xfe,0xc0]
 ; X86-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_pd2udq_256:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vcvttpd2udq %ymm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0xfc,0x29,0x78,0xc8]
 ; X64-NEXT:    vcvttpd2udq %ymm0, %xmm0 # encoding: [0x62,0xf1,0xfc,0x28,0x78,0xc0]
-; X64-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x6f,0xc8]
 ; X64-NEXT:    vpaddd %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xfe,0xc0]
 ; X64-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <4 x i32> @llvm.x86.avx512.mask.cvttpd2udq.256(<4 x double> %x0, <4 x i32> %x1, i8 %x2)
   %res1 = call <4 x i32> @llvm.x86.avx512.mask.cvttpd2udq.256(<4 x double> %x0, <4 x i32> %x1, i8 -1)
   %res2 = add <4 x i32> %res, %res1
   ret <4 x i32> %res2
 }
 
 declare <4 x i32> @llvm.x86.avx512.mask.cvttps2udq.128(<4 x float>, <4 x i32>, i8)
 
 define <4 x i32>@test_int_x86_avx512_mask_cvtt_ps2udq_128(<4 x float> %x0, <4 x i32> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2udq_128:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
 ; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vcvttps2udq %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7c,0x09,0x78,0xc8]
 ; X86-NEXT:    vcvttps2udq %xmm0, %xmm0 # encoding: [0x62,0xf1,0x7c,0x08,0x78,0xc0]
-; X86-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x6f,0xc8]
 ; X86-NEXT:    vpaddd %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xfe,0xc0]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2udq_128:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vcvttps2udq %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7c,0x09,0x78,0xc8]
 ; X64-NEXT:    vcvttps2udq %xmm0, %xmm0 # encoding: [0x62,0xf1,0x7c,0x08,0x78,0xc0]
-; X64-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x6f,0xc8]
 ; X64-NEXT:    vpaddd %xmm0, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0xfe,0xc0]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <4 x i32> @llvm.x86.avx512.mask.cvttps2udq.128(<4 x float> %x0, <4 x i32> %x1, i8 %x2)
   %res1 = call <4 x i32> @llvm.x86.avx512.mask.cvttps2udq.128(<4 x float> %x0, <4 x i32> %x1, i8 -1)
   %res2 = add <4 x i32> %res, %res1
   ret <4 x i32> %res2
 }
 
 declare <8 x i32> @llvm.x86.avx512.mask.cvttps2udq.256(<8 x float>, <8 x i32>, i8)
 
 define <8 x i32>@test_int_x86_avx512_mask_cvtt_ps2udq_256(<8 x float> %x0, <8 x i32> %x1, i8 %x2) {
 ; X86-LABEL: test_int_x86_avx512_mask_cvtt_ps2udq_256:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
 ; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vcvttps2udq %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7c,0x29,0x78,0xc8]
 ; X86-NEXT:    vcvttps2udq %ymm0, %ymm0 # encoding: [0x62,0xf1,0x7c,0x28,0x78,0xc0]
-; X86-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0x6f,0xc8]
 ; X86-NEXT:    vpaddd %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xfe,0xc0]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_int_x86_avx512_mask_cvtt_ps2udq_256:
 ; X64:       # %bb.0:
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vcvttps2udq %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7c,0x29,0x78,0xc8]
 ; X64-NEXT:    vcvttps2udq %ymm0, %ymm0 # encoding: [0x62,0xf1,0x7c,0x28,0x78,0xc0]
-; X64-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0x6f,0xc8]
 ; X64-NEXT:    vpaddd %ymm0, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xf5,0xfe,0xc0]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %res = call <8 x i32> @llvm.x86.avx512.mask.cvttps2udq.256(<8 x float> %x0, <8 x i32> %x1, i8 %x2)
   %res1 = call <8 x i32> @llvm.x86.avx512.mask.cvttps2udq.256(<8 x float> %x0, <8 x i32> %x1, i8 -1)
   %res2 = add <8 x i32> %res, %res1
   ret <8 x i32> %res2
 }