[RISCV] Inline scalar ceil/floor/trunc/rint/round/roundeven.

[craig.topper]

Summary

This avoids the call overhead as well as the the save/restore of
fflags and the snan handling in the libm function.

The save/restore of fflags and snan handling are needed to be
correct for -ftrapping-math. I think we can ignore them in the
default environment.

The inline sequence will generate an invalid exception for nan
and an inexact exception if fractional bits are discarded.

I've used a custom inserter to explicitly create the control flow
around the float->int->float conversion.

We can probably avoid the final fsgnj after the conversion for
no signed zeros FMF, but I'll leave that for future work.

Note the comparison constant is slightly different than glibc uses.
They use 1<<53 for double, I'm using 1<<52. I believe either are valid.
Numbers >= 1<<52 can't have any fractional bits. It's ok to do the
float->int->float conversion on numbers between 1<<53 and 1<<52 since
they will all fit in 64. We only have a problem if the double can't fit
in i64

[reames]

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

4102

Can you precommit a change to add the vector naming so that disappears from this diff?

10813

Are SinkMBB and TailMBB the same?

llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp

528

Can you precommit the style change here?

[craig.topper]

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

10813

Yeah. I was looking at multiple custom inserters across multiple targets and must have copied the comment from one and used the name from another.

[craig.topper]

Rebase

[reames]

LGTM

llvm/lib/Target/RISCV/RISCVISelLowering.h

enum NodeType : unsigned {
   FCVT_X,
   FCVT_XU,
   // FP to 32 bit int conversions for RV64. These are used to keep track of the
   // result being sign extended to 64 bit. These saturate out of range inputs.
   // Used for FP_TO_S/UINT and FP_TO_S/UINT_SAT lowering. Rounding mode
   // is passed as a TargetConstant operand using the RISCVFPRndMode enum.
   FCVT_W_RV64,
   FCVT_WU_RV64,
+
+  // Rounds an FP value to its corresponding integer in the same FP format.
+  // First operand is the value to round, the second operand is the largest
+  // integer that can be represented exactly in the FP format. This will be
+  // expanded into multiple instructions and basic blocks with a custom
+  // inserter.
+  FROUND,
+
   // READ_CYCLE_WIDE - A read of the 64-bit cycle CSR on a 32-bit target
   // (returns (Lo, Hi)). It takes a chain operand.
   READ_CYCLE_WIDE,
   // brev8, orc.b, zip, and unzip from Zbb and Zbkb. All operands are i32 or
   // XLenVT.
   BREV8,
   ORC_B,
   ZIP,

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

static const unsigned FPOpToExpand[] = {
       ISD::FSIN, ISD::FCOS,       ISD::FSINCOS,   ISD::FPOW,
       ISD::FREM, ISD::FP16_TO_FP, ISD::FP_TO_FP16};
 
   if (Subtarget.hasStdExtZfh())
     setOperationAction(ISD::BITCAST, MVT::i16, Custom);
 
   if (Subtarget.hasStdExtZfh()) {
     setOperationAction(FPLegalNodeTypes, MVT::f16, Legal);
+    setOperationAction(ISD::FCEIL, MVT::f16, Custom);
+    setOperationAction(ISD::FFLOOR, MVT::f16, Custom);
+    setOperationAction(ISD::FTRUNC, MVT::f16, Custom);
+    setOperationAction(ISD::FRINT, MVT::f16, Custom);
+    setOperationAction(ISD::FROUND, MVT::f16, Custom);
+    setOperationAction(ISD::FROUNDEVEN, MVT::f16, Custom);
     setOperationAction(ISD::STRICT_FP_ROUND, MVT::f16, Legal);
     setOperationAction(ISD::STRICT_FP_EXTEND, MVT::f32, Legal);
     setCondCodeAction(FPCCToExpand, MVT::f16, Expand);
     setOperationAction(ISD::SELECT_CC, MVT::f16, Expand);
     setOperationAction(ISD::SELECT, MVT::f16, Custom);
     setOperationAction(ISD::BR_CC, MVT::f16, Expand);
 
-    setOperationAction({ISD::FREM, ISD::FCEIL, ISD::FFLOOR, ISD::FNEARBYINT,
-                        ISD::FRINT, ISD::FROUND, ISD::FROUNDEVEN, ISD::FTRUNC,
-                        ISD::FPOW, ISD::FPOWI, ISD::FCOS, ISD::FSIN,
-                        ISD::FSINCOS, ISD::FEXP, ISD::FEXP2, ISD::FLOG,
-                        ISD::FLOG2, ISD::FLOG10},
+    setOperationAction({ISD::FREM, ISD::FNEARBYINT, ISD::FPOW, ISD::FPOWI,
+                        ISD::FCOS, ISD::FSIN, ISD::FSINCOS, ISD::FEXP,
+                        ISD::FEXP2, ISD::FLOG, ISD::FLOG2, ISD::FLOG10},
                        MVT::f16, Promote);
 
     // FIXME: Need to promote f16 STRICT_* to f32 libcalls, but we don't have
     // complete support for all operations in LegalizeDAG.
     setOperationAction({ISD::STRICT_FCEIL, ISD::STRICT_FFLOOR,
                         ISD::STRICT_FNEARBYINT, ISD::STRICT_FRINT,
                         ISD::STRICT_FROUND, ISD::STRICT_FROUNDEVEN,
                         ISD::STRICT_FTRUNC},
                        MVT::f16, Promote);
 
     // We need to custom promote this.
     if (Subtarget.is64Bit())
       setOperationAction(ISD::FPOWI, MVT::i32, Custom);
   }
 
   if (Subtarget.hasStdExtF()) {
     setOperationAction(FPLegalNodeTypes, MVT::f32, Legal);
+    setOperationAction(ISD::FCEIL, MVT::f32, Custom);
+    setOperationAction(ISD::FFLOOR, MVT::f32, Custom);
+    setOperationAction(ISD::FTRUNC, MVT::f32, Custom);
+    setOperationAction(ISD::FRINT, MVT::f32, Custom);
+    setOperationAction(ISD::FROUND, MVT::f32, Custom);
+    setOperationAction(ISD::FROUNDEVEN, MVT::f32, Custom);
     setCondCodeAction(FPCCToExpand, MVT::f32, Expand);
     setOperationAction(ISD::SELECT_CC, MVT::f32, Expand);
     setOperationAction(ISD::SELECT, MVT::f32, Custom);
     setOperationAction(ISD::BR_CC, MVT::f32, Expand);
     setOperationAction(FPOpToExpand, MVT::f32, Expand);
     setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::f16, Expand);
     setTruncStoreAction(MVT::f32, MVT::f16, Expand);
   }
 
   if (Subtarget.hasStdExtF() && Subtarget.is64Bit())
     setOperationAction(ISD::BITCAST, MVT::i32, Custom);
 
   if (Subtarget.hasStdExtD()) {
     setOperationAction(FPLegalNodeTypes, MVT::f64, Legal);
+    if (Subtarget.is64Bit()) {
+      setOperationAction(ISD::FCEIL, MVT::f64, Custom);
+      setOperationAction(ISD::FFLOOR, MVT::f64, Custom);
+      setOperationAction(ISD::FTRUNC, MVT::f64, Custom);
+      setOperationAction(ISD::FRINT, MVT::f64, Custom);
+      setOperationAction(ISD::FROUND, MVT::f64, Custom);
+      setOperationAction(ISD::FROUNDEVEN, MVT::f64, Custom);
+    }
     setOperationAction(ISD::STRICT_FP_ROUND, MVT::f32, Legal);
     setOperationAction(ISD::STRICT_FP_EXTEND, MVT::f64, Legal);
     setCondCodeAction(FPCCToExpand, MVT::f64, Expand);
     setOperationAction(ISD::SELECT_CC, MVT::f64, Expand);
     setOperationAction(ISD::SELECT, MVT::f64, Custom);
     setOperationAction(ISD::BR_CC, MVT::f64, Expand);
     setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f32, Expand);
     setTruncStoreAction(MVT::f64, MVT::f32, Expand);
static RISCVFPRndMode::RoundingMode matchRoundingOp(unsigned Opc) {
   case ISD::VP_FFLOOR:
     return RISCVFPRndMode::RDN;
   case ISD::FCEIL:
   case ISD::VP_FCEIL:
     return RISCVFPRndMode::RUP;
   case ISD::FROUND:
   case ISD::VP_FROUND:
     return RISCVFPRndMode::RMM;
+  case ISD::FRINT:
+    return RISCVFPRndMode::DYN;
   }
 
   return RISCVFPRndMode::Invalid;
 }
 
 // Expand vector FTRUNC, FCEIL, FFLOOR, FROUND, VP_FCEIL, VP_FFLOOR, VP_FROUND
 // VP_FROUNDEVEN and VP_FROUNDTOZERO by converting to the integer domain and
 // back. Taking care to avoid converting values that are nan or already correct.
Truncated = DAG.getNode(RISCVISD::FCOPYSIGN_VL, DL, ContainerVT, Truncated,
                           Src, Src, Mask, VL);
 
   if (!VT.isFixedLengthVector())
     return Truncated;
 
   return convertFromScalableVector(VT, Truncated, DAG, Subtarget);
 }
 
+static SDValue
+lowerFTRUNC_FCEIL_FFLOOR_FROUND(SDValue Op, SelectionDAG &DAG,
+                                const RISCVSubtarget &Subtarget) {
+  MVT VT = Op.getSimpleValueType();
+  if (VT.isVector())
+    return lowerVectorFTRUNC_FCEIL_FFLOOR_FROUND(Op, DAG, Subtarget);
+
+  if (DAG.shouldOptForSize())
+    return SDValue();
+
+  SDLoc DL(Op);
+  SDValue Src = Op.getOperand(0);
+
+  // Create an integer the size of the mantissa with the MSB set. This and all
+  // values larger than it don't have any fractional bits so don't need to be
+  // converted.
+  const fltSemantics &FltSem = DAG.EVTToAPFloatSemantics(VT);
+  unsigned Precision = APFloat::semanticsPrecision(FltSem);
+  APFloat MaxVal = APFloat(FltSem);
+  MaxVal.convertFromAPInt(APInt::getOneBitSet(Precision, Precision - 1),
+                          /*IsSigned*/ false, APFloat::rmNearestTiesToEven);
+  SDValue MaxValNode = DAG.getConstantFP(MaxVal, DL, VT);
+
+  RISCVFPRndMode::RoundingMode FRM = matchRoundingOp(Op.getOpcode());
+  return DAG.getNode(RISCVISD::FROUND, DL, VT, Src, MaxValNode,
+                     DAG.getTargetConstant(FRM, DL, Subtarget.getXLenVT()));
+}
+
 struct VIDSequence {
   int64_t StepNumerator;
   unsigned StepDenominator;
   int64_t Addend;
 };
 
 static Optional<uint64_t> getExactInteger(const APFloat &APF,
                                           uint32_t BitWidth) {
case ISD::UINT_TO_FP: {
     return convertFromScalableVector(VT, Src, DAG, Subtarget);
   }
   case ISD::FP_TO_SINT_SAT:
   case ISD::FP_TO_UINT_SAT:
     return lowerFP_TO_INT_SAT(Op, DAG, Subtarget);
   case ISD::FTRUNC:
   case ISD::FCEIL:
   case ISD::FFLOOR:
+  case ISD::FRINT:
   case ISD::FROUND:
   case ISD::FROUNDEVEN:
-    return lowerVectorFTRUNC_FCEIL_FFLOOR_FROUND(Op, DAG, Subtarget);
+    return lowerFTRUNC_FCEIL_FFLOOR_FROUND(Op, DAG, Subtarget);
   case ISD::VECREDUCE_ADD:
   case ISD::VECREDUCE_UMAX:
   case ISD::VECREDUCE_SMAX:
   case ISD::VECREDUCE_UMIN:
   case ISD::VECREDUCE_SMIN:
     return lowerVECREDUCE(Op, DAG);
   case ISD::VECREDUCE_AND:
   case ISD::VECREDUCE_OR:
case ISD::EXPERIMENTAL_VP_STRIDED_LOAD:
     return lowerVPStridedLoad(Op, DAG);
   case ISD::EXPERIMENTAL_VP_STRIDED_STORE:
     return lowerVPStridedStore(Op, DAG);
   case ISD::VP_FCEIL:
   case ISD::VP_FFLOOR:
   case ISD::VP_FROUND:
   case ISD::VP_FROUNDEVEN:
   case ISD::VP_FROUNDTOZERO:
     return lowerVectorFTRUNC_FCEIL_FFLOOR_FROUND(Op, DAG, Subtarget);
   }
 }
 
 static SDValue getTargetNode(GlobalAddressSDNode *N, SDLoc DL, EVT Ty,
                              SelectionDAG &DAG, unsigned Flags) {
   return DAG.getTargetGlobalAddress(N->getGlobal(), DL, Ty, 0, Flags);
 }
 
emitVFCVT_RM_MASK(MachineInstr &MI, MachineBasicBlock *BB, unsigned Opcode) {
   BuildMI(*BB, MI, DL, TII.get(RISCV::WriteFRM))
       .addReg(SavedFRM, RegState::Kill);
 
   // Erase the pseudoinstruction.
   MI.eraseFromParent();
   return BB;
 }
 
+static MachineBasicBlock *emitFROUND(MachineInstr &MI, MachineBasicBlock *MBB,
+                                     const RISCVSubtarget &Subtarget) {
+  unsigned CmpOpc, F2IOpc, I2FOpc, FSGNJOpc, FSGNJXOpc;
+  const TargetRegisterClass *RC;
+  switch (MI.getOpcode()) {
+  default:
+    llvm_unreachable("Unexpected opcode");
+  case RISCV::PseudoFROUND_H:
+    CmpOpc = RISCV::FLT_H;
+    F2IOpc = RISCV::FCVT_W_H;
+    I2FOpc = RISCV::FCVT_H_W;
+    FSGNJOpc = RISCV::FSGNJ_H;
+    FSGNJXOpc = RISCV::FSGNJX_H;
+    RC = &RISCV::FPR16RegClass;
+    break;
+  case RISCV::PseudoFROUND_S:
+    CmpOpc = RISCV::FLT_S;
+    F2IOpc = RISCV::FCVT_W_S;
+    I2FOpc = RISCV::FCVT_S_W;
+    FSGNJOpc = RISCV::FSGNJ_S;
+    FSGNJXOpc = RISCV::FSGNJX_S;
+    RC = &RISCV::FPR32RegClass;
+    break;
+  case RISCV::PseudoFROUND_D:
+    assert(Subtarget.is64Bit() && "Expected 64-bit GPR.");
+    CmpOpc = RISCV::FLT_D;
+    F2IOpc = RISCV::FCVT_L_D;
+    I2FOpc = RISCV::FCVT_D_L;
+    FSGNJOpc = RISCV::FSGNJ_D;
+    FSGNJXOpc = RISCV::FSGNJX_D;
+    RC = &RISCV::FPR64RegClass;
+    break;
+  }
+
+  const BasicBlock *BB = MBB->getBasicBlock();
+  DebugLoc DL = MI.getDebugLoc();
+  MachineFunction::iterator I = ++MBB->getIterator();
+
+  MachineFunction *F = MBB->getParent();
+  MachineBasicBlock *CvtMBB = F->CreateMachineBasicBlock(BB);
+  MachineBasicBlock *DoneMBB = F->CreateMachineBasicBlock(BB);
+
+  F->insert(I, CvtMBB);
+  F->insert(I, DoneMBB);
+  // Move all instructions after the sequence to DoneMBB.
+  DoneMBB->splice(DoneMBB->end(), MBB, MachineBasicBlock::iterator(MI),
+                  MBB->end());
+  // Update machine-CFG edges by transferring all successors of the current
+  // block to the new block which will contain the Phi nodes for the selects.
+  DoneMBB->transferSuccessorsAndUpdatePHIs(MBB);
+  // Set the successors for MBB.
+  MBB->addSuccessor(CvtMBB);
+  MBB->addSuccessor(DoneMBB);
+
+  Register DstReg = MI.getOperand(0).getReg();
+  Register SrcReg = MI.getOperand(1).getReg();
+  Register MaxReg = MI.getOperand(2).getReg();
+  int64_t FRM = MI.getOperand(3).getImm();
+
+  const RISCVInstrInfo &TII = *Subtarget.getInstrInfo();
+  MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
+
+  Register FabsReg = MRI.createVirtualRegister(RC);
+  BuildMI(MBB, DL, TII.get(FSGNJXOpc), FabsReg).addReg(SrcReg).addReg(SrcReg);
+
+  // Compare the FP value to the max value.
+  Register CmpReg = MRI.createVirtualRegister(&RISCV::GPRRegClass);
+  auto MIB =
+      BuildMI(MBB, DL, TII.get(CmpOpc), CmpReg).addReg(FabsReg).addReg(MaxReg);
+  if (MI.getFlag(MachineInstr::MIFlag::NoFPExcept))
+    MIB->setFlag(MachineInstr::MIFlag::NoFPExcept);
+
+  // Insert branch.
+  BuildMI(MBB, DL, TII.get(RISCV::BEQ))
+      .addReg(CmpReg)
+      .addReg(RISCV::X0)
+      .addMBB(DoneMBB);
+
+  CvtMBB->addSuccessor(DoneMBB);
+
+  // Convert to integer.
+  Register F2IReg = MRI.createVirtualRegister(&RISCV::GPRRegClass);
+  MIB = BuildMI(CvtMBB, DL, TII.get(F2IOpc), F2IReg).addReg(SrcReg).addImm(FRM);
+  if (MI.getFlag(MachineInstr::MIFlag::NoFPExcept))
+    MIB->setFlag(MachineInstr::MIFlag::NoFPExcept);
+
+  // Convert back to FP.
+  Register I2FReg = MRI.createVirtualRegister(RC);
+  MIB = BuildMI(CvtMBB, DL, TII.get(I2FOpc), I2FReg).addReg(F2IReg).addImm(FRM);
+  if (MI.getFlag(MachineInstr::MIFlag::NoFPExcept))
+    MIB->setFlag(MachineInstr::MIFlag::NoFPExcept);
+
+  // Restore the sign bit.
+  Register CvtReg = MRI.createVirtualRegister(RC);
+  BuildMI(CvtMBB, DL, TII.get(FSGNJOpc), CvtReg).addReg(I2FReg).addReg(SrcReg);
+
+  // Merge the results.
+  BuildMI(*DoneMBB, DoneMBB->begin(), DL, TII.get(RISCV::PHI), DstReg)
+      .addReg(SrcReg)
+      .addMBB(MBB)
+      .addReg(CvtReg)
+      .addMBB(CvtMBB);
+
+  MI.eraseFromParent();
+  return DoneMBB;
+}
+
 MachineBasicBlock *
 RISCVTargetLowering::EmitInstrWithCustomInserter(MachineInstr &MI,
                                                  MachineBasicBlock *BB) const {
   switch (MI.getOpcode()) {
   default:
     llvm_unreachable("Unexpected instr type to insert");
   case RISCV::ReadCycleWide:
     assert(!Subtarget.is64Bit() &&
RISCVTargetLowering::EmitInstrWithCustomInserter(MachineInstr &MI,
   case RISCV::PseudoVFCVT_RM_X_F_V_M4_MASK:
     return emitVFCVT_RM_MASK(MI, BB, RISCV::PseudoVFCVT_X_F_V_M4_MASK);
   case RISCV::PseudoVFCVT_RM_X_F_V_M8_MASK:
     return emitVFCVT_RM_MASK(MI, BB, RISCV::PseudoVFCVT_X_F_V_M8_MASK);
   case RISCV::PseudoVFCVT_RM_X_F_V_MF2_MASK:
     return emitVFCVT_RM_MASK(MI, BB, RISCV::PseudoVFCVT_X_F_V_MF2_MASK);
   case RISCV::PseudoVFCVT_RM_X_F_V_MF4_MASK:
     return emitVFCVT_RM_MASK(MI, BB, RISCV::PseudoVFCVT_X_F_V_MF4_MASK);
+  case RISCV::PseudoFROUND_H:
+  case RISCV::PseudoFROUND_S:
+  case RISCV::PseudoFROUND_D:
+    return emitFROUND(MI, BB, Subtarget);
   }
 }
 
 void RISCVTargetLowering::AdjustInstrPostInstrSelection(MachineInstr &MI,
                                                         SDNode *Node) const {
   // Add FRM dependency to any instructions with dynamic rounding mode.
   unsigned Opc = MI.getOpcode();
   auto Idx = RISCV::getNamedOperandIdx(Opc, RISCV::OpName::frm);
#define NODE_NAME_CASE(NODE)                                                   \
   NODE_NAME_CASE(FMV_W_X_RV64)
   NODE_NAME_CASE(FMV_X_ANYEXTW_RV64)
   NODE_NAME_CASE(FCVT_X)
   NODE_NAME_CASE(FCVT_XU)
   NODE_NAME_CASE(FCVT_W_RV64)
   NODE_NAME_CASE(FCVT_WU_RV64)
   NODE_NAME_CASE(STRICT_FCVT_W_RV64)
   NODE_NAME_CASE(STRICT_FCVT_WU_RV64)
+  NODE_NAME_CASE(FROUND)
   NODE_NAME_CASE(READ_CYCLE_WIDE)
   NODE_NAME_CASE(BREV8)
   NODE_NAME_CASE(ORC_B)
   NODE_NAME_CASE(ZIP)
   NODE_NAME_CASE(UNZIP)
   NODE_NAME_CASE(VMV_V_X_VL)
   NODE_NAME_CASE(VFMV_V_F_VL)
   NODE_NAME_CASE(VMV_X_S)

llvm/lib/Target/RISCV/RISCVInstrInfoD.td

 
 def : PatSetCC<FPR64, any_fsetccs, SETLT, FLT_D>;
 def : PatSetCC<FPR64, any_fsetccs, SETOLT, FLT_D>;
 def : PatSetCC<FPR64, any_fsetccs, SETLE, FLE_D>;
 def : PatSetCC<FPR64, any_fsetccs, SETOLE, FLE_D>;
 
 defm Select_FPR64 : SelectCC_GPR_rrirr<FPR64>;
 
