from X86ISelLowering.cpp

// LowerUINT_TO_FP_i64 - 64-bit unsigned integer to double expansion.
SDValue X86TargetLowering::LowerUINT_TO_FP_i64(SDValue Op,
                                               SelectionDAG &DAG) const {
  // This algorithm is not obvious. Here it is what we're trying to output:
  /*
     movq       %rax,  %xmm0
     punpckldq  (c0),  %xmm0  // c0: (uint4){ 0x43300000U, 0x45300000U, 0U, 0U }
     subpd      (c1),  %xmm0  // c1: (double2){ 0x1.0p52, 0x1.0p52 * 0x1.0p32 }
     #ifdef __SSE3__
       haddpd   %xmm0, %xmm0
     #else
       pshufd   $0x4e, %xmm0, %xmm1
       addpd    %xmm1, %xmm0
     #endif
  */

  SDLoc dl(Op);
  LLVMContext *Context = DAG.getContext();

  // Build some magic constants.
  static const uint32_t CV0[] = { 0x43300000, 0x45300000, 0, 0 };
  Constant *C0 = ConstantDataVector::get(*Context, CV0);
  auto PtrVT = getPointerTy(DAG.getDataLayout());
  SDValue CPIdx0 = DAG.getConstantPool(C0, PtrVT, 16);

  SmallVector<Constant*,2> CV1;
  CV1.push_back(
    ConstantFP::get(*Context, APFloat(APFloat::IEEEdouble,
                                      APInt(64, 0x4330000000000000ULL))));
  CV1.push_back(
    ConstantFP::get(*Context, APFloat(APFloat::IEEEdouble,
                                      APInt(64, 0x4530000000000000ULL))));
  Constant *C1 = ConstantVector::get(CV1);
  SDValue CPIdx1 = DAG.getConstantPool(C1, PtrVT, 16);

  // Load the 64-bit value into an XMM register.
  SDValue XR1 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2i64,
                            Op.getOperand(0));
  SDValue CLod0 =
      DAG.getLoad(MVT::v4i32, dl, DAG.getEntryNode(), CPIdx0,
                  MachinePointerInfo::getConstantPool(DAG.getMachineFunction()),
                  false, false, false, 16);
  SDValue Unpck1 =
      getUnpackl(DAG, dl, MVT::v4i32, DAG.getBitcast(MVT::v4i32, XR1), CLod0);

  SDValue CLod1 =
      DAG.getLoad(MVT::v2f64, dl, CLod0.getValue(1), CPIdx1,
                  MachinePointerInfo::getConstantPool(DAG.getMachineFunction()),
                  false, false, false, 16);
  SDValue XR2F = DAG.getBitcast(MVT::v2f64, Unpck1);
  SDValue Sub = DAG.getNode(ISD::FSUB, dl, MVT::v2f64, XR2F, CLod1);
  SDValue Result;

  if (Subtarget->hasSSE3()) {
    // FIXME: The 'haddpd' instruction may be slower than 'movhlps + addsd'.
    Result = DAG.getNode(X86ISD::FHADD, dl, MVT::v2f64, Sub, Sub);
  } else {
    SDValue S2F = DAG.getBitcast(MVT::v4i32, Sub);
    SDValue Shuffle = getTargetShuffleNode(X86ISD::PSHUFD, dl, MVT::v4i32,
                                           S2F, 0x4E, DAG);
    Result = DAG.getNode(ISD::FADD, dl, MVT::v2f64,
                         DAG.getBitcast(MVT::v2f64, Shuffle), Sub);
  }

  return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64, Result,
                     DAG.getIntPtrConstant(0, dl));
}