[RISCV] Support strict FP conversion operations.

[craig.topper]

Summary

This adds support for strict conversions between fp types and between
integer and fp.

[craig.topper]

Remove some stale test comments copied from non-strict tests.

[luismarques]

Is it out of scope of this patch to check the rounding mode metadata argument? I've checked that changing the r.m. in a test case of e.g. fptrunc doesn't do anything, it just keeps emitting an fcvt with dynamic rounding. It's probably a good idea to add a short note describing the status/limitations of that to the tests / code / commit message, as appropriate.

Nit: is there any reason for not using a hard-float ABI in the tests? It seems like we could cut down on the loads/conversion noise in the tests. This probably also applies to D115680.

[craig.topper]

In D115997#3206395, @luismarques wrote:

Is it out of scope of this patch to check the rounding mode metadata argument? I've checked that changing the r.m. in a test case of e.g. fptrunc doesn't do anything, it just keeps emitting an fcvt with dynamic rounding. It's probably a good idea to add a short note describing the status/limitations of that to the tests / code / commit message, as appropriate.

Rounding mode doesn't make it through SelectionDAGBuilder so isel can't use it. I think the metadata is intended to convey the compiler's knowledge about the most recent call to fesetround for constant folding purposes. (Though I don't think that information is captured at all today.) It is not intended to select static rounding mode. The dynamic mode was already updated so using a static instruction wouldn't do much unless the compiler was able to optimize out the dynamic change if all operations were able to use static rounding mode instruction.

I can add comments for this.

Nit: is there any reason for not using a hard-float ABI in the tests? It seems like we could cut down on the loads/conversion noise in the tests. This probably also applies to D115680.

Command lines and test case names came from the equivalent test files without "-strict". I only added the -disable-strictnode-mutation to make sure isel would fail if STRICT nodes were not handled. I diffed the files during patch development to make sure the same code was produced for the equivalent test without constrained intrinsics.

[luismarques]

In D115997#3206508, @craig.topper wrote:

Command lines and test case names came from the equivalent test files without "-strict". I only added the -disable-strictnode-mutation to make sure isel would fail if STRICT nodes were not handled.

At first glance it looks like the non-strict files would also benefit from that ABI tweak, so both could be changed, but it's not very important.

I diffed the files during patch development to make sure the same code was produced for the equivalent test without constrained intrinsic.

I wish we could keep that testable with FileCheck but I suppose there's no good way to do that... :/

[luismarques]

LGTM.

llvm/lib/Target/RISCV/RISCVInstrInfoF.td

22–25

I missed these in D106346 :)

llvm/lib/Target/RISCV/RISCVISelLowering.h

 //===-- RISCVISelLowering.h - RISCV DAG Lowering Interface ------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
enum NodeType : unsigned {
   // chain pointer.
   WRITE_CSR,
   // Read and write value of CSR.
   // The first operand is a chain pointer, the second specifies address of the
   // required CSR and the third is the value to write. Two results are produced,
   // the value read before the modification and the new chain pointer.
   SWAP_CSR,
 
+  // 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.
+  STRICT_FCVT_W_RTZ_RV64 = ISD::FIRST_TARGET_STRICTFP_OPCODE,
+  STRICT_FCVT_WU_RTZ_RV64,
+
   // Memory opcodes start here.
   VLE_VL = ISD::FIRST_TARGET_MEMORY_OPCODE,
   VSE_VL,
 
   // WARNING: Do not add anything in the end unless you want the node to
   // have memop! In fact, starting from FIRST_TARGET_MEMORY_OPCODE all
   // opcodes will be thought as target memory ops!
 };