+def PseudoFROUND_D : PseudoFROUND<FPR64>;
+
 /// Loads
 
 defm : LdPat<load, FLD, f64>;
 
 /// Stores
 
 defm : StPat<store, FSD, FPR64, f64>;
 

llvm/lib/Target/RISCV/RISCVInstrInfoF.td

def SDT_RISCVFMV_X_ANYEXTW_RV64
     : SDTypeProfile<1, 1, [SDTCisVT<0, i64>, SDTCisVT<1, f32>]>;
 def SDT_RISCVFCVT_W_RV64
     : SDTypeProfile<1, 2, [SDTCisVT<0, i64>, SDTCisFP<1>,
                            SDTCisVT<2, i64>]>;
 def SDT_RISCVFCVT_X
     : SDTypeProfile<1, 2, [SDTCisVT<0, XLenVT>, SDTCisFP<1>,
                            SDTCisVT<2, XLenVT>]>;
 
+def SDT_RISCVFROUND
+    : SDTypeProfile<1, 3, [SDTCisFP<0>, SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>,
+                           SDTCisVT<3, XLenVT>]>;
+
+def riscv_fround
+    : SDNode<"RISCVISD::FROUND", SDT_RISCVFROUND>;
+
 def riscv_fmv_w_x_rv64
     : SDNode<"RISCVISD::FMV_W_X_RV64", SDT_RISCVFMV_W_X_RV64>;
 def riscv_fmv_x_anyextw_rv64
     : SDNode<"RISCVISD::FMV_X_ANYEXTW_RV64", SDT_RISCVFMV_X_ANYEXTW_RV64>;
 def riscv_fcvt_w_rv64
     : SDNode<"RISCVISD::FCVT_W_RV64", SDT_RISCVFCVT_W_RV64>;
 def riscv_fcvt_wu_rv64
     : SDNode<"RISCVISD::FCVT_WU_RV64", SDT_RISCVFCVT_W_RV64>;

 }
 multiclass FPCmp_rr_m<bits<7> funct7, bits<3> funct3, string opcodestr,
                       list<ExtInfo_r> Exts, bit Commutable = 0> {
   foreach Ext = Exts in
     let Predicates = Ext.Predicates, DecoderNamespace = Ext.Space in
     def Ext.Suffix : FPCmp_rr<funct7, funct3, opcodestr, Ext.Reg, Commutable>;
 }
 
+class PseudoFROUND<RegisterClass Ty>
+    : Pseudo<(outs Ty:$rd), (ins Ty:$rs1, Ty:$rs2, ixlenimm:$rm),
+      [(set Ty:$rd, (riscv_fround Ty:$rs1, Ty:$rs2, timm:$rm))]> {
+  let hasSideEffects = 0;
+  let mayLoad = 0;
+  let mayStore = 0;
+  let usesCustomInserter = 1;
+  let mayRaiseFPException = 1;
+}
+
 //===----------------------------------------------------------------------===//
 // Instructions
 //===----------------------------------------------------------------------===//
 
 let Predicates = [HasStdExtF] in {
 def FLW : FPLoad_r<0b010, "flw", FPR32, WriteFLD32>;
 
 // Operands for stores are in the order srcreg, base, offset rather than

 
 def : PatSetCC<FPR32, any_fsetccs, SETLT, FLT_S>;
 def : PatSetCC<FPR32, any_fsetccs, SETOLT, FLT_S>;
 def : PatSetCC<FPR32, any_fsetccs, SETLE, FLE_S>;
 def : PatSetCC<FPR32, any_fsetccs, SETOLE, FLE_S>;
 
 defm Select_FPR32 : SelectCC_GPR_rrirr<FPR32>;
 
+def PseudoFROUND_S : PseudoFROUND<FPR32>;
+
 /// Loads
 
 defm : LdPat<load, FLW, f32>;
 
 /// Stores
 
 defm : StPat<store, FSW, FPR32, f32>;
 

llvm/lib/Target/RISCV/RISCVInstrInfoZfh.td

def : Pat<(strict_fsetccs FPR16:$rs1, FPR16:$rs1, SETOEQ),
           (FLE_H $rs1, $rs1)>;
 
 def : PatSetCC<FPR16, any_fsetccs, SETLT, FLT_H>;
 def : PatSetCC<FPR16, any_fsetccs, SETOLT, FLT_H>;
 def : PatSetCC<FPR16, any_fsetccs, SETLE, FLE_H>;
 def : PatSetCC<FPR16, any_fsetccs, SETOLE, FLE_H>;
 
 defm Select_FPR16 : SelectCC_GPR_rrirr<FPR16>;
+
+def PseudoFROUND_H : PseudoFROUND<FPR16>;
 } // Predicates = [HasStdExtZfh]
 
 let Predicates = [HasStdExtZfhOrZfhmin] in {
 /// Loads
 
 defm : LdPat<load, FLH, f16>;
 
 /// Stores

llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp

static const CostTblEntry VectorIntrinsicCostTable[]{
     {Intrinsic::ctpop, MVT::nxv8i64, 21},
 };
 
 InstructionCost
 RISCVTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
                                     TTI::TargetCostKind CostKind) {
   auto *RetTy = ICA.getReturnType();
   switch (ICA.getID()) {
+  case Intrinsic::ceil:
+  case Intrinsic::floor:
+  case Intrinsic::trunc:
+  case Intrinsic::rint:
+  case Intrinsic::round:
+  case Intrinsic::roundeven: {
+    // These all use the same code.
+    auto LT = getTypeLegalizationCost(RetTy);
+    if (!LT.second.isVector() && TLI->isOperationCustom(ISD::FCEIL, LT.second))
+      return LT.first * 8;
+    break;
+  }
   case Intrinsic::umin:
   case Intrinsic::umax:
   case Intrinsic::smin:
   case Intrinsic::smax: {
     auto LT = getTypeLegalizationCost(RetTy);
     if ((ST->hasVInstructions() && LT.second.isVector()) ||
         (LT.second.isScalarInteger() && ST->hasStdExtZbb()))
       return LT.first;
RISCVTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
   // TODO: add more intrinsic
   case Intrinsic::experimental_stepvector: {
     unsigned Cost = 1; // vid
     auto LT = getTypeLegalizationCost(RetTy);
     return Cost + (LT.first - 1);
   }
   }
 
   if (ST->hasVInstructions() && RetTy->isVectorTy()) {
     auto LT = getTypeLegalizationCost(RetTy);
     if (const auto *Entry = CostTableLookup(VectorIntrinsicCostTable,
                                             ICA.getID(), LT.second))
       return LT.first * Entry->Cost;
   }
 
   return BaseT::getIntrinsicInstrCost(ICA, CostKind);
 }

llvm/test/Analysis/CostModel/RISCV/fround.ll

 ; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
 ; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -S -mtriple=riscv64 -mattr=+v,+f,+d -riscv-v-vector-bits-min=-1 | FileCheck %s
 
 define void @floor() {
 ; CHECK-LABEL: 'floor'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %1 = call float @llvm.floor.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %1 = call float @llvm.floor.f32(float undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %2 = call <2 x float> @llvm.floor.v2f32(<2 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %3 = call <4 x float> @llvm.floor.v4f32(<4 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %4 = call <8 x float> @llvm.floor.v8f32(<8 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %5 = call <16 x float> @llvm.floor.v16f32(<16 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %6 = call <vscale x 1 x float> @llvm.floor.nxv1f32(<vscale x 1 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %7 = call <vscale x 2 x float> @llvm.floor.nxv2f32(<vscale x 2 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %8 = call <vscale x 4 x float> @llvm.floor.nxv4f32(<vscale x 4 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %9 = call <vscale x 8 x float> @llvm.floor.nxv8f32(<vscale x 8 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %10 = call <vscale x 16 x float> @llvm.floor.nxv16f32(<vscale x 16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %11 = call double @llvm.floor.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %11 = call double @llvm.floor.f64(double undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %12 = call <2 x double> @llvm.floor.v2f64(<2 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %13 = call <4 x double> @llvm.floor.v4f64(<4 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %14 = call <8 x double> @llvm.floor.v8f64(<8 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %15 = call <16 x double> @llvm.floor.v16f64(<16 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %16 = call <vscale x 1 x double> @llvm.floor.nxv1f64(<vscale x 1 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %17 = call <vscale x 2 x double> @llvm.floor.nxv2f64(<vscale x 2 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %18 = call <vscale x 4 x double> @llvm.floor.nxv4f64(<vscale x 4 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %19 = call <vscale x 8 x double> @llvm.floor.nxv8f64(<vscale x 8 x double> undef)
;
   call <vscale x 2 x double> @llvm.floor.nvx2f64(<vscale x 2 x double> undef)
   call <vscale x 4 x double> @llvm.floor.nvx5f64(<vscale x 4 x double> undef)
   call <vscale x 8 x double> @llvm.floor.nvx8f64(<vscale x 8 x double> undef)
   ret void
 }
 
 define void @ceil() {
 ; CHECK-LABEL: 'ceil'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %1 = call float @llvm.ceil.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %1 = call float @llvm.ceil.f32(float undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %2 = call <2 x float> @llvm.ceil.v2f32(<2 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %3 = call <4 x float> @llvm.ceil.v4f32(<4 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %4 = call <8 x float> @llvm.ceil.v8f32(<8 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %5 = call <16 x float> @llvm.ceil.v16f32(<16 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %6 = call <vscale x 1 x float> @llvm.ceil.nxv1f32(<vscale x 1 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %7 = call <vscale x 2 x float> @llvm.ceil.nxv2f32(<vscale x 2 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %8 = call <vscale x 4 x float> @llvm.ceil.nxv4f32(<vscale x 4 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %9 = call <vscale x 8 x float> @llvm.ceil.nxv8f32(<vscale x 8 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %10 = call <vscale x 16 x float> @llvm.ceil.nxv16f32(<vscale x 16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %11 = call double @llvm.ceil.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %11 = call double @llvm.ceil.f64(double undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %12 = call <2 x double> @llvm.ceil.v2f64(<2 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %13 = call <4 x double> @llvm.ceil.v4f64(<4 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %14 = call <8 x double> @llvm.ceil.v8f64(<8 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %15 = call <16 x double> @llvm.ceil.v16f64(<16 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %16 = call <vscale x 1 x double> @llvm.ceil.nxv1f64(<vscale x 1 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %17 = call <vscale x 2 x double> @llvm.ceil.nxv2f64(<vscale x 2 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %18 = call <vscale x 4 x double> @llvm.ceil.nxv4f64(<vscale x 4 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %19 = call <vscale x 8 x double> @llvm.ceil.nxv8f64(<vscale x 8 x double> undef)
;
   call <vscale x 2 x double> @llvm.ceil.nvx2f64(<vscale x 2 x double> undef)
   call <vscale x 4 x double> @llvm.ceil.nvx5f64(<vscale x 4 x double> undef)
   call <vscale x 8 x double> @llvm.ceil.nvx8f64(<vscale x 8 x double> undef)
   ret void
 }
 
 define void @trunc() {
 ; CHECK-LABEL: 'trunc'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %1 = call float @llvm.trunc.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %1 = call float @llvm.trunc.f32(float undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %2 = call <2 x float> @llvm.trunc.v2f32(<2 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %3 = call <4 x float> @llvm.trunc.v4f32(<4 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %4 = call <8 x float> @llvm.trunc.v8f32(<8 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %5 = call <16 x float> @llvm.trunc.v16f32(<16 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %6 = call <vscale x 1 x float> @llvm.trunc.nxv1f32(<vscale x 1 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %7 = call <vscale x 2 x float> @llvm.trunc.nxv2f32(<vscale x 2 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %8 = call <vscale x 4 x float> @llvm.trunc.nxv4f32(<vscale x 4 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %9 = call <vscale x 8 x float> @llvm.trunc.nxv8f32(<vscale x 8 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %10 = call <vscale x 16 x float> @llvm.trunc.nxv16f32(<vscale x 16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %11 = call double @llvm.trunc.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %11 = call double @llvm.trunc.f64(double undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %12 = call <2 x double> @llvm.trunc.v2f64(<2 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %13 = call <4 x double> @llvm.trunc.v4f64(<4 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %14 = call <8 x double> @llvm.trunc.v8f64(<8 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %15 = call <16 x double> @llvm.trunc.v16f64(<16 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %16 = call <vscale x 1 x double> @llvm.trunc.nxv1f64(<vscale x 1 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %17 = call <vscale x 2 x double> @llvm.trunc.nxv2f64(<vscale x 2 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %18 = call <vscale x 4 x double> @llvm.trunc.nxv4f64(<vscale x 4 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %19 = call <vscale x 8 x double> @llvm.trunc.nxv8f64(<vscale x 8 x double> undef)
;
   call <vscale x 2 x double> @llvm.trunc.nvx2f64(<vscale x 2 x double> undef)
   call <vscale x 4 x double> @llvm.trunc.nvx5f64(<vscale x 4 x double> undef)
   call <vscale x 8 x double> @llvm.trunc.nvx8f64(<vscale x 8 x double> undef)
   ret void
 }
 
 define void @rint() {
 ; CHECK-LABEL: 'rint'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %1 = call float @llvm.rint.f32(float undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %2 = call <2 x float> @llvm.rint.v2f32(<2 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %3 = call <4 x float> @llvm.rint.v4f32(<4 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %4 = call <8 x float> @llvm.rint.v8f32(<8 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 191 for instruction: %5 = call <16 x float> @llvm.rint.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %1 = call float @llvm.rint.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %2 = call <2 x float> @llvm.rint.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %3 = call <4 x float> @llvm.rint.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 79 for instruction: %4 = call <8 x float> @llvm.rint.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 159 for instruction: %5 = call <16 x float> @llvm.rint.v16f32(<16 x float> undef)
 ; CHECK-NEXT:  Cost Model: Invalid cost for instruction: %6 = call <vscale x 1 x float> @llvm.rint.nxv1f32(<vscale x 1 x float> undef)
 ; CHECK-NEXT:  Cost Model: Invalid cost for instruction: %7 = call <vscale x 2 x float> @llvm.rint.nxv2f32(<vscale x 2 x float> undef)
 ; CHECK-NEXT:  Cost Model: Invalid cost for instruction: %8 = call <vscale x 4 x float> @llvm.rint.nxv4f32(<vscale x 4 x float> undef)
 ; CHECK-NEXT:  Cost Model: Invalid cost for instruction: %9 = call <vscale x 8 x float> @llvm.rint.nxv8f32(<vscale x 8 x float> undef)
 ; CHECK-NEXT:  Cost Model: Invalid cost for instruction: %10 = call <vscale x 16 x float> @llvm.rint.nxv16f32(<vscale x 16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %11 = call double @llvm.rint.f64(double undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %12 = call <2 x double> @llvm.rint.v2f64(<2 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %13 = call <4 x double> @llvm.rint.v4f64(<4 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %14 = call <8 x double> @llvm.rint.v8f64(<8 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 191 for instruction: %15 = call <16 x double> @llvm.rint.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %11 = call double @llvm.rint.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %12 = call <2 x double> @llvm.rint.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %13 = call <4 x double> @llvm.rint.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 79 for instruction: %14 = call <8 x double> @llvm.rint.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 159 for instruction: %15 = call <16 x double> @llvm.rint.v16f64(<16 x double> undef)
 ; CHECK-NEXT:  Cost Model: Invalid cost for instruction: %16 = call <vscale x 1 x double> @llvm.rint.nxv1f64(<vscale x 1 x double> undef)
 ; CHECK-NEXT:  Cost Model: Invalid cost for instruction: %17 = call <vscale x 2 x double> @llvm.rint.nxv2f64(<vscale x 2 x double> undef)
 ; CHECK-NEXT:  Cost Model: Invalid cost for instruction: %18 = call <vscale x 4 x double> @llvm.rint.nxv4f64(<vscale x 4 x double> undef)
 ; CHECK-NEXT:  Cost Model: Invalid cost for instruction: %19 = call <vscale x 8 x double> @llvm.rint.nxv8f64(<vscale x 8 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   call float @llvm.rint.f32(float undef)
   call <2 x float> @llvm.rint.v2f32(<2 x float> undef)
;
   call <vscale x 2 x double> @llvm.nearbyint.nvx2f64(<vscale x 2 x double> undef)
   call <vscale x 4 x double> @llvm.nearbyint.nvx5f64(<vscale x 4 x double> undef)
   call <vscale x 8 x double> @llvm.nearbyint.nvx8f64(<vscale x 8 x double> undef)
   ret void
 }
 
 define void @round() {
 ; CHECK-LABEL: 'round'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %1 = call float @llvm.round.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %1 = call float @llvm.round.f32(float undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %2 = call <2 x float> @llvm.round.v2f32(<2 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %3 = call <4 x float> @llvm.round.v4f32(<4 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %4 = call <8 x float> @llvm.round.v8f32(<8 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %5 = call <16 x float> @llvm.round.v16f32(<16 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %6 = call <vscale x 1 x float> @llvm.round.nxv1f32(<vscale x 1 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %7 = call <vscale x 2 x float> @llvm.round.nxv2f32(<vscale x 2 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %8 = call <vscale x 4 x float> @llvm.round.nxv4f32(<vscale x 4 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %9 = call <vscale x 8 x float> @llvm.round.nxv8f32(<vscale x 8 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %10 = call <vscale x 16 x float> @llvm.round.nxv16f32(<vscale x 16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %11 = call double @llvm.round.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %11 = call double @llvm.round.f64(double undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %12 = call <2 x double> @llvm.round.v2f64(<2 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %13 = call <4 x double> @llvm.round.v4f64(<4 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %14 = call <8 x double> @llvm.round.v8f64(<8 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %15 = call <16 x double> @llvm.round.v16f64(<16 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %16 = call <vscale x 1 x double> @llvm.round.nxv1f64(<vscale x 1 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %17 = call <vscale x 2 x double> @llvm.round.nxv2f64(<vscale x 2 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %18 = call <vscale x 4 x double> @llvm.round.nxv4f64(<vscale x 4 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %19 = call <vscale x 8 x double> @llvm.round.nxv8f64(<vscale x 8 x double> undef)
;
   call <vscale x 2 x double> @llvm.round.nvx2f64(<vscale x 2 x double> undef)
   call <vscale x 4 x double> @llvm.round.nvx5f64(<vscale x 4 x double> undef)
   call <vscale x 8 x double> @llvm.round.nvx8f64(<vscale x 8 x double> undef)
   ret void
 }
 
 define void @roundeven() {
 ; CHECK-LABEL: 'roundeven'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %1 = call float @llvm.roundeven.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %1 = call float @llvm.roundeven.f32(float undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %2 = call <2 x float> @llvm.roundeven.v2f32(<2 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %3 = call <4 x float> @llvm.roundeven.v4f32(<4 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %4 = call <8 x float> @llvm.roundeven.v8f32(<8 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %5 = call <16 x float> @llvm.roundeven.v16f32(<16 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %6 = call <vscale x 1 x float> @llvm.roundeven.nxv1f32(<vscale x 1 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %7 = call <vscale x 2 x float> @llvm.roundeven.nxv2f32(<vscale x 2 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %8 = call <vscale x 4 x float> @llvm.roundeven.nxv4f32(<vscale x 4 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %9 = call <vscale x 8 x float> @llvm.roundeven.nxv8f32(<vscale x 8 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %10 = call <vscale x 16 x float> @llvm.roundeven.nxv16f32(<vscale x 16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %11 = call double @llvm.roundeven.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %11 = call double @llvm.roundeven.f64(double undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %12 = call <2 x double> @llvm.roundeven.v2f64(<2 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %13 = call <4 x double> @llvm.roundeven.v4f64(<4 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %14 = call <8 x double> @llvm.roundeven.v8f64(<8 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %15 = call <16 x double> @llvm.roundeven.v16f64(<16 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %16 = call <vscale x 1 x double> @llvm.roundeven.nxv1f64(<vscale x 1 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %17 = call <vscale x 2 x double> @llvm.roundeven.nxv2f64(<vscale x 2 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %18 = call <vscale x 4 x double> @llvm.roundeven.nxv4f64(<vscale x 4 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %19 = call <vscale x 8 x double> @llvm.roundeven.nxv8f64(<vscale x 8 x double> undef)

llvm/test/CodeGen/RISCV/double-intrinsics.ll

 ; RV64I-NEXT:    ret
   %1 = call double @llvm.copysign.f64(double %a, double %b)
   ret double %1
 }
 
 declare double @llvm.floor.f64(double)
 
 define double @floor_f64(double %a) nounwind {
-; CHECKIFD-LABEL: floor_f64:
-; CHECKIFD:       # %bb.0:
-; CHECKIFD-NEXT:    tail floor@plt
+; RV32IFD-LABEL: floor_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    tail floor@plt
+;
+; RV64IFD-LABEL: floor_f64:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    lui a0, %hi(.LCPI17_0)
+; RV64IFD-NEXT:    fld ft0, %lo(.LCPI17_0)(a0)
+; RV64IFD-NEXT:    fabs.d ft1, fa0
+; RV64IFD-NEXT:    flt.d a0, ft1, ft0
+; RV64IFD-NEXT:    beqz a0, .LBB17_2
+; RV64IFD-NEXT:  # %bb.1:
+; RV64IFD-NEXT:    fcvt.l.d a0, fa0, rdn
+; RV64IFD-NEXT:    fcvt.d.l ft0, a0, rdn
+; RV64IFD-NEXT:    fsgnj.d fa0, ft0, fa0
+; RV64IFD-NEXT:  .LBB17_2:
+; RV64IFD-NEXT:    ret
 ;
 ; RV32I-LABEL: floor_f64:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    call floor@plt
 ; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-NEXT:    addi sp, sp, 16

