[X86] Lower avx2/avx512f gather intrinsics to X86MaskedGatherSDNode instead of going directly to MachineSDNode.

[craig.topper]

Summary

This sends these intrinsics through isel in a much more normal way. This should allow addressing mode matching in isel to make better use of the displacement field.

[craig.topper]

Update to include tests

[RKSimon]

LGTM

lib/Target/X86/X86ISelLowering.cpp

case TRUNCATE_TO_MEM_VI32: {
     else if (IntrData->Type == TRUNCATE_TO_MEM_VI32)
       ScalarVT = MVT::i32;
 
     Info.memVT = MVT::getVectorVT(ScalarVT, VT.getVectorNumElements());
     Info.align = 1;
     Info.flags |= MachineMemOperand::MOStore;
     break;
   }
+  case GATHER:
+  case GATHER_AVX2: {
+    Info.ptrVal = nullptr;
+    MVT DataVT = MVT::getVT(I.getType());
+    MVT IndexVT = MVT::getVT(I.getArgOperand(2)->getType());
+    unsigned NumElts = std::min(DataVT.getVectorNumElements(),
+                                IndexVT.getVectorNumElements());
+    Info.memVT = MVT::getVectorVT(DataVT.getVectorElementType(), NumElts);
+    Info.align = 1;
+    Info.flags |= MachineMemOperand::MOLoad;
+    break;
+  }
   default:
     return false;
   }
 
   return true;
 }
 
 /// Returns true if the target can instruction select the
static SDValue getAVX2GatherNode(unsigned Opc, SDValue Op, SelectionDAG &DAG,
                                  SDValue Index, SDValue ScaleOp, SDValue Chain,
                                  const X86Subtarget &Subtarget) {
   SDLoc dl(Op);
   auto *C = dyn_cast<ConstantSDNode>(ScaleOp);
   // Scale must be constant.
   if (!C)
     return SDValue();
   SDValue Scale = DAG.getTargetConstant(C->getZExtValue(), dl, MVT::i8);
-  EVT MaskVT = Mask.getValueType();
+  EVT MaskVT = Mask.getValueType().changeVectorElementTypeToInteger();
   SDVTList VTs = DAG.getVTList(Op.getValueType(), MaskVT, MVT::Other);
-  SDValue Disp = DAG.getTargetConstant(0, dl, MVT::i32);
-  SDValue Segment = DAG.getRegister(0, MVT::i32);
   // If source is undef or we know it won't be used, use a zero vector
   // to break register dependency.
   // TODO: use undef instead and let BreakFalseDeps deal with it?
   if (Src.isUndef() || ISD::isBuildVectorAllOnes(Mask.getNode()))
     Src = getZeroVector(Op.getSimpleValueType(), Subtarget, DAG, dl);
-  SDValue Ops[] = {Src, Base, Scale, Index, Disp, Segment, Mask, Chain};
-  SDNode *Res = DAG.getMachineNode(Opc, dl, VTs, Ops);
-  SDValue RetOps[] = { SDValue(Res, 0), SDValue(Res, 2) };
-  return DAG.getMergeValues(RetOps, dl);
+
+  MemIntrinsicSDNode *MemIntr = cast<MemIntrinsicSDNode>(Op);
+
+  SDValue Ops[] = {Chain, Src, Mask, Base, Index, Scale };
+  SDValue Res = DAG.getTargetMemSDNode<X86MaskedGatherSDNode>(
+    VTs, Ops, dl, MemIntr->getMemoryVT(), MemIntr->getMemOperand());
+  return DAG.getMergeValues({ Res, Res.getValue(2) }, dl);
 }
 