 ; RV64I-NEXT:    ret
   %1 = call double @llvm.floor.f64(double %a)
   ret double %1
 }
 
 declare double @llvm.ceil.f64(double)
 
 define double @ceil_f64(double %a) nounwind {
-; CHECKIFD-LABEL: ceil_f64:
-; CHECKIFD:       # %bb.0:
-; CHECKIFD-NEXT:    tail ceil@plt
+; RV32IFD-LABEL: ceil_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    tail ceil@plt
+;
+; RV64IFD-LABEL: ceil_f64:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    lui a0, %hi(.LCPI18_0)
+; RV64IFD-NEXT:    fld ft0, %lo(.LCPI18_0)(a0)
+; RV64IFD-NEXT:    fabs.d ft1, fa0
+; RV64IFD-NEXT:    flt.d a0, ft1, ft0
+; RV64IFD-NEXT:    beqz a0, .LBB18_2
+; RV64IFD-NEXT:  # %bb.1:
+; RV64IFD-NEXT:    fcvt.l.d a0, fa0, rup
+; RV64IFD-NEXT:    fcvt.d.l ft0, a0, rup
+; RV64IFD-NEXT:    fsgnj.d fa0, ft0, fa0
+; RV64IFD-NEXT:  .LBB18_2:
+; RV64IFD-NEXT:    ret
 ;
 ; RV32I-LABEL: ceil_f64:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    call ceil@plt
 ; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-NEXT:    addi sp, sp, 16

 ; RV64I-NEXT:    ret
   %1 = call double @llvm.ceil.f64(double %a)
   ret double %1
 }
 
 declare double @llvm.trunc.f64(double)
 
 define double @trunc_f64(double %a) nounwind {
-; CHECKIFD-LABEL: trunc_f64:
-; CHECKIFD:       # %bb.0:
-; CHECKIFD-NEXT:    tail trunc@plt
+; RV32IFD-LABEL: trunc_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    tail trunc@plt
+;
+; RV64IFD-LABEL: trunc_f64:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    lui a0, %hi(.LCPI19_0)
+; RV64IFD-NEXT:    fld ft0, %lo(.LCPI19_0)(a0)
+; RV64IFD-NEXT:    fabs.d ft1, fa0
+; RV64IFD-NEXT:    flt.d a0, ft1, ft0
+; RV64IFD-NEXT:    beqz a0, .LBB19_2
+; RV64IFD-NEXT:  # %bb.1:
+; RV64IFD-NEXT:    fcvt.l.d a0, fa0, rtz
+; RV64IFD-NEXT:    fcvt.d.l ft0, a0, rtz
+; RV64IFD-NEXT:    fsgnj.d fa0, ft0, fa0
+; RV64IFD-NEXT:  .LBB19_2:
+; RV64IFD-NEXT:    ret
 ;
 ; RV32I-LABEL: trunc_f64:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    call trunc@plt
 ; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-NEXT:    addi sp, sp, 16

 ; RV64I-NEXT:    ret
   %1 = call double @llvm.trunc.f64(double %a)
   ret double %1
 }
 
 declare double @llvm.rint.f64(double)
 
 define double @rint_f64(double %a) nounwind {
-; CHECKIFD-LABEL: rint_f64:
-; CHECKIFD:       # %bb.0:
-; CHECKIFD-NEXT:    tail rint@plt
+; RV32IFD-LABEL: rint_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    tail rint@plt
+;
+; RV64IFD-LABEL: rint_f64:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    lui a0, %hi(.LCPI20_0)
+; RV64IFD-NEXT:    fld ft0, %lo(.LCPI20_0)(a0)
+; RV64IFD-NEXT:    fabs.d ft1, fa0
+; RV64IFD-NEXT:    flt.d a0, ft1, ft0
+; RV64IFD-NEXT:    beqz a0, .LBB20_2
+; RV64IFD-NEXT:  # %bb.1:
+; RV64IFD-NEXT:    fcvt.l.d a0, fa0
+; RV64IFD-NEXT:    fcvt.d.l ft0, a0
+; RV64IFD-NEXT:    fsgnj.d fa0, ft0, fa0
+; RV64IFD-NEXT:  .LBB20_2:
+; RV64IFD-NEXT:    ret
 ;
 ; RV32I-LABEL: rint_f64:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    call rint@plt
 ; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-NEXT:    addi sp, sp, 16

 ; RV64I-NEXT:    ret
   %1 = call double @llvm.nearbyint.f64(double %a)
   ret double %1
 }
 
 declare double @llvm.round.f64(double)
 
 define double @round_f64(double %a) nounwind {
-; CHECKIFD-LABEL: round_f64:
-; CHECKIFD:       # %bb.0:
-; CHECKIFD-NEXT:    tail round@plt
+; RV32IFD-LABEL: round_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    tail round@plt
+;
+; RV64IFD-LABEL: round_f64:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    lui a0, %hi(.LCPI22_0)
+; RV64IFD-NEXT:    fld ft0, %lo(.LCPI22_0)(a0)
+; RV64IFD-NEXT:    fabs.d ft1, fa0
+; RV64IFD-NEXT:    flt.d a0, ft1, ft0
+; RV64IFD-NEXT:    beqz a0, .LBB22_2
+; RV64IFD-NEXT:  # %bb.1:
+; RV64IFD-NEXT:    fcvt.l.d a0, fa0, rmm
+; RV64IFD-NEXT:    fcvt.d.l ft0, a0, rmm
+; RV64IFD-NEXT:    fsgnj.d fa0, ft0, fa0
+; RV64IFD-NEXT:  .LBB22_2:
+; RV64IFD-NEXT:    ret
 ;
 ; RV32I-LABEL: round_f64:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    call round@plt
 ; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-NEXT:    addi sp, sp, 16

 ; RV64I-NEXT:    ret
   %1 = call double @llvm.round.f64(double %a)
   ret double %1
 }
 
 declare double @llvm.roundeven.f64(double)
 
 define double @roundeven_f64(double %a) nounwind {
-; CHECKIFD-LABEL: roundeven_f64:
-; CHECKIFD:       # %bb.0:
-; CHECKIFD-NEXT:    tail roundeven@plt
+; RV32IFD-LABEL: roundeven_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    tail roundeven@plt
+;
+; RV64IFD-LABEL: roundeven_f64:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    lui a0, %hi(.LCPI23_0)
+; RV64IFD-NEXT:    fld ft0, %lo(.LCPI23_0)(a0)
+; RV64IFD-NEXT:    fabs.d ft1, fa0
+; RV64IFD-NEXT:    flt.d a0, ft1, ft0
+; RV64IFD-NEXT:    beqz a0, .LBB23_2
+; RV64IFD-NEXT:  # %bb.1:
+; RV64IFD-NEXT:    fcvt.l.d a0, fa0, rne
+; RV64IFD-NEXT:    fcvt.d.l ft0, a0, rne
+; RV64IFD-NEXT:    fsgnj.d fa0, ft0, fa0
+; RV64IFD-NEXT:  .LBB23_2:
+; RV64IFD-NEXT:    ret
 ;
 ; RV32I-LABEL: roundeven_f64:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    call roundeven@plt
 ; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-NEXT:    addi sp, sp, 16

llvm/test/CodeGen/RISCV/double-round-conv.ll

 ; RV64IFD-NEXT:    fcvt.lu.d a0, fa0, rne
 ; RV64IFD-NEXT:    ret
   %a = call double @llvm.roundeven.f64(double %x)
   %b = fptoui double %a to i64
   ret i64 %b
 }
 
 define double @test_floor_double(double %x) {
-; CHECKIFD-LABEL: test_floor_double:
-; CHECKIFD:       # %bb.0:
-; CHECKIFD-NEXT:    tail floor@plt
+; RV32IFD-LABEL: test_floor_double:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    tail floor@plt
+;
+; RV64IFD-LABEL: test_floor_double:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    lui a0, %hi(.LCPI40_0)
+; RV64IFD-NEXT:    fld ft0, %lo(.LCPI40_0)(a0)
+; RV64IFD-NEXT:    fabs.d ft1, fa0
+; RV64IFD-NEXT:    flt.d a0, ft1, ft0
+; RV64IFD-NEXT:    beqz a0, .LBB40_2
+; RV64IFD-NEXT:  # %bb.1:
+; RV64IFD-NEXT:    fcvt.l.d a0, fa0, rdn
+; RV64IFD-NEXT:    fcvt.d.l ft0, a0, rdn
+; RV64IFD-NEXT:    fsgnj.d fa0, ft0, fa0
+; RV64IFD-NEXT:  .LBB40_2:
+; RV64IFD-NEXT:    ret
   %a = call double @llvm.floor.f64(double %x)
   ret double %a
 }
 
 define double @test_ceil_double(double %x) {
-; CHECKIFD-LABEL: test_ceil_double:
-; CHECKIFD:       # %bb.0:
-; CHECKIFD-NEXT:    tail ceil@plt
+; RV32IFD-LABEL: test_ceil_double:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    tail ceil@plt
+;
+; RV64IFD-LABEL: test_ceil_double:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    lui a0, %hi(.LCPI41_0)
+; RV64IFD-NEXT:    fld ft0, %lo(.LCPI41_0)(a0)
+; RV64IFD-NEXT:    fabs.d ft1, fa0
+; RV64IFD-NEXT:    flt.d a0, ft1, ft0
+; RV64IFD-NEXT:    beqz a0, .LBB41_2
+; RV64IFD-NEXT:  # %bb.1:
+; RV64IFD-NEXT:    fcvt.l.d a0, fa0, rup
+; RV64IFD-NEXT:    fcvt.d.l ft0, a0, rup
+; RV64IFD-NEXT:    fsgnj.d fa0, ft0, fa0
+; RV64IFD-NEXT:  .LBB41_2:
+; RV64IFD-NEXT:    ret
   %a = call double @llvm.ceil.f64(double %x)
   ret double %a
 }
 
 define double @test_trunc_double(double %x) {
-; CHECKIFD-LABEL: test_trunc_double:
-; CHECKIFD:       # %bb.0:
-; CHECKIFD-NEXT:    tail trunc@plt
+; RV32IFD-LABEL: test_trunc_double:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    tail trunc@plt
+;
+; RV64IFD-LABEL: test_trunc_double:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    lui a0, %hi(.LCPI42_0)
+; RV64IFD-NEXT:    fld ft0, %lo(.LCPI42_0)(a0)
+; RV64IFD-NEXT:    fabs.d ft1, fa0
+; RV64IFD-NEXT:    flt.d a0, ft1, ft0
+; RV64IFD-NEXT:    beqz a0, .LBB42_2
+; RV64IFD-NEXT:  # %bb.1:
+; RV64IFD-NEXT:    fcvt.l.d a0, fa0, rtz
+; RV64IFD-NEXT:    fcvt.d.l ft0, a0, rtz
+; RV64IFD-NEXT:    fsgnj.d fa0, ft0, fa0
+; RV64IFD-NEXT:  .LBB42_2:
+; RV64IFD-NEXT:    ret
   %a = call double @llvm.trunc.f64(double %x)
   ret double %a
 }
 
 define double @test_round_double(double %x) {
-; CHECKIFD-LABEL: test_round_double:
-; CHECKIFD:       # %bb.0:
-; CHECKIFD-NEXT:    tail round@plt
+; RV32IFD-LABEL: test_round_double:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    tail round@plt
+;
+; RV64IFD-LABEL: test_round_double:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    lui a0, %hi(.LCPI43_0)
+; RV64IFD-NEXT:    fld ft0, %lo(.LCPI43_0)(a0)
+; RV64IFD-NEXT:    fabs.d ft1, fa0
+; RV64IFD-NEXT:    flt.d a0, ft1, ft0
+; RV64IFD-NEXT:    beqz a0, .LBB43_2
+; RV64IFD-NEXT:  # %bb.1:
+; RV64IFD-NEXT:    fcvt.l.d a0, fa0, rmm
+; RV64IFD-NEXT:    fcvt.d.l ft0, a0, rmm
+; RV64IFD-NEXT:    fsgnj.d fa0, ft0, fa0
+; RV64IFD-NEXT:  .LBB43_2:
+; RV64IFD-NEXT:    ret
   %a = call double @llvm.round.f64(double %x)
   ret double %a
 }
 
 define double @test_roundeven_double(double %x) {
-; CHECKIFD-LABEL: test_roundeven_double:
-; CHECKIFD:       # %bb.0:
-; CHECKIFD-NEXT:    tail roundeven@plt
+; RV32IFD-LABEL: test_roundeven_double:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    tail roundeven@plt
+;
+; RV64IFD-LABEL: test_roundeven_double:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    lui a0, %hi(.LCPI44_0)
+; RV64IFD-NEXT:    fld ft0, %lo(.LCPI44_0)(a0)
+; RV64IFD-NEXT:    fabs.d ft1, fa0
+; RV64IFD-NEXT:    flt.d a0, ft1, ft0
+; RV64IFD-NEXT:    beqz a0, .LBB44_2
+; RV64IFD-NEXT:  # %bb.1:
+; RV64IFD-NEXT:    fcvt.l.d a0, fa0, rne
+; RV64IFD-NEXT:    fcvt.d.l ft0, a0, rne
+; RV64IFD-NEXT:    fsgnj.d fa0, ft0, fa0
+; RV64IFD-NEXT:  .LBB44_2:
+; RV64IFD-NEXT:    ret
   %a = call double @llvm.roundeven.f64(double %x)
   ret double %a
 }
 
 declare double @llvm.floor.f64(double)
 declare double @llvm.ceil.f64(double)
 declare double @llvm.trunc.f64(double)
 declare double @llvm.round.f64(double)
 declare double @llvm.roundeven.f64(double)

llvm/test/CodeGen/RISCV/float-intrinsics.ll

; RV64I-NEXT:    ret
   ret float %1
 }
 
 declare float @llvm.floor.f32(float)
 
 define float @floor_f32(float %a) nounwind {
 ; RV32IF-LABEL: floor_f32:
 ; RV32IF:       # %bb.0:
-; RV32IF-NEXT:    tail floorf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI17_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI17_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB17_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rdn
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rdn
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB17_2:
+; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: floor_f32:
 ; RV64IF:       # %bb.0:
-; RV64IF-NEXT:    tail floorf@plt
+; RV64IF-NEXT:    lui a0, %hi(.LCPI17_0)
+; RV64IF-NEXT:    flw ft0, %lo(.LCPI17_0)(a0)
+; RV64IF-NEXT:    fabs.s ft1, fa0
+; RV64IF-NEXT:    flt.s a0, ft1, ft0
+; RV64IF-NEXT:    beqz a0, .LBB17_2
+; RV64IF-NEXT:  # %bb.1:
+; RV64IF-NEXT:    fcvt.w.s a0, fa0, rdn
+; RV64IF-NEXT:    fcvt.s.w ft0, a0, rdn
+; RV64IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV64IF-NEXT:  .LBB17_2:
+; RV64IF-NEXT:    ret
 ;
 ; RV32I-LABEL: floor_f32:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    call floorf@plt
 ; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-NEXT:    addi sp, sp, 16
; RV64I-NEXT:    ret
   ret float %1
 }
 
 declare float @llvm.ceil.f32(float)
 
 define float @ceil_f32(float %a) nounwind {
 ; RV32IF-LABEL: ceil_f32:
 ; RV32IF:       # %bb.0:
-; RV32IF-NEXT:    tail ceilf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI18_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI18_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB18_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rup
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rup
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB18_2:
+; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: ceil_f32:
 ; RV64IF:       # %bb.0:
-; RV64IF-NEXT:    tail ceilf@plt
+; RV64IF-NEXT:    lui a0, %hi(.LCPI18_0)
+; RV64IF-NEXT:    flw ft0, %lo(.LCPI18_0)(a0)
+; RV64IF-NEXT:    fabs.s ft1, fa0
+; RV64IF-NEXT:    flt.s a0, ft1, ft0
+; RV64IF-NEXT:    beqz a0, .LBB18_2
+; RV64IF-NEXT:  # %bb.1:
+; RV64IF-NEXT:    fcvt.w.s a0, fa0, rup
+; RV64IF-NEXT:    fcvt.s.w ft0, a0, rup
+; RV64IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV64IF-NEXT:  .LBB18_2:
+; RV64IF-NEXT:    ret
 ;
 ; RV32I-LABEL: ceil_f32:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    call ceilf@plt
 ; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-NEXT:    addi sp, sp, 16
; RV64I-NEXT:    ret
   ret float %1
 }
 
 declare float @llvm.trunc.f32(float)
 
 define float @trunc_f32(float %a) nounwind {
 ; RV32IF-LABEL: trunc_f32:
 ; RV32IF:       # %bb.0:
-; RV32IF-NEXT:    tail truncf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI19_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI19_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB19_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rtz
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rtz
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB19_2:
+; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: trunc_f32:
 ; RV64IF:       # %bb.0:
-; RV64IF-NEXT:    tail truncf@plt
+; RV64IF-NEXT:    lui a0, %hi(.LCPI19_0)
+; RV64IF-NEXT:    flw ft0, %lo(.LCPI19_0)(a0)
+; RV64IF-NEXT:    fabs.s ft1, fa0
+; RV64IF-NEXT:    flt.s a0, ft1, ft0
+; RV64IF-NEXT:    beqz a0, .LBB19_2
+; RV64IF-NEXT:  # %bb.1:
+; RV64IF-NEXT:    fcvt.w.s a0, fa0, rtz
+; RV64IF-NEXT:    fcvt.s.w ft0, a0, rtz
+; RV64IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV64IF-NEXT:  .LBB19_2:
+; RV64IF-NEXT:    ret
 ;
 ; RV32I-LABEL: trunc_f32:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    call truncf@plt
 ; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-NEXT:    addi sp, sp, 16
; RV64I-NEXT:    ret
   ret float %1
 }
 
 declare float @llvm.rint.f32(float)
 
 define float @rint_f32(float %a) nounwind {
 ; RV32IF-LABEL: rint_f32:
 ; RV32IF:       # %bb.0:
-; RV32IF-NEXT:    tail rintf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI20_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI20_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB20_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0
+; RV32IF-NEXT:    fcvt.s.w ft0, a0
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB20_2:
+; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: rint_f32:
 ; RV64IF:       # %bb.0:
-; RV64IF-NEXT:    tail rintf@plt
+; RV64IF-NEXT:    lui a0, %hi(.LCPI20_0)
+; RV64IF-NEXT:    flw ft0, %lo(.LCPI20_0)(a0)
+; RV64IF-NEXT:    fabs.s ft1, fa0
+; RV64IF-NEXT:    flt.s a0, ft1, ft0
+; RV64IF-NEXT:    beqz a0, .LBB20_2
+; RV64IF-NEXT:  # %bb.1:
+; RV64IF-NEXT:    fcvt.w.s a0, fa0
+; RV64IF-NEXT:    fcvt.s.w ft0, a0
+; RV64IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV64IF-NEXT:  .LBB20_2:
+; RV64IF-NEXT:    ret
 ;
 ; RV32I-LABEL: rint_f32:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    call rintf@plt
 ; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-NEXT:    addi sp, sp, 16
; RV64I-NEXT:    ret
   ret float %1
 }
 
 declare float @llvm.round.f32(float)
 
 define float @round_f32(float %a) nounwind {
 ; RV32IF-LABEL: round_f32:
 ; RV32IF:       # %bb.0:
-; RV32IF-NEXT:    tail roundf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI22_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI22_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB22_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rmm
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rmm
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB22_2:
+; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: round_f32:
 ; RV64IF:       # %bb.0:
-; RV64IF-NEXT:    tail roundf@plt
+; RV64IF-NEXT:    lui a0, %hi(.LCPI22_0)
+; RV64IF-NEXT:    flw ft0, %lo(.LCPI22_0)(a0)
+; RV64IF-NEXT:    fabs.s ft1, fa0
+; RV64IF-NEXT:    flt.s a0, ft1, ft0
+; RV64IF-NEXT:    beqz a0, .LBB22_2
+; RV64IF-NEXT:  # %bb.1:
+; RV64IF-NEXT:    fcvt.w.s a0, fa0, rmm
+; RV64IF-NEXT:    fcvt.s.w ft0, a0, rmm
+; RV64IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV64IF-NEXT:  .LBB22_2:
+; RV64IF-NEXT:    ret
 ;
 ; RV32I-LABEL: round_f32:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    call roundf@plt
 ; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-NEXT:    addi sp, sp, 16
; RV64I-NEXT:    ret
   ret float %1
 }
 
 declare float @llvm.roundeven.f32(float)
 
 define float @roundeven_f32(float %a) nounwind {
 ; RV32IF-LABEL: roundeven_f32:
 ; RV32IF:       # %bb.0:
-; RV32IF-NEXT:    tail roundevenf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI23_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI23_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB23_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rne
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rne
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB23_2:
+; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: roundeven_f32:
 ; RV64IF:       # %bb.0:
-; RV64IF-NEXT:    tail roundevenf@plt
+; RV64IF-NEXT:    lui a0, %hi(.LCPI23_0)
+; RV64IF-NEXT:    flw ft0, %lo(.LCPI23_0)(a0)
+; RV64IF-NEXT:    fabs.s ft1, fa0
+; RV64IF-NEXT:    flt.s a0, ft1, ft0
+; RV64IF-NEXT:    beqz a0, .LBB23_2
+; RV64IF-NEXT:  # %bb.1:
+; RV64IF-NEXT:    fcvt.w.s a0, fa0, rne
+; RV64IF-NEXT:    fcvt.s.w ft0, a0, rne
+; RV64IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV64IF-NEXT:  .LBB23_2:
+; RV64IF-NEXT:    ret
 ;
 ; RV32I-LABEL: roundeven_f32:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    call roundevenf@plt
 ; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-NEXT:    addi sp, sp, 16

llvm/test/CodeGen/RISCV/float-round-conv-sat.ll

 
 define i64 @test_floor_si64(float %x) nounwind {
 ; RV32IF-LABEL: test_floor_si64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
-; RV32IF-NEXT:    call floorf@plt
 ; RV32IF-NEXT:    lui a0, %hi(.LCPI1_0)
 ; RV32IF-NEXT:    flw ft0, %lo(.LCPI1_0)(a0)
 ; RV32IF-NEXT:    fmv.s fs0, fa0
-; RV32IF-NEXT:    fle.s s0, ft0, fa0
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB1_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fs0, rdn
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rdn
+; RV32IF-NEXT:    fsgnj.s fs0, ft0, fs0
+; RV32IF-NEXT:  .LBB1_2:
+; RV32IF-NEXT:    lui a0, %hi(.LCPI1_1)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI1_1)(a0)
+; RV32IF-NEXT:    fle.s s0, ft0, fs0
+; RV32IF-NEXT:    fmv.s fa0, fs0
 ; RV32IF-NEXT:    call __fixsfdi@plt
 ; RV32IF-NEXT:    lui a3, 524288
-; RV32IF-NEXT:    bnez s0, .LBB1_2
-; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    bnez s0, .LBB1_4
+; RV32IF-NEXT:  # %bb.3:
 ; RV32IF-NEXT:    lui a1, 524288
-; RV32IF-NEXT:  .LBB1_2:
-; RV32IF-NEXT:    lui a2, %hi(.LCPI1_1)
-; RV32IF-NEXT:    flw ft0, %lo(.LCPI1_1)(a2)
+; RV32IF-NEXT:  .LBB1_4:
+; RV32IF-NEXT:    lui a2, %hi(.LCPI1_2)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI1_2)(a2)
 ; RV32IF-NEXT:    flt.s a2, ft0, fs0
-; RV32IF-NEXT:    beqz a2, .LBB1_4
-; RV32IF-NEXT:  # %bb.3:
+; RV32IF-NEXT:    beqz a2, .LBB1_6
+; RV32IF-NEXT:  # %bb.5:
 ; RV32IF-NEXT:    addi a1, a3, -1
-; RV32IF-NEXT:  .LBB1_4:
+; RV32IF-NEXT:  .LBB1_6:
 ; RV32IF-NEXT:    feq.s a3, fs0, fs0
 ; RV32IF-NEXT:    seqz a3, a3
 ; RV32IF-NEXT:    addi a3, a3, -1
 ; RV32IF-NEXT:    and a1, a3, a1
-; RV32IF-NEXT:    neg a2, a2
 ; RV32IF-NEXT:    neg a4, s0
 ; RV32IF-NEXT:    and a0, a4, a0
+; RV32IF-NEXT:    neg a2, a2
 ; RV32IF-NEXT:    or a0, a2, a0
 ; RV32IF-NEXT:    and a0, a3, a0
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;

 }
 
 define i64 @test_floor_ui64(float %x) nounwind {
 ; RV32IF-LABEL: test_floor_ui64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
-; RV32IF-NEXT:    sw s1, 4(sp) # 4-byte Folded Spill
-; RV32IF-NEXT:    call floorf@plt
+; RV32IF-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    lui a0, %hi(.LCPI3_0)
 ; RV32IF-NEXT:    flw ft0, %lo(.LCPI3_0)(a0)
-; RV32IF-NEXT:    flt.s a0, ft0, fa0
-; RV32IF-NEXT:    neg s0, a0
+; RV32IF-NEXT:    fmv.s fs0, fa0
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB3_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fs0, rdn
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rdn
+; RV32IF-NEXT:    fsgnj.s fs0, ft0, fs0
+; RV32IF-NEXT:  .LBB3_2:
 ; RV32IF-NEXT:    fmv.w.x ft0, zero
-; RV32IF-NEXT:    fle.s a0, ft0, fa0
-; RV32IF-NEXT:    neg s1, a0
+; RV32IF-NEXT:    fle.s a0, ft0, fs0
+; RV32IF-NEXT:    neg s0, a0
+; RV32IF-NEXT:    fmv.s fa0, fs0
 ; RV32IF-NEXT:    call __fixunssfdi@plt
-; RV32IF-NEXT:    and a0, s1, a0
-; RV32IF-NEXT:    or a0, s0, a0
-; RV32IF-NEXT:    and a1, s1, a1
-; RV32IF-NEXT:    or a1, s0, a1
+; RV32IF-NEXT:    lui a2, %hi(.LCPI3_1)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI3_1)(a2)
+; RV32IF-NEXT:    and a0, s0, a0
+; RV32IF-NEXT:    flt.s a2, ft0, fs0
+; RV32IF-NEXT:    neg a2, a2
+; RV32IF-NEXT:    or a0, a2, a0
+; RV32IF-NEXT:    and a1, s0, a1
+; RV32IF-NEXT:    or a1, a2, a1
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
-; RV32IF-NEXT:    lw s1, 4(sp) # 4-byte Folded Reload
+; RV32IF-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_floor_ui64:
 ; RV64IF:       # %bb.0:
 ; RV64IF-NEXT:    fcvt.lu.s a0, fa0, rdn
 ; RV64IF-NEXT:    feq.s a1, fa0, fa0
 ; RV64IF-NEXT:    seqz a1, a1

 
 define i64 @test_ceil_si64(float %x) nounwind {
 ; RV32IF-LABEL: test_ceil_si64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
-; RV32IF-NEXT:    call ceilf@plt
 ; RV32IF-NEXT:    lui a0, %hi(.LCPI5_0)
 ; RV32IF-NEXT:    flw ft0, %lo(.LCPI5_0)(a0)
 ; RV32IF-NEXT:    fmv.s fs0, fa0
-; RV32IF-NEXT:    fle.s s0, ft0, fa0
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB5_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fs0, rup
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rup
+; RV32IF-NEXT:    fsgnj.s fs0, ft0, fs0
+; RV32IF-NEXT:  .LBB5_2:
+; RV32IF-NEXT:    lui a0, %hi(.LCPI5_1)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI5_1)(a0)
+; RV32IF-NEXT:    fle.s s0, ft0, fs0
+; RV32IF-NEXT:    fmv.s fa0, fs0
 ; RV32IF-NEXT:    call __fixsfdi@plt
 ; RV32IF-NEXT:    lui a3, 524288
-; RV32IF-NEXT:    bnez s0, .LBB5_2
-; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    bnez s0, .LBB5_4
+; RV32IF-NEXT:  # %bb.3:
 ; RV32IF-NEXT:    lui a1, 524288
-; RV32IF-NEXT:  .LBB5_2:
-; RV32IF-NEXT:    lui a2, %hi(.LCPI5_1)
-; RV32IF-NEXT:    flw ft0, %lo(.LCPI5_1)(a2)
+; RV32IF-NEXT:  .LBB5_4:
+; RV32IF-NEXT:    lui a2, %hi(.LCPI5_2)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI5_2)(a2)
 ; RV32IF-NEXT:    flt.s a2, ft0, fs0
-; RV32IF-NEXT:    beqz a2, .LBB5_4
-; RV32IF-NEXT:  # %bb.3:
+; RV32IF-NEXT:    beqz a2, .LBB5_6
+; RV32IF-NEXT:  # %bb.5:
 ; RV32IF-NEXT:    addi a1, a3, -1
-; RV32IF-NEXT:  .LBB5_4:
+; RV32IF-NEXT:  .LBB5_6:
 ; RV32IF-NEXT:    feq.s a3, fs0, fs0
 ; RV32IF-NEXT:    seqz a3, a3
 ; RV32IF-NEXT:    addi a3, a3, -1
 ; RV32IF-NEXT:    and a1, a3, a1
-; RV32IF-NEXT:    neg a2, a2
 ; RV32IF-NEXT:    neg a4, s0
 ; RV32IF-NEXT:    and a0, a4, a0
+; RV32IF-NEXT:    neg a2, a2
 ; RV32IF-NEXT:    or a0, a2, a0
 ; RV32IF-NEXT:    and a0, a3, a0
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;

 }
 
 define i64 @test_ceil_ui64(float %x) nounwind {
 ; RV32IF-LABEL: test_ceil_ui64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
-; RV32IF-NEXT:    sw s1, 4(sp) # 4-byte Folded Spill
-; RV32IF-NEXT:    call ceilf@plt
+; RV32IF-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    lui a0, %hi(.LCPI7_0)
 ; RV32IF-NEXT:    flw ft0, %lo(.LCPI7_0)(a0)
-; RV32IF-NEXT:    flt.s a0, ft0, fa0
-; RV32IF-NEXT:    neg s0, a0
+; RV32IF-NEXT:    fmv.s fs0, fa0
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB7_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fs0, rup
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rup
+; RV32IF-NEXT:    fsgnj.s fs0, ft0, fs0
+; RV32IF-NEXT:  .LBB7_2:
 ; RV32IF-NEXT:    fmv.w.x ft0, zero
-; RV32IF-NEXT:    fle.s a0, ft0, fa0
-; RV32IF-NEXT:    neg s1, a0
+; RV32IF-NEXT:    fle.s a0, ft0, fs0
+; RV32IF-NEXT:    neg s0, a0
+; RV32IF-NEXT:    fmv.s fa0, fs0
 ; RV32IF-NEXT:    call __fixunssfdi@plt
-; RV32IF-NEXT:    and a0, s1, a0
-; RV32IF-NEXT:    or a0, s0, a0
-; RV32IF-NEXT:    and a1, s1, a1
-; RV32IF-NEXT:    or a1, s0, a1
+; RV32IF-NEXT:    lui a2, %hi(.LCPI7_1)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI7_1)(a2)
+; RV32IF-NEXT:    and a0, s0, a0
+; RV32IF-NEXT:    flt.s a2, ft0, fs0
+; RV32IF-NEXT:    neg a2, a2
+; RV32IF-NEXT:    or a0, a2, a0
+; RV32IF-NEXT:    and a1, s0, a1
+; RV32IF-NEXT:    or a1, a2, a1
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
-; RV32IF-NEXT:    lw s1, 4(sp) # 4-byte Folded Reload
+; RV32IF-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_ceil_ui64:
 ; RV64IF:       # %bb.0:
 ; RV64IF-NEXT:    fcvt.lu.s a0, fa0, rup
 ; RV64IF-NEXT:    feq.s a1, fa0, fa0
 ; RV64IF-NEXT:    seqz a1, a1

 
 define i64 @test_trunc_si64(float %x) nounwind {
 ; RV32IF-LABEL: test_trunc_si64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
-; RV32IF-NEXT:    call truncf@plt
 ; RV32IF-NEXT:    lui a0, %hi(.LCPI9_0)
 ; RV32IF-NEXT:    flw ft0, %lo(.LCPI9_0)(a0)
 ; RV32IF-NEXT:    fmv.s fs0, fa0
-; RV32IF-NEXT:    fle.s s0, ft0, fa0
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB9_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fs0, rtz
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rtz
+; RV32IF-NEXT:    fsgnj.s fs0, ft0, fs0
+; RV32IF-NEXT:  .LBB9_2:
+; RV32IF-NEXT:    lui a0, %hi(.LCPI9_1)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI9_1)(a0)
+; RV32IF-NEXT:    fle.s s0, ft0, fs0
+; RV32IF-NEXT:    fmv.s fa0, fs0
 ; RV32IF-NEXT:    call __fixsfdi@plt
 ; RV32IF-NEXT:    lui a3, 524288
-; RV32IF-NEXT:    bnez s0, .LBB9_2
-; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    bnez s0, .LBB9_4
+; RV32IF-NEXT:  # %bb.3:
 ; RV32IF-NEXT:    lui a1, 524288
-; RV32IF-NEXT:  .LBB9_2:
-; RV32IF-NEXT:    lui a2, %hi(.LCPI9_1)
-; RV32IF-NEXT:    flw ft0, %lo(.LCPI9_1)(a2)
+; RV32IF-NEXT:  .LBB9_4:
+; RV32IF-NEXT:    lui a2, %hi(.LCPI9_2)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI9_2)(a2)
 ; RV32IF-NEXT:    flt.s a2, ft0, fs0
-; RV32IF-NEXT:    beqz a2, .LBB9_4
-; RV32IF-NEXT:  # %bb.3:
+; RV32IF-NEXT:    beqz a2, .LBB9_6
+; RV32IF-NEXT:  # %bb.5:
 ; RV32IF-NEXT:    addi a1, a3, -1
-; RV32IF-NEXT:  .LBB9_4:
+; RV32IF-NEXT:  .LBB9_6:
 ; RV32IF-NEXT:    feq.s a3, fs0, fs0
 ; RV32IF-NEXT:    seqz a3, a3
 ; RV32IF-NEXT:    addi a3, a3, -1
 ; RV32IF-NEXT:    and a1, a3, a1
-; RV32IF-NEXT:    neg a2, a2
 ; RV32IF-NEXT:    neg a4, s0
 ; RV32IF-NEXT:    and a0, a4, a0
+; RV32IF-NEXT:    neg a2, a2
 ; RV32IF-NEXT:    or a0, a2, a0
 ; RV32IF-NEXT:    and a0, a3, a0
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;

 }
 
 define i64 @test_trunc_ui64(float %x) nounwind {
 ; RV32IF-LABEL: test_trunc_ui64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
-; RV32IF-NEXT:    sw s1, 4(sp) # 4-byte Folded Spill
-; RV32IF-NEXT:    call truncf@plt
+; RV32IF-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    lui a0, %hi(.LCPI11_0)
 ; RV32IF-NEXT:    flw ft0, %lo(.LCPI11_0)(a0)
-; RV32IF-NEXT:    flt.s a0, ft0, fa0
-; RV32IF-NEXT:    neg s0, a0
+; RV32IF-NEXT:    fmv.s fs0, fa0
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB11_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fs0, rtz
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rtz
+; RV32IF-NEXT:    fsgnj.s fs0, ft0, fs0
+; RV32IF-NEXT:  .LBB11_2:
 ; RV32IF-NEXT:    fmv.w.x ft0, zero
-; RV32IF-NEXT:    fle.s a0, ft0, fa0
-; RV32IF-NEXT:    neg s1, a0
+; RV32IF-NEXT:    fle.s a0, ft0, fs0
+; RV32IF-NEXT:    neg s0, a0
+; RV32IF-NEXT:    fmv.s fa0, fs0
 ; RV32IF-NEXT:    call __fixunssfdi@plt
-; RV32IF-NEXT:    and a0, s1, a0
-; RV32IF-NEXT:    or a0, s0, a0
-; RV32IF-NEXT:    and a1, s1, a1
-; RV32IF-NEXT:    or a1, s0, a1
+; RV32IF-NEXT:    lui a2, %hi(.LCPI11_1)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI11_1)(a2)
+; RV32IF-NEXT:    and a0, s0, a0
+; RV32IF-NEXT:    flt.s a2, ft0, fs0
+; RV32IF-NEXT:    neg a2, a2
+; RV32IF-NEXT:    or a0, a2, a0
+; RV32IF-NEXT:    and a1, s0, a1
+; RV32IF-NEXT:    or a1, a2, a1
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
-; RV32IF-NEXT:    lw s1, 4(sp) # 4-byte Folded Reload
+; RV32IF-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_trunc_ui64:
 ; RV64IF:       # %bb.0:
 ; RV64IF-NEXT:    fcvt.lu.s a0, fa0, rtz
 ; RV64IF-NEXT:    feq.s a1, fa0, fa0
 ; RV64IF-NEXT:    seqz a1, a1

 
 define i64 @test_round_si64(float %x) nounwind {
 ; RV32IF-LABEL: test_round_si64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
-; RV32IF-NEXT:    call roundf@plt
 ; RV32IF-NEXT:    lui a0, %hi(.LCPI13_0)
 ; RV32IF-NEXT:    flw ft0, %lo(.LCPI13_0)(a0)
 ; RV32IF-NEXT:    fmv.s fs0, fa0
-; RV32IF-NEXT:    fle.s s0, ft0, fa0
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB13_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fs0, rmm
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rmm
+; RV32IF-NEXT:    fsgnj.s fs0, ft0, fs0
+; RV32IF-NEXT:  .LBB13_2:
+; RV32IF-NEXT:    lui a0, %hi(.LCPI13_1)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI13_1)(a0)
+; RV32IF-NEXT:    fle.s s0, ft0, fs0
+; RV32IF-NEXT:    fmv.s fa0, fs0
 ; RV32IF-NEXT:    call __fixsfdi@plt
 ; RV32IF-NEXT:    lui a3, 524288
-; RV32IF-NEXT:    bnez s0, .LBB13_2
-; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    bnez s0, .LBB13_4
+; RV32IF-NEXT:  # %bb.3:
 ; RV32IF-NEXT:    lui a1, 524288
-; RV32IF-NEXT:  .LBB13_2:
-; RV32IF-NEXT:    lui a2, %hi(.LCPI13_1)
-; RV32IF-NEXT:    flw ft0, %lo(.LCPI13_1)(a2)
+; RV32IF-NEXT:  .LBB13_4:
+; RV32IF-NEXT:    lui a2, %hi(.LCPI13_2)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI13_2)(a2)
 ; RV32IF-NEXT:    flt.s a2, ft0, fs0
-; RV32IF-NEXT:    beqz a2, .LBB13_4
-; RV32IF-NEXT:  # %bb.3:
+; RV32IF-NEXT:    beqz a2, .LBB13_6
+; RV32IF-NEXT:  # %bb.5:
 ; RV32IF-NEXT:    addi a1, a3, -1
-; RV32IF-NEXT:  .LBB13_4:
+; RV32IF-NEXT:  .LBB13_6:
 ; RV32IF-NEXT:    feq.s a3, fs0, fs0
 ; RV32IF-NEXT:    seqz a3, a3
 ; RV32IF-NEXT:    addi a3, a3, -1
 ; RV32IF-NEXT:    and a1, a3, a1
-; RV32IF-NEXT:    neg a2, a2
 ; RV32IF-NEXT:    neg a4, s0
 ; RV32IF-NEXT:    and a0, a4, a0
+; RV32IF-NEXT:    neg a2, a2
 ; RV32IF-NEXT:    or a0, a2, a0
 ; RV32IF-NEXT:    and a0, a3, a0
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;