-static SDValue getGatherNode(unsigned Opc, SDValue Op, SelectionDAG &DAG,
-                              SDValue Src, SDValue Mask, SDValue Base,
-                              SDValue Index, SDValue ScaleOp, SDValue Chain,
-                              const X86Subtarget &Subtarget) {
+static SDValue getGatherNode(SDValue Op, SelectionDAG &DAG,
   MVT VT = Op.getSimpleValueType();
   SDLoc dl(Op);
   auto *C = dyn_cast<ConstantSDNode>(ScaleOp);
   // Scale must be constant.
   if (!C)
     return SDValue();
   SDValue Scale = DAG.getTargetConstant(C->getZExtValue(), dl, MVT::i8);
   unsigned MinElts = std::min(Index.getSimpleValueType().getVectorNumElements(),
                               VT.getVectorNumElements());
   MVT MaskVT = MVT::getVectorVT(MVT::i1, MinElts);
 
   // We support two versions of the gather intrinsics. One with scalar mask and
   // one with vXi1 mask. Convert scalar to vXi1 if necessary.
   if (Mask.getValueType() != MaskVT)
     Mask = getMaskNode(Mask, MaskVT, Subtarget, DAG, dl);
 
   SDVTList VTs = DAG.getVTList(Op.getValueType(), MaskVT, MVT::Other);
-  SDValue Disp = DAG.getTargetConstant(0, dl, MVT::i32);
-  SDValue Segment = DAG.getRegister(0, MVT::i32);
   // If source is undef or we know it won't be used, use a zero vector
   // to break register dependency.
   // TODO: use undef instead and let BreakFalseDeps deal with it?
   if (Src.isUndef() || ISD::isBuildVectorAllOnes(Mask.getNode()))
     Src = getZeroVector(Op.getSimpleValueType(), Subtarget, DAG, dl);
-  SDValue Ops[] = {Src, Mask, Base, Scale, Index, Disp, Segment, Chain};
-  SDNode *Res = DAG.getMachineNode(Opc, dl, VTs, Ops);
-  SDValue RetOps[] = { SDValue(Res, 0), SDValue(Res, 2) };
-  return DAG.getMergeValues(RetOps, dl);
+
+  MemIntrinsicSDNode *MemIntr = cast<MemIntrinsicSDNode>(Op);
+
+  SDValue Ops[] = {Chain, Src, Mask, Base, Index, Scale };
+  SDValue Res = DAG.getTargetMemSDNode<X86MaskedGatherSDNode>(
+    VTs, Ops, dl, MemIntr->getMemoryVT(), MemIntr->getMemOperand());
+  return DAG.getMergeValues({ Res, Res.getValue(2) }, dl);
 }
 
 static SDValue getScatterNode(unsigned Opc, SDValue Op, SelectionDAG &DAG,
                                SDValue Src, SDValue Mask, SDValue Base,
                                SDValue Index, SDValue ScaleOp, SDValue Chain,
                                const X86Subtarget &Subtarget) {
   SDLoc dl(Op);
   auto *C = dyn_cast<ConstantSDNode>(ScaleOp);
static SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, const X86Subtarget &Subtarget,
   case GATHER: {
   //gather(v1, mask, index, base, scale);
     SDValue Chain = Op.getOperand(0);
     SDValue Src   = Op.getOperand(2);
     SDValue Base  = Op.getOperand(3);
     SDValue Index = Op.getOperand(4);
     SDValue Mask  = Op.getOperand(5);
     SDValue Scale = Op.getOperand(6);
-    return getGatherNode(IntrData->Opc0, Op, DAG, Src, Mask, Base, Index, Scale,
+    return getGatherNode(Op, DAG, Src, Mask, Base, Index, Scale,
                          Chain, Subtarget);
   }
   case SCATTER: {
   //scatter(base, mask, index, v1, scale);
     SDValue Chain = Op.getOperand(0);
     SDValue Base  = Op.getOperand(2);
     SDValue Mask  = Op.getOperand(3);
     SDValue Index = Op.getOperand(4);

lib/Target/X86/X86InstrSSE.td

def Yrm : AVX28I<opc, MRMSrcMem4VOp3, (outs RC256:$dst, RC256:$mask_wb),
             !strconcat(OpcodeStr,
               "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
             [(set (VTy RC256:$dst), (MTy RC256:$mask_wb),
                   (GatherNode256 RC256:$src1, RC256:$mask,
                                 vectoraddr:$src2))]>,
             VEX, VEX_L, Sched<[WriteLoad]>;
 }
 