 }
 
 define i64 @test_round_ui64(float %x) nounwind {
 ; RV32IF-LABEL: test_round_ui64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
-; RV32IF-NEXT:    sw s1, 4(sp) # 4-byte Folded Spill
-; RV32IF-NEXT:    call roundf@plt
+; RV32IF-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    lui a0, %hi(.LCPI15_0)
 ; RV32IF-NEXT:    flw ft0, %lo(.LCPI15_0)(a0)
-; RV32IF-NEXT:    flt.s a0, ft0, fa0
-; RV32IF-NEXT:    neg s0, a0
+; RV32IF-NEXT:    fmv.s fs0, fa0
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB15_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fs0, rmm
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rmm
+; RV32IF-NEXT:    fsgnj.s fs0, ft0, fs0
+; RV32IF-NEXT:  .LBB15_2:
 ; RV32IF-NEXT:    fmv.w.x ft0, zero
-; RV32IF-NEXT:    fle.s a0, ft0, fa0
-; RV32IF-NEXT:    neg s1, a0
+; RV32IF-NEXT:    fle.s a0, ft0, fs0
+; RV32IF-NEXT:    neg s0, a0
+; RV32IF-NEXT:    fmv.s fa0, fs0
 ; RV32IF-NEXT:    call __fixunssfdi@plt
-; RV32IF-NEXT:    and a0, s1, a0
-; RV32IF-NEXT:    or a0, s0, a0
-; RV32IF-NEXT:    and a1, s1, a1
-; RV32IF-NEXT:    or a1, s0, a1
+; RV32IF-NEXT:    lui a2, %hi(.LCPI15_1)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI15_1)(a2)
+; RV32IF-NEXT:    and a0, s0, a0
+; RV32IF-NEXT:    flt.s a2, ft0, fs0
+; RV32IF-NEXT:    neg a2, a2
+; RV32IF-NEXT:    or a0, a2, a0
+; RV32IF-NEXT:    and a1, s0, a1
+; RV32IF-NEXT:    or a1, a2, a1
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
-; RV32IF-NEXT:    lw s1, 4(sp) # 4-byte Folded Reload
+; RV32IF-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_round_ui64:
 ; RV64IF:       # %bb.0:
 ; RV64IF-NEXT:    fcvt.lu.s a0, fa0, rmm
 ; RV64IF-NEXT:    feq.s a1, fa0, fa0
 ; RV64IF-NEXT:    seqz a1, a1

 
 define i64 @test_roundeven_si64(float %x) nounwind {
 ; RV32IF-LABEL: test_roundeven_si64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
-; RV32IF-NEXT:    call roundevenf@plt
 ; RV32IF-NEXT:    lui a0, %hi(.LCPI17_0)
 ; RV32IF-NEXT:    flw ft0, %lo(.LCPI17_0)(a0)
 ; RV32IF-NEXT:    fmv.s fs0, fa0
-; RV32IF-NEXT:    fle.s s0, ft0, fa0
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB17_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fs0, rne
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rne
+; RV32IF-NEXT:    fsgnj.s fs0, ft0, fs0
+; RV32IF-NEXT:  .LBB17_2:
+; RV32IF-NEXT:    lui a0, %hi(.LCPI17_1)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI17_1)(a0)
+; RV32IF-NEXT:    fle.s s0, ft0, fs0
+; RV32IF-NEXT:    fmv.s fa0, fs0
 ; RV32IF-NEXT:    call __fixsfdi@plt
 ; RV32IF-NEXT:    lui a3, 524288
-; RV32IF-NEXT:    bnez s0, .LBB17_2
-; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    bnez s0, .LBB17_4
+; RV32IF-NEXT:  # %bb.3:
 ; RV32IF-NEXT:    lui a1, 524288
-; RV32IF-NEXT:  .LBB17_2:
-; RV32IF-NEXT:    lui a2, %hi(.LCPI17_1)
-; RV32IF-NEXT:    flw ft0, %lo(.LCPI17_1)(a2)
+; RV32IF-NEXT:  .LBB17_4:
+; RV32IF-NEXT:    lui a2, %hi(.LCPI17_2)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI17_2)(a2)
 ; RV32IF-NEXT:    flt.s a2, ft0, fs0
-; RV32IF-NEXT:    beqz a2, .LBB17_4
-; RV32IF-NEXT:  # %bb.3:
+; RV32IF-NEXT:    beqz a2, .LBB17_6
+; RV32IF-NEXT:  # %bb.5:
 ; RV32IF-NEXT:    addi a1, a3, -1
-; RV32IF-NEXT:  .LBB17_4:
+; RV32IF-NEXT:  .LBB17_6:
 ; RV32IF-NEXT:    feq.s a3, fs0, fs0
 ; RV32IF-NEXT:    seqz a3, a3
 ; RV32IF-NEXT:    addi a3, a3, -1
 ; RV32IF-NEXT:    and a1, a3, a1
-; RV32IF-NEXT:    neg a2, a2
 ; RV32IF-NEXT:    neg a4, s0
 ; RV32IF-NEXT:    and a0, a4, a0
+; RV32IF-NEXT:    neg a2, a2
 ; RV32IF-NEXT:    or a0, a2, a0
 ; RV32IF-NEXT:    and a0, a3, a0
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;

 }
 
 define i64 @test_roundeven_ui64(float %x) nounwind {
 ; RV32IF-LABEL: test_roundeven_ui64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
-; RV32IF-NEXT:    sw s1, 4(sp) # 4-byte Folded Spill
-; RV32IF-NEXT:    call roundevenf@plt
+; RV32IF-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    lui a0, %hi(.LCPI19_0)
 ; RV32IF-NEXT:    flw ft0, %lo(.LCPI19_0)(a0)
-; RV32IF-NEXT:    flt.s a0, ft0, fa0
-; RV32IF-NEXT:    neg s0, a0
+; RV32IF-NEXT:    fmv.s fs0, fa0
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB19_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fs0, rne
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rne
+; RV32IF-NEXT:    fsgnj.s fs0, ft0, fs0
+; RV32IF-NEXT:  .LBB19_2:
 ; RV32IF-NEXT:    fmv.w.x ft0, zero
-; RV32IF-NEXT:    fle.s a0, ft0, fa0
-; RV32IF-NEXT:    neg s1, a0
+; RV32IF-NEXT:    fle.s a0, ft0, fs0
+; RV32IF-NEXT:    neg s0, a0
+; RV32IF-NEXT:    fmv.s fa0, fs0
 ; RV32IF-NEXT:    call __fixunssfdi@plt
-; RV32IF-NEXT:    and a0, s1, a0
-; RV32IF-NEXT:    or a0, s0, a0
-; RV32IF-NEXT:    and a1, s1, a1
-; RV32IF-NEXT:    or a1, s0, a1
+; RV32IF-NEXT:    lui a2, %hi(.LCPI19_1)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI19_1)(a2)
+; RV32IF-NEXT:    and a0, s0, a0
+; RV32IF-NEXT:    flt.s a2, ft0, fs0
+; RV32IF-NEXT:    neg a2, a2
+; RV32IF-NEXT:    or a0, a2, a0
+; RV32IF-NEXT:    and a1, s0, a1
+; RV32IF-NEXT:    or a1, a2, a1
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
-; RV32IF-NEXT:    lw s1, 4(sp) # 4-byte Folded Reload
+; RV32IF-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_roundeven_ui64:
 ; RV64IF:       # %bb.0:
 ; RV64IF-NEXT:    fcvt.lu.s a0, fa0, rne
 ; RV64IF-NEXT:    feq.s a1, fa0, fa0
 ; RV64IF-NEXT:    seqz a1, a1

llvm/test/CodeGen/RISCV/float-round-conv.ll

 
 define i64 @test_floor_si64(float %x) {
 ; RV32IF-LABEL: test_floor_si64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    .cfi_offset ra, -4
-; RV32IF-NEXT:    call floorf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI3_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI3_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB3_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rdn
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rdn
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB3_2:
 ; RV32IF-NEXT:    call __fixsfdi@plt
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_floor_si64:
 ; RV64IF:       # %bb.0:
 ; RV64IF-NEXT:    fcvt.l.s a0, fa0, rdn

 
 define i64 @test_floor_ui64(float %x) {
 ; RV32IF-LABEL: test_floor_ui64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    .cfi_offset ra, -4
-; RV32IF-NEXT:    call floorf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI7_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI7_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB7_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rdn
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rdn
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB7_2:
 ; RV32IF-NEXT:    call __fixunssfdi@plt
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_floor_ui64:
 ; RV64IF:       # %bb.0:
 ; RV64IF-NEXT:    fcvt.lu.s a0, fa0, rdn

 
 define i64 @test_ceil_si64(float %x) {
 ; RV32IF-LABEL: test_ceil_si64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    .cfi_offset ra, -4
-; RV32IF-NEXT:    call ceilf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI11_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI11_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB11_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rup
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rup
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB11_2:
 ; RV32IF-NEXT:    call __fixsfdi@plt
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_ceil_si64:
 ; RV64IF:       # %bb.0:
 ; RV64IF-NEXT:    fcvt.l.s a0, fa0, rup

 
 define i64 @test_ceil_ui64(float %x) {
 ; RV32IF-LABEL: test_ceil_ui64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    .cfi_offset ra, -4
-; RV32IF-NEXT:    call ceilf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI15_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI15_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB15_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rup
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rup
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB15_2:
 ; RV32IF-NEXT:    call __fixunssfdi@plt
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_ceil_ui64:
 ; RV64IF:       # %bb.0:
 ; RV64IF-NEXT:    fcvt.lu.s a0, fa0, rup

 
 define i64 @test_trunc_si64(float %x) {
 ; RV32IF-LABEL: test_trunc_si64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    .cfi_offset ra, -4
-; RV32IF-NEXT:    call truncf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI19_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI19_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB19_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rtz
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rtz
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB19_2:
 ; RV32IF-NEXT:    call __fixsfdi@plt
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_trunc_si64:
 ; RV64IF:       # %bb.0:
 ; RV64IF-NEXT:    fcvt.l.s a0, fa0, rtz

 
 define i64 @test_trunc_ui64(float %x) {
 ; RV32IF-LABEL: test_trunc_ui64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    .cfi_offset ra, -4
-; RV32IF-NEXT:    call truncf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI23_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI23_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB23_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rtz
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rtz
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB23_2:
 ; RV32IF-NEXT:    call __fixunssfdi@plt
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_trunc_ui64:
 ; RV64IF:       # %bb.0:
 ; RV64IF-NEXT:    fcvt.lu.s a0, fa0, rtz

 
 define i64 @test_round_si64(float %x) {
 ; RV32IF-LABEL: test_round_si64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    .cfi_offset ra, -4
-; RV32IF-NEXT:    call roundf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI27_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI27_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB27_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rmm
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rmm
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB27_2:
 ; RV32IF-NEXT:    call __fixsfdi@plt
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_round_si64:
 ; RV64IF:       # %bb.0:
 ; RV64IF-NEXT:    fcvt.l.s a0, fa0, rmm

 
 define i64 @test_round_ui64(float %x) {
 ; RV32IF-LABEL: test_round_ui64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    .cfi_offset ra, -4
-; RV32IF-NEXT:    call roundf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI31_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI31_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB31_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rmm
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rmm
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB31_2:
 ; RV32IF-NEXT:    call __fixunssfdi@plt
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_round_ui64:
 ; RV64IF:       # %bb.0:
 ; RV64IF-NEXT:    fcvt.lu.s a0, fa0, rmm

 
 define i64 @test_roundeven_si64(float %x) {
 ; RV32IF-LABEL: test_roundeven_si64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    .cfi_offset ra, -4
-; RV32IF-NEXT:    call roundevenf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI35_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI35_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB35_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rne
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rne
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB35_2:
 ; RV32IF-NEXT:    call __fixsfdi@plt
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_roundeven_si64:
 ; RV64IF:       # %bb.0:
 ; RV64IF-NEXT:    fcvt.l.s a0, fa0, rne

 
 define i64 @test_roundeven_ui64(float %x) {
 ; RV32IF-LABEL: test_roundeven_ui64:
 ; RV32IF:       # %bb.0:
 ; RV32IF-NEXT:    addi sp, sp, -16
 ; RV32IF-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IF-NEXT:    .cfi_offset ra, -4
-; RV32IF-NEXT:    call roundevenf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI39_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI39_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB39_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rne
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rne
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB39_2:
 ; RV32IF-NEXT:    call __fixunssfdi@plt
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    addi sp, sp, 16
 ; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_roundeven_ui64:
 ; RV64IF:       # %bb.0:
 ; RV64IF-NEXT:    fcvt.lu.s a0, fa0, rne

 ; RV64IFD-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64IFD-NEXT:    .cfi_offset ra, -8
 ; RV64IFD-NEXT:    call floor@plt
 ; RV64IFD-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64IFD-NEXT:    addi sp, sp, 16
 ; RV64IFD-NEXT:    ret
 ; RV32IF-LABEL: test_floor_float:
 ; RV32IF:       # %bb.0:
-; RV32IF-NEXT:    tail floorf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI40_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI40_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB40_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rdn
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rdn
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB40_2:
+; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_floor_float:
 ; RV64IF:       # %bb.0:
-; RV64IF-NEXT:    tail floorf@plt
+; RV64IF-NEXT:    lui a0, %hi(.LCPI40_0)
+; RV64IF-NEXT:    flw ft0, %lo(.LCPI40_0)(a0)
+; RV64IF-NEXT:    fabs.s ft1, fa0
+; RV64IF-NEXT:    flt.s a0, ft1, ft0
+; RV64IF-NEXT:    beqz a0, .LBB40_2
+; RV64IF-NEXT:  # %bb.1:
+; RV64IF-NEXT:    fcvt.w.s a0, fa0, rdn
+; RV64IF-NEXT:    fcvt.s.w ft0, a0, rdn
+; RV64IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV64IF-NEXT:  .LBB40_2:
+; RV64IF-NEXT:    ret
   %a = call float @llvm.floor.f32(float %x)
   ret float %a
 }
 
 define float @test_ceil_float(float %x) {
 ; RV32IFD-LABEL: test_ceil_float:
 ; RV32IFD:       # %bb.0:
 ; RV32IFD-NEXT:    addi sp, sp, -16

 ; RV64IFD-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64IFD-NEXT:    .cfi_offset ra, -8
 ; RV64IFD-NEXT:    call ceil@plt
 ; RV64IFD-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64IFD-NEXT:    addi sp, sp, 16
 ; RV64IFD-NEXT:    ret
 ; RV32IF-LABEL: test_ceil_float:
 ; RV32IF:       # %bb.0:
-; RV32IF-NEXT:    tail ceilf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI41_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI41_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB41_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rup
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rup
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB41_2:
+; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_ceil_float:
 ; RV64IF:       # %bb.0:
-; RV64IF-NEXT:    tail ceilf@plt
+; RV64IF-NEXT:    lui a0, %hi(.LCPI41_0)
+; RV64IF-NEXT:    flw ft0, %lo(.LCPI41_0)(a0)
+; RV64IF-NEXT:    fabs.s ft1, fa0
+; RV64IF-NEXT:    flt.s a0, ft1, ft0
+; RV64IF-NEXT:    beqz a0, .LBB41_2
+; RV64IF-NEXT:  # %bb.1:
+; RV64IF-NEXT:    fcvt.w.s a0, fa0, rup
+; RV64IF-NEXT:    fcvt.s.w ft0, a0, rup
+; RV64IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV64IF-NEXT:  .LBB41_2:
+; RV64IF-NEXT:    ret
   %a = call float @llvm.ceil.f32(float %x)
   ret float %a
 }
 
 define float @test_trunc_float(float %x) {
 ; RV32IFD-LABEL: test_trunc_float:
 ; RV32IFD:       # %bb.0:
 ; RV32IFD-NEXT:    addi sp, sp, -16

 ; RV64IFD-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64IFD-NEXT:    .cfi_offset ra, -8
 ; RV64IFD-NEXT:    call trunc@plt
 ; RV64IFD-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64IFD-NEXT:    addi sp, sp, 16
 ; RV64IFD-NEXT:    ret
 ; RV32IF-LABEL: test_trunc_float:
 ; RV32IF:       # %bb.0:
-; RV32IF-NEXT:    tail truncf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI42_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI42_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB42_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rtz
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rtz
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB42_2:
+; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_trunc_float:
 ; RV64IF:       # %bb.0:
-; RV64IF-NEXT:    tail truncf@plt
+; RV64IF-NEXT:    lui a0, %hi(.LCPI42_0)
+; RV64IF-NEXT:    flw ft0, %lo(.LCPI42_0)(a0)
+; RV64IF-NEXT:    fabs.s ft1, fa0
+; RV64IF-NEXT:    flt.s a0, ft1, ft0
+; RV64IF-NEXT:    beqz a0, .LBB42_2
+; RV64IF-NEXT:  # %bb.1:
+; RV64IF-NEXT:    fcvt.w.s a0, fa0, rtz
+; RV64IF-NEXT:    fcvt.s.w ft0, a0, rtz
+; RV64IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV64IF-NEXT:  .LBB42_2:
+; RV64IF-NEXT:    ret
   %a = call float @llvm.trunc.f32(float %x)
   ret float %a
 }
 
 define float @test_round_float(float %x) {
 ; RV32IFD-LABEL: test_round_float:
 ; RV32IFD:       # %bb.0:
 ; RV32IFD-NEXT:    addi sp, sp, -16

 ; RV64IFD-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64IFD-NEXT:    .cfi_offset ra, -8
 ; RV64IFD-NEXT:    call round@plt
 ; RV64IFD-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64IFD-NEXT:    addi sp, sp, 16
 ; RV64IFD-NEXT:    ret
 ; RV32IF-LABEL: test_round_float:
 ; RV32IF:       # %bb.0:
-; RV32IF-NEXT:    tail roundf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI43_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI43_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB43_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rmm
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rmm
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB43_2:
+; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_round_float:
 ; RV64IF:       # %bb.0:
-; RV64IF-NEXT:    tail roundf@plt
+; RV64IF-NEXT:    lui a0, %hi(.LCPI43_0)
+; RV64IF-NEXT:    flw ft0, %lo(.LCPI43_0)(a0)
+; RV64IF-NEXT:    fabs.s ft1, fa0
+; RV64IF-NEXT:    flt.s a0, ft1, ft0
+; RV64IF-NEXT:    beqz a0, .LBB43_2
+; RV64IF-NEXT:  # %bb.1:
+; RV64IF-NEXT:    fcvt.w.s a0, fa0, rmm
+; RV64IF-NEXT:    fcvt.s.w ft0, a0, rmm
+; RV64IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV64IF-NEXT:  .LBB43_2:
+; RV64IF-NEXT:    ret
   %a = call float @llvm.round.f32(float %x)
   ret float %a
 }
 
 define float @test_roundeven_float(float %x) {
 ; RV32IFD-LABEL: test_roundeven_float:
 ; RV32IFD:       # %bb.0:
 ; RV32IFD-NEXT:    addi sp, sp, -16

 ; RV64IFD-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64IFD-NEXT:    .cfi_offset ra, -8
 ; RV64IFD-NEXT:    call roundeven@plt
 ; RV64IFD-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64IFD-NEXT:    addi sp, sp, 16
 ; RV64IFD-NEXT:    ret
 ; RV32IF-LABEL: test_roundeven_float:
 ; RV32IF:       # %bb.0:
-; RV32IF-NEXT:    tail roundevenf@plt
+; RV32IF-NEXT:    lui a0, %hi(.LCPI44_0)
+; RV32IF-NEXT:    flw ft0, %lo(.LCPI44_0)(a0)
+; RV32IF-NEXT:    fabs.s ft1, fa0
+; RV32IF-NEXT:    flt.s a0, ft1, ft0
+; RV32IF-NEXT:    beqz a0, .LBB44_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    fcvt.w.s a0, fa0, rne
+; RV32IF-NEXT:    fcvt.s.w ft0, a0, rne
+; RV32IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV32IF-NEXT:  .LBB44_2:
+; RV32IF-NEXT:    ret
 ;
 ; RV64IF-LABEL: test_roundeven_float:
 ; RV64IF:       # %bb.0:
-; RV64IF-NEXT:    tail roundevenf@plt
+; RV64IF-NEXT:    lui a0, %hi(.LCPI44_0)
+; RV64IF-NEXT:    flw ft0, %lo(.LCPI44_0)(a0)
+; RV64IF-NEXT:    fabs.s ft1, fa0
+; RV64IF-NEXT:    flt.s a0, ft1, ft0
+; RV64IF-NEXT:    beqz a0, .LBB44_2
+; RV64IF-NEXT:  # %bb.1:
+; RV64IF-NEXT:    fcvt.w.s a0, fa0, rne
+; RV64IF-NEXT:    fcvt.s.w ft0, a0, rne
+; RV64IF-NEXT:    fsgnj.s fa0, ft0, fa0
+; RV64IF-NEXT:  .LBB44_2:
+; RV64IF-NEXT:    ret
   %a = call float @llvm.roundeven.f32(float %x)
   ret float %a
 }
 
 declare float @llvm.floor.f32(float)
 declare float @llvm.ceil.f32(float)
 declare float @llvm.trunc.f32(float)
 declare float @llvm.round.f32(float)
 declare float @llvm.roundeven.f32(float)

llvm/test/CodeGen/RISCV/half-intrinsics.ll

 ; RV64I-NEXT:    ret
   %1 = call half @llvm.copysign.f16(half %a, half %b)
   ret half %1
 }
 
 declare half @llvm.floor.f16(half)
 
 define half @floor_f16(half %a) nounwind {
-; RV32IZFH-LABEL: floor_f16:
-; RV32IZFH:       # %bb.0:
-; RV32IZFH-NEXT:    addi sp, sp, -16
-; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call floorf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    addi sp, sp, 16
-; RV32IZFH-NEXT:    ret
-;
-; RV64IZFH-LABEL: floor_f16:
-; RV64IZFH:       # %bb.0:
-; RV64IZFH-NEXT:    addi sp, sp, -16
-; RV64IZFH-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
-; RV64IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV64IZFH-NEXT:    call floorf@plt
-; RV64IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV64IZFH-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
-; RV64IZFH-NEXT:    addi sp, sp, 16
-; RV64IZFH-NEXT:    ret
+; CHECKIZFH-LABEL: floor_f16:
+; CHECKIZFH:       # %bb.0:
+; CHECKIZFH-NEXT:    lui a0, %hi(.LCPI17_0)
+; CHECKIZFH-NEXT:    flh ft0, %lo(.LCPI17_0)(a0)
+; CHECKIZFH-NEXT:    fabs.h ft1, fa0
+; CHECKIZFH-NEXT:    flt.h a0, ft1, ft0
+; CHECKIZFH-NEXT:    beqz a0, .LBB17_2
+; CHECKIZFH-NEXT:  # %bb.1:
+; CHECKIZFH-NEXT:    fcvt.w.h a0, fa0, rdn
+; CHECKIZFH-NEXT:    fcvt.h.w ft0, a0, rdn
+; CHECKIZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; CHECKIZFH-NEXT:  .LBB17_2:
+; CHECKIZFH-NEXT:    ret
 ;
 ; RV32I-LABEL: floor_f16:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    slli a0, a0, 16
 ; RV32I-NEXT:    srli a0, a0, 16
 ; RV32I-NEXT:    call __extendhfsf2@plt

 ; RV64I-NEXT:    ret
   %1 = call half @llvm.floor.f16(half %a)
   ret half %1
 }
 
 declare half @llvm.ceil.f16(half)
 
 define half @ceil_f16(half %a) nounwind {
-; RV32IZFH-LABEL: ceil_f16:
-; RV32IZFH:       # %bb.0:
-; RV32IZFH-NEXT:    addi sp, sp, -16
-; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call ceilf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    addi sp, sp, 16
-; RV32IZFH-NEXT:    ret
-;
-; RV64IZFH-LABEL: ceil_f16:
-; RV64IZFH:       # %bb.0:
-; RV64IZFH-NEXT:    addi sp, sp, -16
-; RV64IZFH-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
-; RV64IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV64IZFH-NEXT:    call ceilf@plt
-; RV64IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV64IZFH-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
-; RV64IZFH-NEXT:    addi sp, sp, 16
-; RV64IZFH-NEXT:    ret
+; CHECKIZFH-LABEL: ceil_f16:
+; CHECKIZFH:       # %bb.0:
+; CHECKIZFH-NEXT:    lui a0, %hi(.LCPI18_0)
+; CHECKIZFH-NEXT:    flh ft0, %lo(.LCPI18_0)(a0)
+; CHECKIZFH-NEXT:    fabs.h ft1, fa0
+; CHECKIZFH-NEXT:    flt.h a0, ft1, ft0
+; CHECKIZFH-NEXT:    beqz a0, .LBB18_2
+; CHECKIZFH-NEXT:  # %bb.1:
+; CHECKIZFH-NEXT:    fcvt.w.h a0, fa0, rup
+; CHECKIZFH-NEXT:    fcvt.h.w ft0, a0, rup
+; CHECKIZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; CHECKIZFH-NEXT:  .LBB18_2:
+; CHECKIZFH-NEXT:    ret
 ;
 ; RV32I-LABEL: ceil_f16:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    slli a0, a0, 16
 ; RV32I-NEXT:    srli a0, a0, 16
 ; RV32I-NEXT:    call __extendhfsf2@plt

 ; RV64I-NEXT:    ret
   %1 = call half @llvm.ceil.f16(half %a)
   ret half %1
 }
 
 declare half @llvm.trunc.f16(half)
 
 define half @trunc_f16(half %a) nounwind {
-; RV32IZFH-LABEL: trunc_f16:
-; RV32IZFH:       # %bb.0:
-; RV32IZFH-NEXT:    addi sp, sp, -16
-; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call truncf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    addi sp, sp, 16
-; RV32IZFH-NEXT:    ret
-;
-; RV64IZFH-LABEL: trunc_f16:
-; RV64IZFH:       # %bb.0:
-; RV64IZFH-NEXT:    addi sp, sp, -16
-; RV64IZFH-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
-; RV64IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV64IZFH-NEXT:    call truncf@plt
-; RV64IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV64IZFH-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
-; RV64IZFH-NEXT:    addi sp, sp, 16
-; RV64IZFH-NEXT:    ret
+; CHECKIZFH-LABEL: trunc_f16:
+; CHECKIZFH:       # %bb.0:
+; CHECKIZFH-NEXT:    lui a0, %hi(.LCPI19_0)
+; CHECKIZFH-NEXT:    flh ft0, %lo(.LCPI19_0)(a0)
+; CHECKIZFH-NEXT:    fabs.h ft1, fa0
+; CHECKIZFH-NEXT:    flt.h a0, ft1, ft0
+; CHECKIZFH-NEXT:    beqz a0, .LBB19_2
+; CHECKIZFH-NEXT:  # %bb.1:
+; CHECKIZFH-NEXT:    fcvt.w.h a0, fa0, rtz
+; CHECKIZFH-NEXT:    fcvt.h.w ft0, a0, rtz
+; CHECKIZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; CHECKIZFH-NEXT:  .LBB19_2:
+; CHECKIZFH-NEXT:    ret
 ;
 ; RV32I-LABEL: trunc_f16:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    slli a0, a0, 16
 ; RV32I-NEXT:    srli a0, a0, 16
 ; RV32I-NEXT:    call __extendhfsf2@plt

 ; RV64I-NEXT:    ret
   %1 = call half @llvm.trunc.f16(half %a)
   ret half %1
 }
 
 declare half @llvm.rint.f16(half)
 
 define half @rint_f16(half %a) nounwind {
-; RV32IZFH-LABEL: rint_f16:
-; RV32IZFH:       # %bb.0:
-; RV32IZFH-NEXT:    addi sp, sp, -16
-; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call rintf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    addi sp, sp, 16
-; RV32IZFH-NEXT:    ret
-;
-; RV64IZFH-LABEL: rint_f16:
-; RV64IZFH:       # %bb.0:
-; RV64IZFH-NEXT:    addi sp, sp, -16
-; RV64IZFH-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
-; RV64IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV64IZFH-NEXT:    call rintf@plt
-; RV64IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV64IZFH-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
-; RV64IZFH-NEXT:    addi sp, sp, 16
-; RV64IZFH-NEXT:    ret
+; CHECKIZFH-LABEL: rint_f16:
+; CHECKIZFH:       # %bb.0:
+; CHECKIZFH-NEXT:    lui a0, %hi(.LCPI20_0)
+; CHECKIZFH-NEXT:    flh ft0, %lo(.LCPI20_0)(a0)
+; CHECKIZFH-NEXT:    fabs.h ft1, fa0
+; CHECKIZFH-NEXT:    flt.h a0, ft1, ft0
+; CHECKIZFH-NEXT:    beqz a0, .LBB20_2
+; CHECKIZFH-NEXT:  # %bb.1:
+; CHECKIZFH-NEXT:    fcvt.w.h a0, fa0
+; CHECKIZFH-NEXT:    fcvt.h.w ft0, a0
+; CHECKIZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; CHECKIZFH-NEXT:  .LBB20_2:
+; CHECKIZFH-NEXT:    ret
 ;
 ; RV32I-LABEL: rint_f16:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    slli a0, a0, 16
 ; RV32I-NEXT:    srli a0, a0, 16
 ; RV32I-NEXT:    call __extendhfsf2@plt

 ; RV64I-NEXT:    ret
   %1 = call half @llvm.nearbyint.f16(half %a)
   ret half %1
 }
 
 declare half @llvm.round.f16(half)
 
 define half @round_f16(half %a) nounwind {
-; RV32IZFH-LABEL: round_f16:
-; RV32IZFH:       # %bb.0:
-; RV32IZFH-NEXT:    addi sp, sp, -16
-; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call roundf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    addi sp, sp, 16
-; RV32IZFH-NEXT:    ret
-;
-; RV64IZFH-LABEL: round_f16:
-; RV64IZFH:       # %bb.0:
-; RV64IZFH-NEXT:    addi sp, sp, -16
-; RV64IZFH-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
-; RV64IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV64IZFH-NEXT:    call roundf@plt
-; RV64IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV64IZFH-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
-; RV64IZFH-NEXT:    addi sp, sp, 16
-; RV64IZFH-NEXT:    ret
+; CHECKIZFH-LABEL: round_f16:
+; CHECKIZFH:       # %bb.0:
+; CHECKIZFH-NEXT:    lui a0, %hi(.LCPI22_0)
+; CHECKIZFH-NEXT:    flh ft0, %lo(.LCPI22_0)(a0)
+; CHECKIZFH-NEXT:    fabs.h ft1, fa0
+; CHECKIZFH-NEXT:    flt.h a0, ft1, ft0
+; CHECKIZFH-NEXT:    beqz a0, .LBB22_2
+; CHECKIZFH-NEXT:  # %bb.1:
+; CHECKIZFH-NEXT:    fcvt.w.h a0, fa0, rmm
+; CHECKIZFH-NEXT:    fcvt.h.w ft0, a0, rmm
+; CHECKIZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; CHECKIZFH-NEXT:  .LBB22_2:
+; CHECKIZFH-NEXT:    ret
 ;
 ; RV32I-LABEL: round_f16:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    slli a0, a0, 16
 ; RV32I-NEXT:    srli a0, a0, 16
 ; RV32I-NEXT:    call __extendhfsf2@plt

 ; RV64I-NEXT:    ret
   %1 = call half @llvm.round.f16(half %a)
   ret half %1
 }
 
 declare half @llvm.roundeven.f16(half)
 
 define half @roundeven_f16(half %a) nounwind {
-; RV32IZFH-LABEL: roundeven_f16:
-; RV32IZFH:       # %bb.0:
-; RV32IZFH-NEXT:    addi sp, sp, -16
-; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call roundevenf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    addi sp, sp, 16
-; RV32IZFH-NEXT:    ret
-;
-; RV64IZFH-LABEL: roundeven_f16:
-; RV64IZFH:       # %bb.0:
-; RV64IZFH-NEXT:    addi sp, sp, -16
-; RV64IZFH-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
-; RV64IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV64IZFH-NEXT:    call roundevenf@plt
-; RV64IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV64IZFH-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
-; RV64IZFH-NEXT:    addi sp, sp, 16
-; RV64IZFH-NEXT:    ret
+; CHECKIZFH-LABEL: roundeven_f16:
+; CHECKIZFH:       # %bb.0:
+; CHECKIZFH-NEXT:    lui a0, %hi(.LCPI23_0)
+; CHECKIZFH-NEXT:    flh ft0, %lo(.LCPI23_0)(a0)
+; CHECKIZFH-NEXT:    fabs.h ft1, fa0
+; CHECKIZFH-NEXT:    flt.h a0, ft1, ft0
+; CHECKIZFH-NEXT:    beqz a0, .LBB23_2
+; CHECKIZFH-NEXT:  # %bb.1:
+; CHECKIZFH-NEXT:    fcvt.w.h a0, fa0, rne
+; CHECKIZFH-NEXT:    fcvt.h.w ft0, a0, rne
+; CHECKIZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; CHECKIZFH-NEXT:  .LBB23_2:
+; CHECKIZFH-NEXT:    ret
 ;
 ; RV32I-LABEL: roundeven_f16:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    addi sp, sp, -16
 ; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-NEXT:    slli a0, a0, 16
 ; RV32I-NEXT:    srli a0, a0, 16
 ; RV32I-NEXT:    call __extendhfsf2@plt

llvm/test/CodeGen/RISCV/half-round-conv-sat.ll

 
 define i64 @test_floor_si64(half %x) nounwind {
 ; RV32IZFH-LABEL: test_floor_si64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call floorf@plt
 ; RV32IZFH-NEXT:    lui a0, %hi(.LCPI1_0)
-; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI1_0)(a0)
-; RV32IZFH-NEXT:    fcvt.h.s ft1, fa0
-; RV32IZFH-NEXT:    fcvt.s.h fs0, ft1
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI1_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB1_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rdn
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rdn
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB1_2:
+; RV32IZFH-NEXT:    lui a0, %hi(.LCPI1_1)
+; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI1_1)(a0)
+; RV32IZFH-NEXT:    fcvt.s.h fs0, fa0
 ; RV32IZFH-NEXT:    fle.s s0, ft0, fs0
 ; RV32IZFH-NEXT:    fmv.s fa0, fs0
 ; RV32IZFH-NEXT:    call __fixsfdi@plt
 ; RV32IZFH-NEXT:    lui a3, 524288
-; RV32IZFH-NEXT:    bnez s0, .LBB1_2
-; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    bnez s0, .LBB1_4
+; RV32IZFH-NEXT:  # %bb.3:
 ; RV32IZFH-NEXT:    lui a1, 524288
-; RV32IZFH-NEXT:  .LBB1_2:
-; RV32IZFH-NEXT:    lui a2, %hi(.LCPI1_1)
-; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI1_1)(a2)
+; RV32IZFH-NEXT:  .LBB1_4:
+; RV32IZFH-NEXT:    lui a2, %hi(.LCPI1_2)
+; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI1_2)(a2)
 ; RV32IZFH-NEXT:    flt.s a2, ft0, fs0
-; RV32IZFH-NEXT:    beqz a2, .LBB1_4
-; RV32IZFH-NEXT:  # %bb.3:
+; RV32IZFH-NEXT:    beqz a2, .LBB1_6
+; RV32IZFH-NEXT:  # %bb.5:
 ; RV32IZFH-NEXT:    addi a1, a3, -1
-; RV32IZFH-NEXT:  .LBB1_4:
+; RV32IZFH-NEXT:  .LBB1_6:
 ; RV32IZFH-NEXT:    feq.s a3, fs0, fs0
 ; RV32IZFH-NEXT:    seqz a3, a3
 ; RV32IZFH-NEXT:    addi a3, a3, -1
 ; RV32IZFH-NEXT:    and a1, a3, a1
-; RV32IZFH-NEXT:    neg a2, a2
 ; RV32IZFH-NEXT:    neg a4, s0
 ; RV32IZFH-NEXT:    and a0, a4, a0
+; RV32IZFH-NEXT:    neg a2, a2
 ; RV32IZFH-NEXT:    or a0, a2, a0
 ; RV32IZFH-NEXT:    and a0, a3, a0
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;