-let Predicates = [UseAVX2] in {
+let Predicates = [HasAVX2] in {
   let mayLoad = 1, hasSideEffects = 0, Constraints
     = "@earlyclobber $dst,@earlyclobber $mask_wb, $src1 = $dst, $mask = $mask_wb"
     in {
     defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", v2i64, v4i64, mgatherv4i32,
                         mgatherv4i32, VR256, vx128mem, vx256mem>, VEX_W;
     defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", v2i64, v4i64, mgatherv2i64,
                         mgatherv4i64, VR256, vx128mem, vy256mem>, VEX_W;
     defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", v4i32, v8i32, mgatherv4i32,

lib/Target/X86/X86IntrinsicsInfo.h

 #define X86_INTRINSIC_DATA(id, type, op0, op1) \
   { Intrinsic::x86_##id, type, op0, op1 }
 
 /*
  * IntrinsicsWithChain - the table should be sorted by Intrinsic ID - in
  * the alphabetical order.
  */
 static const IntrinsicData IntrinsicsWithChain[] = {
-  X86_INTRINSIC_DATA(avx2_gather_d_d,      GATHER_AVX2, X86::VPGATHERDDrm, 0),
-  X86_INTRINSIC_DATA(avx2_gather_d_d_256,  GATHER_AVX2, X86::VPGATHERDDYrm, 0),
-  X86_INTRINSIC_DATA(avx2_gather_d_pd,     GATHER_AVX2, X86::VGATHERDPDrm, 0),
-  X86_INTRINSIC_DATA(avx2_gather_d_pd_256, GATHER_AVX2, X86::VGATHERDPDYrm, 0),
-  X86_INTRINSIC_DATA(avx2_gather_d_ps,     GATHER_AVX2, X86::VGATHERDPSrm, 0),
-  X86_INTRINSIC_DATA(avx2_gather_d_ps_256, GATHER_AVX2, X86::VGATHERDPSYrm, 0),
-  X86_INTRINSIC_DATA(avx2_gather_d_q,      GATHER_AVX2, X86::VPGATHERDQrm, 0),
-  X86_INTRINSIC_DATA(avx2_gather_d_q_256,  GATHER_AVX2, X86::VPGATHERDQYrm, 0),
-  X86_INTRINSIC_DATA(avx2_gather_q_d,      GATHER_AVX2, X86::VPGATHERQDrm, 0),
-  X86_INTRINSIC_DATA(avx2_gather_q_d_256,  GATHER_AVX2, X86::VPGATHERQDYrm, 0),
-  X86_INTRINSIC_DATA(avx2_gather_q_pd,     GATHER_AVX2, X86::VGATHERQPDrm, 0),
-  X86_INTRINSIC_DATA(avx2_gather_q_pd_256, GATHER_AVX2, X86::VGATHERQPDYrm, 0),
-  X86_INTRINSIC_DATA(avx2_gather_q_ps,     GATHER_AVX2, X86::VGATHERQPSrm, 0),
-  X86_INTRINSIC_DATA(avx2_gather_q_ps_256, GATHER_AVX2, X86::VGATHERQPSYrm, 0),
-  X86_INTRINSIC_DATA(avx2_gather_q_q,      GATHER_AVX2, X86::VPGATHERQQrm, 0),
-  X86_INTRINSIC_DATA(avx2_gather_q_q_256,  GATHER_AVX2, X86::VPGATHERQQYrm, 0),
-
-  X86_INTRINSIC_DATA(avx512_gather_dpd_512, GATHER, X86::VGATHERDPDZrm, 0),
-  X86_INTRINSIC_DATA(avx512_gather_dpi_512, GATHER, X86::VPGATHERDDZrm, 0),
-  X86_INTRINSIC_DATA(avx512_gather_dpq_512, GATHER, X86::VPGATHERDQZrm, 0),
-  X86_INTRINSIC_DATA(avx512_gather_dps_512, GATHER, X86::VGATHERDPSZrm, 0),
-  X86_INTRINSIC_DATA(avx512_gather_qpd_512, GATHER, X86::VGATHERQPDZrm, 0),
-  X86_INTRINSIC_DATA(avx512_gather_qpi_512, GATHER, X86::VPGATHERQDZrm, 0),
-  X86_INTRINSIC_DATA(avx512_gather_qpq_512, GATHER, X86::VPGATHERQQZrm, 0),
-  X86_INTRINSIC_DATA(avx512_gather_qps_512, GATHER, X86::VGATHERQPSZrm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3div2_df, GATHER, X86::VGATHERQPDZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3div2_di, GATHER, X86::VPGATHERQQZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3div4_df, GATHER, X86::VGATHERQPDZ256rm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3div4_di, GATHER, X86::VPGATHERQQZ256rm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3div4_sf, GATHER, X86::VGATHERQPSZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3div4_si, GATHER, X86::VPGATHERQDZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