 }
 
 define i64 @test_floor_ui64(half %x) nounwind {
 ; RV32IZFH-LABEL: test_floor_ui64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    sw s1, 4(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call floorf@plt
+; RV32IZFH-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    lui a0, %hi(.LCPI3_0)
-; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI3_0)(a0)
-; RV32IZFH-NEXT:    fcvt.h.s ft1, fa0
-; RV32IZFH-NEXT:    fcvt.s.h fa0, ft1
-; RV32IZFH-NEXT:    flt.s a0, ft0, fa0
-; RV32IZFH-NEXT:    neg s0, a0
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI3_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB3_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rdn
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rdn
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB3_2:
+; RV32IZFH-NEXT:    fcvt.s.h fs0, fa0
 ; RV32IZFH-NEXT:    fmv.w.x ft0, zero
-; RV32IZFH-NEXT:    fle.s a0, ft0, fa0
-; RV32IZFH-NEXT:    neg s1, a0
+; RV32IZFH-NEXT:    fle.s a0, ft0, fs0
+; RV32IZFH-NEXT:    neg s0, a0
+; RV32IZFH-NEXT:    fmv.s fa0, fs0
 ; RV32IZFH-NEXT:    call __fixunssfdi@plt
-; RV32IZFH-NEXT:    and a0, s1, a0
-; RV32IZFH-NEXT:    or a0, s0, a0
-; RV32IZFH-NEXT:    and a1, s1, a1
-; RV32IZFH-NEXT:    or a1, s0, a1
+; RV32IZFH-NEXT:    lui a2, %hi(.LCPI3_1)
+; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI3_1)(a2)
+; RV32IZFH-NEXT:    and a0, s0, a0
+; RV32IZFH-NEXT:    flt.s a2, ft0, fs0
+; RV32IZFH-NEXT:    neg a2, a2
+; RV32IZFH-NEXT:    or a0, a2, a0
+; RV32IZFH-NEXT:    and a1, s0, a1
+; RV32IZFH-NEXT:    or a1, a2, a1
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    lw s1, 4(sp) # 4-byte Folded Reload
+; RV32IZFH-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;
 ; RV64IZFH-LABEL: test_floor_ui64:
 ; RV64IZFH:       # %bb.0:
 ; RV64IZFH-NEXT:    fcvt.lu.h a0, fa0, rdn
 ; RV64IZFH-NEXT:    feq.h a1, fa0, fa0
 ; RV64IZFH-NEXT:    seqz a1, a1

 
 define i64 @test_ceil_si64(half %x) nounwind {
 ; RV32IZFH-LABEL: test_ceil_si64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call ceilf@plt
 ; RV32IZFH-NEXT:    lui a0, %hi(.LCPI5_0)
-; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI5_0)(a0)
-; RV32IZFH-NEXT:    fcvt.h.s ft1, fa0
-; RV32IZFH-NEXT:    fcvt.s.h fs0, ft1
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI5_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB5_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rup
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rup
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB5_2:
+; RV32IZFH-NEXT:    lui a0, %hi(.LCPI5_1)
+; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI5_1)(a0)
+; RV32IZFH-NEXT:    fcvt.s.h fs0, fa0
 ; RV32IZFH-NEXT:    fle.s s0, ft0, fs0
 ; RV32IZFH-NEXT:    fmv.s fa0, fs0
 ; RV32IZFH-NEXT:    call __fixsfdi@plt
 ; RV32IZFH-NEXT:    lui a3, 524288
-; RV32IZFH-NEXT:    bnez s0, .LBB5_2
-; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    bnez s0, .LBB5_4
+; RV32IZFH-NEXT:  # %bb.3:
 ; RV32IZFH-NEXT:    lui a1, 524288
-; RV32IZFH-NEXT:  .LBB5_2:
-; RV32IZFH-NEXT:    lui a2, %hi(.LCPI5_1)
-; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI5_1)(a2)
+; RV32IZFH-NEXT:  .LBB5_4:
+; RV32IZFH-NEXT:    lui a2, %hi(.LCPI5_2)
+; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI5_2)(a2)
 ; RV32IZFH-NEXT:    flt.s a2, ft0, fs0
-; RV32IZFH-NEXT:    beqz a2, .LBB5_4
-; RV32IZFH-NEXT:  # %bb.3:
+; RV32IZFH-NEXT:    beqz a2, .LBB5_6
+; RV32IZFH-NEXT:  # %bb.5:
 ; RV32IZFH-NEXT:    addi a1, a3, -1
-; RV32IZFH-NEXT:  .LBB5_4:
+; RV32IZFH-NEXT:  .LBB5_6:
 ; RV32IZFH-NEXT:    feq.s a3, fs0, fs0
 ; RV32IZFH-NEXT:    seqz a3, a3
 ; RV32IZFH-NEXT:    addi a3, a3, -1
 ; RV32IZFH-NEXT:    and a1, a3, a1
-; RV32IZFH-NEXT:    neg a2, a2
 ; RV32IZFH-NEXT:    neg a4, s0
 ; RV32IZFH-NEXT:    and a0, a4, a0
+; RV32IZFH-NEXT:    neg a2, a2
 ; RV32IZFH-NEXT:    or a0, a2, a0
 ; RV32IZFH-NEXT:    and a0, a3, a0
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;

 }
 
 define i64 @test_ceil_ui64(half %x) nounwind {
 ; RV32IZFH-LABEL: test_ceil_ui64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    sw s1, 4(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call ceilf@plt
+; RV32IZFH-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    lui a0, %hi(.LCPI7_0)
-; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI7_0)(a0)
-; RV32IZFH-NEXT:    fcvt.h.s ft1, fa0
-; RV32IZFH-NEXT:    fcvt.s.h fa0, ft1
-; RV32IZFH-NEXT:    flt.s a0, ft0, fa0
-; RV32IZFH-NEXT:    neg s0, a0
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI7_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB7_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rup
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rup
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB7_2:
+; RV32IZFH-NEXT:    fcvt.s.h fs0, fa0
 ; RV32IZFH-NEXT:    fmv.w.x ft0, zero
-; RV32IZFH-NEXT:    fle.s a0, ft0, fa0
-; RV32IZFH-NEXT:    neg s1, a0
+; RV32IZFH-NEXT:    fle.s a0, ft0, fs0
+; RV32IZFH-NEXT:    neg s0, a0
+; RV32IZFH-NEXT:    fmv.s fa0, fs0
 ; RV32IZFH-NEXT:    call __fixunssfdi@plt
-; RV32IZFH-NEXT:    and a0, s1, a0
-; RV32IZFH-NEXT:    or a0, s0, a0
-; RV32IZFH-NEXT:    and a1, s1, a1
-; RV32IZFH-NEXT:    or a1, s0, a1
+; RV32IZFH-NEXT:    lui a2, %hi(.LCPI7_1)
+; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI7_1)(a2)
+; RV32IZFH-NEXT:    and a0, s0, a0
+; RV32IZFH-NEXT:    flt.s a2, ft0, fs0
+; RV32IZFH-NEXT:    neg a2, a2
+; RV32IZFH-NEXT:    or a0, a2, a0
+; RV32IZFH-NEXT:    and a1, s0, a1
+; RV32IZFH-NEXT:    or a1, a2, a1
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    lw s1, 4(sp) # 4-byte Folded Reload
+; RV32IZFH-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;
 ; RV64IZFH-LABEL: test_ceil_ui64:
 ; RV64IZFH:       # %bb.0:
 ; RV64IZFH-NEXT:    fcvt.lu.h a0, fa0, rup
 ; RV64IZFH-NEXT:    feq.h a1, fa0, fa0
 ; RV64IZFH-NEXT:    seqz a1, a1

 
 define i64 @test_trunc_si64(half %x) nounwind {
 ; RV32IZFH-LABEL: test_trunc_si64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call truncf@plt
 ; RV32IZFH-NEXT:    lui a0, %hi(.LCPI9_0)
-; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI9_0)(a0)
-; RV32IZFH-NEXT:    fcvt.h.s ft1, fa0
-; RV32IZFH-NEXT:    fcvt.s.h fs0, ft1
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI9_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB9_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rtz
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rtz
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB9_2:
+; RV32IZFH-NEXT:    lui a0, %hi(.LCPI9_1)
+; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI9_1)(a0)
+; RV32IZFH-NEXT:    fcvt.s.h fs0, fa0
 ; RV32IZFH-NEXT:    fle.s s0, ft0, fs0
 ; RV32IZFH-NEXT:    fmv.s fa0, fs0
 ; RV32IZFH-NEXT:    call __fixsfdi@plt
 ; RV32IZFH-NEXT:    lui a3, 524288
-; RV32IZFH-NEXT:    bnez s0, .LBB9_2
-; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    bnez s0, .LBB9_4
+; RV32IZFH-NEXT:  # %bb.3:
 ; RV32IZFH-NEXT:    lui a1, 524288
-; RV32IZFH-NEXT:  .LBB9_2:
-; RV32IZFH-NEXT:    lui a2, %hi(.LCPI9_1)
-; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI9_1)(a2)
+; RV32IZFH-NEXT:  .LBB9_4:
+; RV32IZFH-NEXT:    lui a2, %hi(.LCPI9_2)
+; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI9_2)(a2)
 ; RV32IZFH-NEXT:    flt.s a2, ft0, fs0
-; RV32IZFH-NEXT:    beqz a2, .LBB9_4
-; RV32IZFH-NEXT:  # %bb.3:
+; RV32IZFH-NEXT:    beqz a2, .LBB9_6
+; RV32IZFH-NEXT:  # %bb.5:
 ; RV32IZFH-NEXT:    addi a1, a3, -1
-; RV32IZFH-NEXT:  .LBB9_4:
+; RV32IZFH-NEXT:  .LBB9_6:
 ; RV32IZFH-NEXT:    feq.s a3, fs0, fs0
 ; RV32IZFH-NEXT:    seqz a3, a3
 ; RV32IZFH-NEXT:    addi a3, a3, -1
 ; RV32IZFH-NEXT:    and a1, a3, a1
-; RV32IZFH-NEXT:    neg a2, a2
 ; RV32IZFH-NEXT:    neg a4, s0
 ; RV32IZFH-NEXT:    and a0, a4, a0
+; RV32IZFH-NEXT:    neg a2, a2
 ; RV32IZFH-NEXT:    or a0, a2, a0
 ; RV32IZFH-NEXT:    and a0, a3, a0
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;

 }
 
 define i64 @test_trunc_ui64(half %x) nounwind {
 ; RV32IZFH-LABEL: test_trunc_ui64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    sw s1, 4(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call truncf@plt
+; RV32IZFH-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    lui a0, %hi(.LCPI11_0)
-; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI11_0)(a0)
-; RV32IZFH-NEXT:    fcvt.h.s ft1, fa0
-; RV32IZFH-NEXT:    fcvt.s.h fa0, ft1
-; RV32IZFH-NEXT:    flt.s a0, ft0, fa0
-; RV32IZFH-NEXT:    neg s0, a0
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI11_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB11_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rtz
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rtz
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB11_2:
+; RV32IZFH-NEXT:    fcvt.s.h fs0, fa0
 ; RV32IZFH-NEXT:    fmv.w.x ft0, zero
-; RV32IZFH-NEXT:    fle.s a0, ft0, fa0
-; RV32IZFH-NEXT:    neg s1, a0
+; RV32IZFH-NEXT:    fle.s a0, ft0, fs0
+; RV32IZFH-NEXT:    neg s0, a0
+; RV32IZFH-NEXT:    fmv.s fa0, fs0
 ; RV32IZFH-NEXT:    call __fixunssfdi@plt
-; RV32IZFH-NEXT:    and a0, s1, a0
-; RV32IZFH-NEXT:    or a0, s0, a0
-; RV32IZFH-NEXT:    and a1, s1, a1
-; RV32IZFH-NEXT:    or a1, s0, a1
+; RV32IZFH-NEXT:    lui a2, %hi(.LCPI11_1)
+; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI11_1)(a2)
+; RV32IZFH-NEXT:    and a0, s0, a0
+; RV32IZFH-NEXT:    flt.s a2, ft0, fs0
+; RV32IZFH-NEXT:    neg a2, a2
+; RV32IZFH-NEXT:    or a0, a2, a0
+; RV32IZFH-NEXT:    and a1, s0, a1
+; RV32IZFH-NEXT:    or a1, a2, a1
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    lw s1, 4(sp) # 4-byte Folded Reload
+; RV32IZFH-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;
 ; RV64IZFH-LABEL: test_trunc_ui64:
 ; RV64IZFH:       # %bb.0:
 ; RV64IZFH-NEXT:    fcvt.lu.h a0, fa0, rtz
 ; RV64IZFH-NEXT:    feq.h a1, fa0, fa0
 ; RV64IZFH-NEXT:    seqz a1, a1

 
 define i64 @test_round_si64(half %x) nounwind {
 ; RV32IZFH-LABEL: test_round_si64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call roundf@plt
 ; RV32IZFH-NEXT:    lui a0, %hi(.LCPI13_0)
-; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI13_0)(a0)
-; RV32IZFH-NEXT:    fcvt.h.s ft1, fa0
-; RV32IZFH-NEXT:    fcvt.s.h fs0, ft1
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI13_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB13_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rmm
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rmm
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB13_2:
+; RV32IZFH-NEXT:    lui a0, %hi(.LCPI13_1)
+; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI13_1)(a0)
+; RV32IZFH-NEXT:    fcvt.s.h fs0, fa0
 ; RV32IZFH-NEXT:    fle.s s0, ft0, fs0
 ; RV32IZFH-NEXT:    fmv.s fa0, fs0
 ; RV32IZFH-NEXT:    call __fixsfdi@plt
 ; RV32IZFH-NEXT:    lui a3, 524288
-; RV32IZFH-NEXT:    bnez s0, .LBB13_2
-; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    bnez s0, .LBB13_4
+; RV32IZFH-NEXT:  # %bb.3:
 ; RV32IZFH-NEXT:    lui a1, 524288
-; RV32IZFH-NEXT:  .LBB13_2:
-; RV32IZFH-NEXT:    lui a2, %hi(.LCPI13_1)
-; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI13_1)(a2)
+; RV32IZFH-NEXT:  .LBB13_4:
+; RV32IZFH-NEXT:    lui a2, %hi(.LCPI13_2)
+; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI13_2)(a2)
 ; RV32IZFH-NEXT:    flt.s a2, ft0, fs0
-; RV32IZFH-NEXT:    beqz a2, .LBB13_4
-; RV32IZFH-NEXT:  # %bb.3:
+; RV32IZFH-NEXT:    beqz a2, .LBB13_6
+; RV32IZFH-NEXT:  # %bb.5:
 ; RV32IZFH-NEXT:    addi a1, a3, -1
-; RV32IZFH-NEXT:  .LBB13_4:
+; RV32IZFH-NEXT:  .LBB13_6:
 ; RV32IZFH-NEXT:    feq.s a3, fs0, fs0
 ; RV32IZFH-NEXT:    seqz a3, a3
 ; RV32IZFH-NEXT:    addi a3, a3, -1
 ; RV32IZFH-NEXT:    and a1, a3, a1
-; RV32IZFH-NEXT:    neg a2, a2
 ; RV32IZFH-NEXT:    neg a4, s0
 ; RV32IZFH-NEXT:    and a0, a4, a0
+; RV32IZFH-NEXT:    neg a2, a2
 ; RV32IZFH-NEXT:    or a0, a2, a0
 ; RV32IZFH-NEXT:    and a0, a3, a0
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;

 }
 
 define i64 @test_round_ui64(half %x) nounwind {
 ; RV32IZFH-LABEL: test_round_ui64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    sw s1, 4(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call roundf@plt
+; RV32IZFH-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    lui a0, %hi(.LCPI15_0)
-; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI15_0)(a0)
-; RV32IZFH-NEXT:    fcvt.h.s ft1, fa0
-; RV32IZFH-NEXT:    fcvt.s.h fa0, ft1
-; RV32IZFH-NEXT:    flt.s a0, ft0, fa0
-; RV32IZFH-NEXT:    neg s0, a0
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI15_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB15_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rmm
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rmm
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB15_2:
+; RV32IZFH-NEXT:    fcvt.s.h fs0, fa0
 ; RV32IZFH-NEXT:    fmv.w.x ft0, zero
-; RV32IZFH-NEXT:    fle.s a0, ft0, fa0
-; RV32IZFH-NEXT:    neg s1, a0
+; RV32IZFH-NEXT:    fle.s a0, ft0, fs0
+; RV32IZFH-NEXT:    neg s0, a0
+; RV32IZFH-NEXT:    fmv.s fa0, fs0
 ; RV32IZFH-NEXT:    call __fixunssfdi@plt
-; RV32IZFH-NEXT:    and a0, s1, a0
-; RV32IZFH-NEXT:    or a0, s0, a0
-; RV32IZFH-NEXT:    and a1, s1, a1
-; RV32IZFH-NEXT:    or a1, s0, a1
+; RV32IZFH-NEXT:    lui a2, %hi(.LCPI15_1)
+; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI15_1)(a2)
+; RV32IZFH-NEXT:    and a0, s0, a0
+; RV32IZFH-NEXT:    flt.s a2, ft0, fs0
+; RV32IZFH-NEXT:    neg a2, a2
+; RV32IZFH-NEXT:    or a0, a2, a0
+; RV32IZFH-NEXT:    and a1, s0, a1
+; RV32IZFH-NEXT:    or a1, a2, a1
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    lw s1, 4(sp) # 4-byte Folded Reload
+; RV32IZFH-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;
 ; RV64IZFH-LABEL: test_round_ui64:
 ; RV64IZFH:       # %bb.0:
 ; RV64IZFH-NEXT:    fcvt.lu.h a0, fa0, rmm
 ; RV64IZFH-NEXT:    feq.h a1, fa0, fa0
 ; RV64IZFH-NEXT:    seqz a1, a1

 
 define i64 @test_roundeven_si64(half %x) nounwind {
 ; RV32IZFH-LABEL: test_roundeven_si64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call roundevenf@plt
 ; RV32IZFH-NEXT:    lui a0, %hi(.LCPI17_0)
-; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI17_0)(a0)
-; RV32IZFH-NEXT:    fcvt.h.s ft1, fa0
-; RV32IZFH-NEXT:    fcvt.s.h fs0, ft1
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI17_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB17_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rne
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rne
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB17_2:
+; RV32IZFH-NEXT:    lui a0, %hi(.LCPI17_1)
+; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI17_1)(a0)
+; RV32IZFH-NEXT:    fcvt.s.h fs0, fa0
 ; RV32IZFH-NEXT:    fle.s s0, ft0, fs0
 ; RV32IZFH-NEXT:    fmv.s fa0, fs0
 ; RV32IZFH-NEXT:    call __fixsfdi@plt
 ; RV32IZFH-NEXT:    lui a3, 524288
-; RV32IZFH-NEXT:    bnez s0, .LBB17_2
-; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    bnez s0, .LBB17_4
+; RV32IZFH-NEXT:  # %bb.3:
 ; RV32IZFH-NEXT:    lui a1, 524288
-; RV32IZFH-NEXT:  .LBB17_2:
-; RV32IZFH-NEXT:    lui a2, %hi(.LCPI17_1)
-; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI17_1)(a2)
+; RV32IZFH-NEXT:  .LBB17_4:
+; RV32IZFH-NEXT:    lui a2, %hi(.LCPI17_2)
+; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI17_2)(a2)
 ; RV32IZFH-NEXT:    flt.s a2, ft0, fs0
-; RV32IZFH-NEXT:    beqz a2, .LBB17_4
-; RV32IZFH-NEXT:  # %bb.3:
+; RV32IZFH-NEXT:    beqz a2, .LBB17_6
+; RV32IZFH-NEXT:  # %bb.5:
 ; RV32IZFH-NEXT:    addi a1, a3, -1
-; RV32IZFH-NEXT:  .LBB17_4:
+; RV32IZFH-NEXT:  .LBB17_6:
 ; RV32IZFH-NEXT:    feq.s a3, fs0, fs0
 ; RV32IZFH-NEXT:    seqz a3, a3
 ; RV32IZFH-NEXT:    addi a3, a3, -1
 ; RV32IZFH-NEXT:    and a1, a3, a1
-; RV32IZFH-NEXT:    neg a2, a2
 ; RV32IZFH-NEXT:    neg a4, s0
 ; RV32IZFH-NEXT:    and a0, a4, a0
+; RV32IZFH-NEXT:    neg a2, a2
 ; RV32IZFH-NEXT:    or a0, a2, a0
 ; RV32IZFH-NEXT:    and a0, a3, a0
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;

 }
 
 define i64 @test_roundeven_ui64(half %x) nounwind {
 ; RV32IZFH-LABEL: test_roundeven_ui64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    sw s1, 4(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call roundevenf@plt
+; RV32IZFH-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    lui a0, %hi(.LCPI19_0)
-; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI19_0)(a0)
-; RV32IZFH-NEXT:    fcvt.h.s ft1, fa0
-; RV32IZFH-NEXT:    fcvt.s.h fa0, ft1
-; RV32IZFH-NEXT:    flt.s a0, ft0, fa0
-; RV32IZFH-NEXT:    neg s0, a0
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI19_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB19_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rne
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rne
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB19_2:
+; RV32IZFH-NEXT:    fcvt.s.h fs0, fa0
 ; RV32IZFH-NEXT:    fmv.w.x ft0, zero
-; RV32IZFH-NEXT:    fle.s a0, ft0, fa0
-; RV32IZFH-NEXT:    neg s1, a0
+; RV32IZFH-NEXT:    fle.s a0, ft0, fs0
+; RV32IZFH-NEXT:    neg s0, a0
+; RV32IZFH-NEXT:    fmv.s fa0, fs0
 ; RV32IZFH-NEXT:    call __fixunssfdi@plt
-; RV32IZFH-NEXT:    and a0, s1, a0
-; RV32IZFH-NEXT:    or a0, s0, a0
-; RV32IZFH-NEXT:    and a1, s1, a1
-; RV32IZFH-NEXT:    or a1, s0, a1
+; RV32IZFH-NEXT:    lui a2, %hi(.LCPI19_1)
+; RV32IZFH-NEXT:    flw ft0, %lo(.LCPI19_1)(a2)
+; RV32IZFH-NEXT:    and a0, s0, a0
+; RV32IZFH-NEXT:    flt.s a2, ft0, fs0
+; RV32IZFH-NEXT:    neg a2, a2
+; RV32IZFH-NEXT:    or a0, a2, a0
+; RV32IZFH-NEXT:    and a1, s0, a1
+; RV32IZFH-NEXT:    or a1, a2, a1
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    lw s1, 4(sp) # 4-byte Folded Reload
+; RV32IZFH-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;
 ; RV64IZFH-LABEL: test_roundeven_ui64:
 ; RV64IZFH:       # %bb.0:
 ; RV64IZFH-NEXT:    fcvt.lu.h a0, fa0, rne
 ; RV64IZFH-NEXT:    feq.h a1, fa0, fa0
 ; RV64IZFH-NEXT:    seqz a1, a1

llvm/test/CodeGen/RISCV/half-round-conv.ll

 
 define i64 @test_floor_si64(half %x) {
 ; RV32IZFH-LABEL: test_floor_si64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    .cfi_offset ra, -4
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call floorf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
+; RV32IZFH-NEXT:    lui a0, %hi(.LCPI3_0)
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI3_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB3_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rdn
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rdn
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB3_2:
 ; RV32IZFH-NEXT:    call __fixhfdi@plt
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;
 ; RV64IZFH-LABEL: test_floor_si64:
 ; RV64IZFH:       # %bb.0:
 ; RV64IZFH-NEXT:    fcvt.l.h a0, fa0, rdn

 
 define i64 @test_floor_ui64(half %x) {
 ; RV32IZFH-LABEL: test_floor_ui64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    .cfi_offset ra, -4
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call floorf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
+; RV32IZFH-NEXT:    lui a0, %hi(.LCPI7_0)
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI7_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB7_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rdn
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rdn
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB7_2:
 ; RV32IZFH-NEXT:    call __fixunshfdi@plt
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;
 ; RV64IZFH-LABEL: test_floor_ui64:
 ; RV64IZFH:       # %bb.0:
 ; RV64IZFH-NEXT:    fcvt.lu.h a0, fa0, rdn

 
 define i64 @test_ceil_si64(half %x) {
 ; RV32IZFH-LABEL: test_ceil_si64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    .cfi_offset ra, -4
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call ceilf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
+; RV32IZFH-NEXT:    lui a0, %hi(.LCPI11_0)
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI11_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB11_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rup
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rup
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB11_2:
 ; RV32IZFH-NEXT:    call __fixhfdi@plt
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;
 ; RV64IZFH-LABEL: test_ceil_si64:
 ; RV64IZFH:       # %bb.0:
 ; RV64IZFH-NEXT:    fcvt.l.h a0, fa0, rup

 
 define i64 @test_ceil_ui64(half %x) {
 ; RV32IZFH-LABEL: test_ceil_ui64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    .cfi_offset ra, -4
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call ceilf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
+; RV32IZFH-NEXT:    lui a0, %hi(.LCPI15_0)
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI15_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB15_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rup
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rup
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB15_2:
 ; RV32IZFH-NEXT:    call __fixunshfdi@plt
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;
 ; RV64IZFH-LABEL: test_ceil_ui64:
 ; RV64IZFH:       # %bb.0:
 ; RV64IZFH-NEXT:    fcvt.lu.h a0, fa0, rup

 
 define i64 @test_trunc_si64(half %x) {
 ; RV32IZFH-LABEL: test_trunc_si64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    .cfi_offset ra, -4
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call truncf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
+; RV32IZFH-NEXT:    lui a0, %hi(.LCPI19_0)
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI19_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB19_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rtz
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rtz
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB19_2:
 ; RV32IZFH-NEXT:    call __fixhfdi@plt
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;
 ; RV64IZFH-LABEL: test_trunc_si64:
 ; RV64IZFH:       # %bb.0:
 ; RV64IZFH-NEXT:    fcvt.l.h a0, fa0, rtz

 
 define i64 @test_trunc_ui64(half %x) {
 ; RV32IZFH-LABEL: test_trunc_ui64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    .cfi_offset ra, -4
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call truncf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
+; RV32IZFH-NEXT:    lui a0, %hi(.LCPI23_0)
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI23_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB23_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rtz
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rtz
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB23_2:
 ; RV32IZFH-NEXT:    call __fixunshfdi@plt
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;
 ; RV64IZFH-LABEL: test_trunc_ui64:
 ; RV64IZFH:       # %bb.0:
 ; RV64IZFH-NEXT:    fcvt.lu.h a0, fa0, rtz

 
 define i64 @test_round_si64(half %x) {
 ; RV32IZFH-LABEL: test_round_si64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    .cfi_offset ra, -4
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call roundf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
+; RV32IZFH-NEXT:    lui a0, %hi(.LCPI27_0)
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI27_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB27_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rmm
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rmm
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB27_2:
 ; RV32IZFH-NEXT:    call __fixhfdi@plt
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;
 ; RV64IZFH-LABEL: test_round_si64:
 ; RV64IZFH:       # %bb.0:
 ; RV64IZFH-NEXT:    fcvt.l.h a0, fa0, rmm

 
 define i64 @test_round_ui64(half %x) {
 ; RV32IZFH-LABEL: test_round_ui64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    .cfi_offset ra, -4
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call roundf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
+; RV32IZFH-NEXT:    lui a0, %hi(.LCPI31_0)
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI31_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB31_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rmm
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rmm
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB31_2:
 ; RV32IZFH-NEXT:    call __fixunshfdi@plt
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;
 ; RV64IZFH-LABEL: test_round_ui64:
 ; RV64IZFH:       # %bb.0:
 ; RV64IZFH-NEXT:    fcvt.lu.h a0, fa0, rmm

 
 define i64 @test_roundeven_si64(half %x) {
 ; RV32IZFH-LABEL: test_roundeven_si64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    .cfi_offset ra, -4
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call roundevenf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
+; RV32IZFH-NEXT:    lui a0, %hi(.LCPI35_0)
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI35_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB35_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rne
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rne
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB35_2:
 ; RV32IZFH-NEXT:    call __fixhfdi@plt
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;
 ; RV64IZFH-LABEL: test_roundeven_si64:
 ; RV64IZFH:       # %bb.0:
 ; RV64IZFH-NEXT:    fcvt.l.h a0, fa0, rne

 
 define i64 @test_roundeven_ui64(half %x) {
 ; RV32IZFH-LABEL: test_roundeven_ui64:
 ; RV32IZFH:       # %bb.0:
 ; RV32IZFH-NEXT:    addi sp, sp, -16
 ; RV32IZFH-NEXT:    .cfi_def_cfa_offset 16
 ; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32IZFH-NEXT:    .cfi_offset ra, -4
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call roundevenf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
+; RV32IZFH-NEXT:    lui a0, %hi(.LCPI39_0)
+; RV32IZFH-NEXT:    flh ft0, %lo(.LCPI39_0)(a0)
+; RV32IZFH-NEXT:    fabs.h ft1, fa0
+; RV32IZFH-NEXT:    flt.h a0, ft1, ft0
+; RV32IZFH-NEXT:    beqz a0, .LBB39_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rne
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0, rne
+; RV32IZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; RV32IZFH-NEXT:  .LBB39_2:
 ; RV32IZFH-NEXT:    call __fixunshfdi@plt
 ; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IZFH-NEXT:    addi sp, sp, 16
 ; RV32IZFH-NEXT:    ret
 ;
 ; RV64IZFH-LABEL: test_roundeven_ui64:
 ; RV64IZFH:       # %bb.0:
 ; RV64IZFH-NEXT:    fcvt.lu.h a0, fa0, rne

 ; RV64IFD-NEXT:    addi sp, sp, -16
 ; RV64IFD-NEXT:    .cfi_def_cfa_offset 16
 ; RV64IFD-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64IFD-NEXT:    .cfi_offset ra, -8
 ; RV64IFD-NEXT:    call floor@plt
 ; RV64IFD-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64IFD-NEXT:    addi sp, sp, 16
 ; RV64IFD-NEXT:    ret
-; RV32IZFH-LABEL: test_floor_half:
-; RV32IZFH:       # %bb.0:
-; RV32IZFH-NEXT:    addi sp, sp, -16
-; RV32IZFH-NEXT:    .cfi_def_cfa_offset 16
-; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    .cfi_offset ra, -4
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call floorf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    addi sp, sp, 16
-; RV32IZFH-NEXT:    ret
-;
-; RV64IZFH-LABEL: test_floor_half:
-; RV64IZFH:       # %bb.0:
-; RV64IZFH-NEXT:    addi sp, sp, -16
-; RV64IZFH-NEXT:    .cfi_def_cfa_offset 16
-; RV64IZFH-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
-; RV64IZFH-NEXT:    .cfi_offset ra, -8
-; RV64IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV64IZFH-NEXT:    call floorf@plt
-; RV64IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV64IZFH-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
-; RV64IZFH-NEXT:    addi sp, sp, 16
-; RV64IZFH-NEXT:    ret
+; CHECKIZFH-LABEL: test_floor_half:
+; CHECKIZFH:       # %bb.0:
+; CHECKIZFH-NEXT:    lui a0, %hi(.LCPI40_0)
+; CHECKIZFH-NEXT:    flh ft0, %lo(.LCPI40_0)(a0)
+; CHECKIZFH-NEXT:    fabs.h ft1, fa0
+; CHECKIZFH-NEXT:    flt.h a0, ft1, ft0
+; CHECKIZFH-NEXT:    beqz a0, .LBB40_2
+; CHECKIZFH-NEXT:  # %bb.1:
+; CHECKIZFH-NEXT:    fcvt.w.h a0, fa0, rdn
+; CHECKIZFH-NEXT:    fcvt.h.w ft0, a0, rdn
+; CHECKIZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; CHECKIZFH-NEXT:  .LBB40_2:
+; CHECKIZFH-NEXT:    ret
   %a = call half @llvm.floor.f16(half %x)
   ret half %a
 }
 
 define half @test_ceil_half(half %x) {
 ; RV32IFD-LABEL: test_ceil_half:
 ; RV32IFD:       # %bb.0:
 ; RV32IFD-NEXT:    addi sp, sp, -16

 ; RV64IFD-NEXT:    addi sp, sp, -16
 ; RV64IFD-NEXT:    .cfi_def_cfa_offset 16
 ; RV64IFD-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64IFD-NEXT:    .cfi_offset ra, -8
 ; RV64IFD-NEXT:    call ceil@plt
 ; RV64IFD-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64IFD-NEXT:    addi sp, sp, 16
 ; RV64IFD-NEXT:    ret
-; RV32IZFH-LABEL: test_ceil_half:
-; RV32IZFH:       # %bb.0:
-; RV32IZFH-NEXT:    addi sp, sp, -16
-; RV32IZFH-NEXT:    .cfi_def_cfa_offset 16
-; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    .cfi_offset ra, -4
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call ceilf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    addi sp, sp, 16
-; RV32IZFH-NEXT:    ret
-;
-; RV64IZFH-LABEL: test_ceil_half:
-; RV64IZFH:       # %bb.0:
-; RV64IZFH-NEXT:    addi sp, sp, -16
-; RV64IZFH-NEXT:    .cfi_def_cfa_offset 16
-; RV64IZFH-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
-; RV64IZFH-NEXT:    .cfi_offset ra, -8
-; RV64IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV64IZFH-NEXT:    call ceilf@plt
-; RV64IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV64IZFH-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
-; RV64IZFH-NEXT:    addi sp, sp, 16
-; RV64IZFH-NEXT:    ret
+; CHECKIZFH-LABEL: test_ceil_half:
+; CHECKIZFH:       # %bb.0:
+; CHECKIZFH-NEXT:    lui a0, %hi(.LCPI41_0)
+; CHECKIZFH-NEXT:    flh ft0, %lo(.LCPI41_0)(a0)
+; CHECKIZFH-NEXT:    fabs.h ft1, fa0
+; CHECKIZFH-NEXT:    flt.h a0, ft1, ft0
+; CHECKIZFH-NEXT:    beqz a0, .LBB41_2
+; CHECKIZFH-NEXT:  # %bb.1:
+; CHECKIZFH-NEXT:    fcvt.w.h a0, fa0, rup
+; CHECKIZFH-NEXT:    fcvt.h.w ft0, a0, rup
+; CHECKIZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; CHECKIZFH-NEXT:  .LBB41_2:
+; CHECKIZFH-NEXT:    ret
   %a = call half @llvm.ceil.f16(half %x)
   ret half %a
 }
 
 define half @test_trunc_half(half %x) {
 ; RV32IFD-LABEL: test_trunc_half:
 ; RV32IFD:       # %bb.0:
 ; RV32IFD-NEXT:    addi sp, sp, -16

 ; RV64IFD-NEXT:    addi sp, sp, -16
 ; RV64IFD-NEXT:    .cfi_def_cfa_offset 16
 ; RV64IFD-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64IFD-NEXT:    .cfi_offset ra, -8
 ; RV64IFD-NEXT:    call trunc@plt
 ; RV64IFD-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64IFD-NEXT:    addi sp, sp, 16
 ; RV64IFD-NEXT:    ret
-; RV32IZFH-LABEL: test_trunc_half:
-; RV32IZFH:       # %bb.0:
-; RV32IZFH-NEXT:    addi sp, sp, -16
-; RV32IZFH-NEXT:    .cfi_def_cfa_offset 16
-; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    .cfi_offset ra, -4
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call truncf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    addi sp, sp, 16
-; RV32IZFH-NEXT:    ret
-;
-; RV64IZFH-LABEL: test_trunc_half:
-; RV64IZFH:       # %bb.0:
-; RV64IZFH-NEXT:    addi sp, sp, -16
-; RV64IZFH-NEXT:    .cfi_def_cfa_offset 16
-; RV64IZFH-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
-; RV64IZFH-NEXT:    .cfi_offset ra, -8
-; RV64IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV64IZFH-NEXT:    call truncf@plt
-; RV64IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV64IZFH-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
-; RV64IZFH-NEXT:    addi sp, sp, 16
-; RV64IZFH-NEXT:    ret
+; CHECKIZFH-LABEL: test_trunc_half:
+; CHECKIZFH:       # %bb.0:
+; CHECKIZFH-NEXT:    lui a0, %hi(.LCPI42_0)
+; CHECKIZFH-NEXT:    flh ft0, %lo(.LCPI42_0)(a0)
+; CHECKIZFH-NEXT:    fabs.h ft1, fa0
+; CHECKIZFH-NEXT:    flt.h a0, ft1, ft0
+; CHECKIZFH-NEXT:    beqz a0, .LBB42_2
+; CHECKIZFH-NEXT:  # %bb.1:
+; CHECKIZFH-NEXT:    fcvt.w.h a0, fa0, rtz
+; CHECKIZFH-NEXT:    fcvt.h.w ft0, a0, rtz
+; CHECKIZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; CHECKIZFH-NEXT:  .LBB42_2:
+; CHECKIZFH-NEXT:    ret
   %a = call half @llvm.trunc.f16(half %x)
   ret half %a
 }
 
 define half @test_round_half(half %x) {
 ; RV32IFD-LABEL: test_round_half:
 ; RV32IFD:       # %bb.0:
 ; RV32IFD-NEXT:    addi sp, sp, -16

 ; RV64IFD-NEXT:    addi sp, sp, -16
 ; RV64IFD-NEXT:    .cfi_def_cfa_offset 16
 ; RV64IFD-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64IFD-NEXT:    .cfi_offset ra, -8
 ; RV64IFD-NEXT:    call round@plt
 ; RV64IFD-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64IFD-NEXT:    addi sp, sp, 16
 ; RV64IFD-NEXT:    ret
-; RV32IZFH-LABEL: test_round_half:
-; RV32IZFH:       # %bb.0:
-; RV32IZFH-NEXT:    addi sp, sp, -16
-; RV32IZFH-NEXT:    .cfi_def_cfa_offset 16
-; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    .cfi_offset ra, -4
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call roundf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    addi sp, sp, 16
-; RV32IZFH-NEXT:    ret
-;
-; RV64IZFH-LABEL: test_round_half:
-; RV64IZFH:       # %bb.0:
-; RV64IZFH-NEXT:    addi sp, sp, -16
-; RV64IZFH-NEXT:    .cfi_def_cfa_offset 16
-; RV64IZFH-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
-; RV64IZFH-NEXT:    .cfi_offset ra, -8
-; RV64IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV64IZFH-NEXT:    call roundf@plt
-; RV64IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV64IZFH-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
-; RV64IZFH-NEXT:    addi sp, sp, 16
-; RV64IZFH-NEXT:    ret
+; CHECKIZFH-LABEL: test_round_half:
+; CHECKIZFH:       # %bb.0:
+; CHECKIZFH-NEXT:    lui a0, %hi(.LCPI43_0)
+; CHECKIZFH-NEXT:    flh ft0, %lo(.LCPI43_0)(a0)
+; CHECKIZFH-NEXT:    fabs.h ft1, fa0
+; CHECKIZFH-NEXT:    flt.h a0, ft1, ft0
+; CHECKIZFH-NEXT:    beqz a0, .LBB43_2
+; CHECKIZFH-NEXT:  # %bb.1:
+; CHECKIZFH-NEXT:    fcvt.w.h a0, fa0, rmm
+; CHECKIZFH-NEXT:    fcvt.h.w ft0, a0, rmm
+; CHECKIZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; CHECKIZFH-NEXT:  .LBB43_2:
+; CHECKIZFH-NEXT:    ret
   %a = call half @llvm.round.f16(half %x)
   ret half %a
 }
 
 define half @test_roundeven_half(half %x) {
 ; RV32IFD-LABEL: test_roundeven_half:
 ; RV32IFD:       # %bb.0:
 ; RV32IFD-NEXT:    addi sp, sp, -16

 ; RV64IFD-NEXT:    addi sp, sp, -16
 ; RV64IFD-NEXT:    .cfi_def_cfa_offset 16
 ; RV64IFD-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64IFD-NEXT:    .cfi_offset ra, -8
 ; RV64IFD-NEXT:    call roundeven@plt
 ; RV64IFD-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64IFD-NEXT:    addi sp, sp, 16
 ; RV64IFD-NEXT:    ret
-; RV32IZFH-LABEL: test_roundeven_half:
-; RV32IZFH:       # %bb.0:
-; RV32IZFH-NEXT:    addi sp, sp, -16
-; RV32IZFH-NEXT:    .cfi_def_cfa_offset 16
-; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
-; RV32IZFH-NEXT:    .cfi_offset ra, -4
-; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV32IZFH-NEXT:    call roundevenf@plt
-; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
-; RV32IZFH-NEXT:    addi sp, sp, 16
-; RV32IZFH-NEXT:    ret
-;
-; RV64IZFH-LABEL: test_roundeven_half:
-; RV64IZFH:       # %bb.0:
-; RV64IZFH-NEXT:    addi sp, sp, -16
-; RV64IZFH-NEXT:    .cfi_def_cfa_offset 16
-; RV64IZFH-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
-; RV64IZFH-NEXT:    .cfi_offset ra, -8
-; RV64IZFH-NEXT:    fcvt.s.h fa0, fa0
-; RV64IZFH-NEXT:    call roundevenf@plt
-; RV64IZFH-NEXT:    fcvt.h.s fa0, fa0
-; RV64IZFH-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
-; RV64IZFH-NEXT:    addi sp, sp, 16
-; RV64IZFH-NEXT:    ret
+; CHECKIZFH-LABEL: test_roundeven_half:
+; CHECKIZFH:       # %bb.0:
+; CHECKIZFH-NEXT:    lui a0, %hi(.LCPI44_0)
+; CHECKIZFH-NEXT:    flh ft0, %lo(.LCPI44_0)(a0)
+; CHECKIZFH-NEXT:    fabs.h ft1, fa0
+; CHECKIZFH-NEXT:    flt.h a0, ft1, ft0
+; CHECKIZFH-NEXT:    beqz a0, .LBB44_2
+; CHECKIZFH-NEXT:  # %bb.1:
+; CHECKIZFH-NEXT:    fcvt.w.h a0, fa0, rne
+; CHECKIZFH-NEXT:    fcvt.h.w ft0, a0, rne
+; CHECKIZFH-NEXT:    fsgnj.h fa0, ft0, fa0
+; CHECKIZFH-NEXT:  .LBB44_2:
+; CHECKIZFH-NEXT:    ret
   %a = call half @llvm.roundeven.f16(half %x)
   ret half %a
 }
 
 declare half @llvm.floor.f16(half)
 declare half @llvm.ceil.f16(half)
 declare half @llvm.trunc.f16(half)
 declare half @llvm.round.f16(half)
 declare half @llvm.roundeven.f16(half)