3div8_sf, GATHER, X86::VGATHERQPSZ256rm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3div8_si, GATHER, X86::VPGATHERQDZ256rm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3siv2_df, GATHER, X86::VGATHERDPDZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3siv2_di, GATHER, X86::VPGATHERDQZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3siv4_df, GATHER, X86::VGATHERDPDZ256rm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3siv4_di, GATHER, X86::VPGATHERDQZ256rm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3siv4_sf, GATHER, X86::VGATHERDPSZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3siv4_si, GATHER, X86::VPGATHERDDZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3siv8_sf, GATHER, X86::VGATHERDPSZ256rm, 0),
-  X86_INTRINSIC_DATA(avx512_gather3siv8_si, GATHER, X86::VPGATHERDDZ256rm, 0),
+  X86_INTRINSIC_DATA(avx2_gather_d_d,      GATHER_AVX2, 0, 0),
+  X86_INTRINSIC_DATA(avx2_gather_d_d_256,  GATHER_AVX2, 0, 0),
+  X86_INTRINSIC_DATA(avx2_gather_d_pd,     GATHER_AVX2, 0, 0),
+  X86_INTRINSIC_DATA(avx2_gather_d_pd_256, GATHER_AVX2, 0, 0),
+  X86_INTRINSIC_DATA(avx2_gather_d_ps,     GATHER_AVX2, 0, 0),
+  X86_INTRINSIC_DATA(avx2_gather_d_ps_256, GATHER_AVX2, 0, 0),
+  X86_INTRINSIC_DATA(avx2_gather_d_q,      GATHER_AVX2, 0, 0),
+  X86_INTRINSIC_DATA(avx2_gather_d_q_256,  GATHER_AVX2, 0, 0),
+  X86_INTRINSIC_DATA(avx2_gather_q_d,      GATHER_AVX2, 0, 0),
+  X86_INTRINSIC_DATA(avx2_gather_q_d_256,  GATHER_AVX2, 0, 0),
+  X86_INTRINSIC_DATA(avx2_gather_q_pd,     GATHER_AVX2, 0, 0),
+  X86_INTRINSIC_DATA(avx2_gather_q_pd_256, GATHER_AVX2, 0, 0),
+  X86_INTRINSIC_DATA(avx2_gather_q_ps,     GATHER_AVX2, 0, 0),
+  X86_INTRINSIC_DATA(avx2_gather_q_ps_256, GATHER_AVX2, 0, 0),
+  X86_INTRINSIC_DATA(avx2_gather_q_q,      GATHER_AVX2, 0, 0),
+  X86_INTRINSIC_DATA(avx2_gather_q_q_256,  GATHER_AVX2, 0, 0),
+
+  X86_INTRINSIC_DATA(avx512_gather_dpd_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather_dpi_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather_dpq_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather_dps_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather_qpd_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather_qpi_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather_qpq_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather_qps_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3div2_df, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3div2_di, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3div4_df, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3div4_di, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3div4_sf, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3div4_si, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3div8_sf, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3div8_si, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3siv2_df, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3siv2_di, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3siv4_df, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3siv4_di, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3siv4_sf, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3siv4_si, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3siv8_sf, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_gather3siv8_si, GATHER, 0, 0),
 
   X86_INTRINSIC_DATA(avx512_gatherpf_dpd_512, PREFETCH,
                      X86::VGATHERPF0DPDm, X86::VGATHERPF1DPDm),
   X86_INTRINSIC_DATA(avx512_gatherpf_dps_512, PREFETCH,
                      X86::VGATHERPF0DPSm, X86::VGATHERPF1DPSm),
   X86_INTRINSIC_DATA(avx512_gatherpf_qpd_512, PREFETCH,
                      X86::VGATHERPF0QPDm, X86::VGATHERPF1QPDm),
   X86_INTRINSIC_DATA(avx512_gatherpf_qps_512, PREFETCH,
                      X86::VGATHERPF0QPSm, X86::VGATHERPF1QPSm),
 
-  X86_INTRINSIC_DATA(avx512_mask_gather_dpd_512, GATHER, X86::VGATHERDPDZrm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather_dpi_512, GATHER, X86::VPGATHERDDZrm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather_dpq_512, GATHER, X86::VPGATHERDQZrm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather_dps_512, GATHER, X86::VGATHERDPSZrm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather_qpd_512, GATHER, X86::VGATHERQPDZrm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather_qpi_512, GATHER, X86::VPGATHERQDZrm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather_qpq_512, GATHER, X86::VPGATHERQQZrm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather_qps_512, GATHER, X86::VGATHERQPSZrm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3div2_df, GATHER, X86::VGATHERQPDZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3div2_di, GATHER, X86::VPGATHERQQZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3div4_df, GATHER, X86::VGATHERQPDZ256rm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3div4_di, GATHER, X86::VPGATHERQQZ256rm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3div4_sf, GATHER, X86::VGATHERQPSZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3div4_si, GATHER, X86::VPGATHERQDZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3div8_sf, GATHER, X86::VGATHERQPSZ256rm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3div8_si, GATHER, X86::VPGATHERQDZ256rm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3siv2_df, GATHER, X86::VGATHERDPDZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3siv2_di, GATHER, X86::VPGATHERDQZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3siv4_df, GATHER, X86::VGATHERDPDZ256rm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3siv4_di, GATHER, X86::VPGATHERDQZ256rm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3siv4_sf, GATHER, X86::VGATHERDPSZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3siv4_si, GATHER, X86::VPGATHERDDZ128rm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3siv8_sf, GATHER, X86::VGATHERDPSZ256rm, 0),
-  X86_INTRINSIC_DATA(avx512_mask_gather3siv8_si, GATHER, X86::VPGATHERDDZ256rm, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather_dpd_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather_dpi_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather_dpq_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather_dps_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather_qpd_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather_qpi_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather_qpq_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather_qps_512, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3div2_df, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3div2_di, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3div4_df, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3div4_di, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3div4_sf, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3div4_si, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3div8_sf, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3div8_si, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3siv2_df, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3siv2_di, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3siv4_df, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3siv4_di, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3siv4_sf, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3siv4_si, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3siv8_sf, GATHER, 0, 0),
+  X86_INTRINSIC_DATA(avx512_mask_gather3siv8_si, GATHER, 0, 0),
 
   X86_INTRINSIC_DATA(avx512_mask_pmov_db_mem_128, TRUNCATE_TO_MEM_VI8,
                      X86ISD::VTRUNC, 0),
   X86_INTRINSIC_DATA(avx512_mask_pmov_db_mem_256, TRUNCATE_TO_MEM_VI8,
                      X86ISD::VTRUNC, 0),
   X86_INTRINSIC_DATA(avx512_mask_pmov_db_mem_512, TRUNCATE_TO_MEM_VI8,
                      X86ISD::VTRUNC, 0),
   X86_INTRINSIC_DATA(avx512_mask_pmov_dw_mem_128, TRUNCATE_TO_MEM_VI16,

test/CodeGen/X86/avx2-gather.ll

 }
 
 @x = global [1024 x float] zeroinitializer, align 16
 
 define <4 x float> @gather_global(<4 x i64>, i32* nocapture readnone) {
 ; X32-LABEL: gather_global:
 ; X32:       # %bb.0:
 ; X32-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
-; X32-NEXT:    movl $x, %eax
 ; X32-NEXT:    vxorps %xmm1, %xmm1, %xmm1
-; X32-NEXT:    vgatherqps %xmm2, (%eax,%ymm0,4), %xmm1
+; X32-NEXT:    vgatherqps %xmm2, x(,%ymm0,4), %xmm1
 ; X32-NEXT:    vmovaps %xmm1, %xmm0
 ; X32-NEXT:    vzeroupper
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: gather_global:
 ; X64:       # %bb.0:
 ; X64-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
-; X64-NEXT:    movl $x, %eax
 ; X64-NEXT:    vxorps %xmm1, %xmm1, %xmm1
-; X64-NEXT:    vgatherqps %xmm2, (%rax,%ymm0,4), %xmm1
+; X64-NEXT:    vgatherqps %xmm2, x(,%ymm0,4), %xmm1
 ; X64-NEXT:    vmovaps %xmm1, %xmm0
 ; X64-NEXT:    vzeroupper
 ; X64-NEXT:    retq
   %3 = tail call <4 x float> @llvm.x86.avx2.gather.q.ps.256(<4 x float> zeroinitializer, i8* bitcast ([1024 x float]* @x to i8*), <4 x i64> %0, <4 x float> <float 0xFFFFFFFFE0000000, float 0xFFFFFFFFE0000000, float 0xFFFFFFFFE0000000, float 0xFFFFFFFFE0000000>, i8 4)
   ret <4 x float> %3
 }
 declare <4 x float> @llvm.x86.avx2.gather.q.ps.256(<4 x float>, i8*, <4 x i64>, <4 x float>, i8)

test/CodeGen/X86/avx512-gather-scatter-intrin.ll

 }
 
 @x = global [1024 x float] zeroinitializer, align 16
 
 define <8 x float> @gather_global(<8 x i64>, i32* nocapture readnone) {
 ; CHECK-LABEL: gather_global:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    kxnorw %k0, %k0, %k1
-; CHECK-NEXT:    movl $x, %eax
 ; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
-; CHECK-NEXT:    vgatherqps (%rax,%zmm0,4), %ymm1 {%k1}
+; CHECK-NEXT:    vgatherqps x(,%zmm0,4), %ymm1 {%k1}
 ; CHECK-NEXT:    vmovaps %ymm1, %ymm0
 ; CHECK-NEXT:    retq
   %3 = tail call <8 x float> @llvm.x86.avx512.mask.gather.qps.512(<8 x float> zeroinitializer, i8* bitcast ([1024 x float]* @x to i8*), <8 x i64> %0, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, i32 4)
   ret <8 x float> %3
 }
 
 declare <16 x float> @llvm.x86.avx512.mask.gather.dps.512(<16 x float>, i8*, <16 x i32>, <16 x i1>, i32)
 declare <8 x double> @llvm.x86.avx512.mask.gather.dpd.512(<8 x double>, i8*, <8 x i32>, <8 x i1>, i32)