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

 //===-- RISCVISelLowering.cpp - RISCV DAG Lowering Implementation  --------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
if (Subtarget.hasStdExtZfh()) {
     setOperationAction(ISD::LLRINT, MVT::f16, Legal);
     setOperationAction(ISD::LROUND, MVT::f16, Legal);
     setOperationAction(ISD::LLROUND, MVT::f16, Legal);
     setOperationAction(ISD::STRICT_FADD, MVT::f16, Legal);
     setOperationAction(ISD::STRICT_FMA, MVT::f16, Legal);
     setOperationAction(ISD::STRICT_FSUB, MVT::f16, Legal);
     setOperationAction(ISD::STRICT_FMUL, MVT::f16, Legal);
     setOperationAction(ISD::STRICT_FDIV, MVT::f16, Legal);
+    setOperationAction(ISD::STRICT_FP_ROUND, MVT::f16, Legal);
+    setOperationAction(ISD::STRICT_FP_EXTEND, MVT::f32, Legal);
     setOperationAction(ISD::STRICT_FSQRT, MVT::f16, Legal);
     for (auto CC : FPCCToExpand)
       setCondCodeAction(CC, 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,       MVT::f16, Promote);
if (Subtarget.hasStdExtD()) {
     setOperationAction(ISD::LLRINT, MVT::f64, Legal);
     setOperationAction(ISD::LROUND, MVT::f64, Legal);
     setOperationAction(ISD::LLROUND, MVT::f64, Legal);
     setOperationAction(ISD::STRICT_FMA, MVT::f64, Legal);
     setOperationAction(ISD::STRICT_FADD, MVT::f64, Legal);
     setOperationAction(ISD::STRICT_FSUB, MVT::f64, Legal);
     setOperationAction(ISD::STRICT_FMUL, MVT::f64, Legal);
     setOperationAction(ISD::STRICT_FDIV, MVT::f64, Legal);
+    setOperationAction(ISD::STRICT_FP_ROUND, MVT::f32, Legal);
+    setOperationAction(ISD::STRICT_FP_EXTEND, MVT::f64, Legal);
     setOperationAction(ISD::STRICT_FSQRT, MVT::f64, Legal);
     for (auto CC : FPCCToExpand)
       setCondCodeAction(CC, 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);
if (Subtarget.is64Bit()) {
     setOperationAction(ISD::STRICT_FP_TO_UINT, MVT::i32, Custom);
     setOperationAction(ISD::STRICT_FP_TO_SINT, MVT::i32, Custom);
   }
 
   if (Subtarget.hasStdExtF()) {
     setOperationAction(ISD::FP_TO_UINT_SAT, XLenVT, Custom);
     setOperationAction(ISD::FP_TO_SINT_SAT, XLenVT, Custom);
 
+    setOperationAction(ISD::STRICT_FP_TO_UINT, XLenVT, Legal);
+    setOperationAction(ISD::STRICT_FP_TO_SINT, XLenVT, Legal);
+    setOperationAction(ISD::STRICT_UINT_TO_FP, XLenVT, Legal);
+    setOperationAction(ISD::STRICT_SINT_TO_FP, XLenVT, Legal);
+
     setOperationAction(ISD::FLT_ROUNDS_, XLenVT, Custom);
     setOperationAction(ISD::SET_ROUNDING, MVT::Other, Custom);
   }
 
   setOperationAction(ISD::GlobalAddress, XLenVT, Custom);
   setOperationAction(ISD::BlockAddress, XLenVT, Custom);
   setOperationAction(ISD::ConstantPool, XLenVT, Custom);
   setOperationAction(ISD::JumpTable, XLenVT, Custom);
case ISD::FP_TO_UINT: {
     assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
            "Unexpected custom legalisation");
     bool IsStrict = N->isStrictFPOpcode();
     bool IsSigned = N->getOpcode() == ISD::FP_TO_SINT ||
                     N->getOpcode() == ISD::STRICT_FP_TO_SINT;
     SDValue Op0 = IsStrict ? N->getOperand(1) : N->getOperand(0);
     if (getTypeAction(*DAG.getContext(), Op0.getValueType()) !=
         TargetLowering::TypeSoftenFloat) {
-      // FIXME: Support strict FP.
-      if (IsStrict)
-        return;
       if (!isTypeLegal(Op0.getValueType()))
         return;
+      if (IsStrict) {
+        unsigned Opc = IsSigned ? RISCVISD::STRICT_FCVT_W_RTZ_RV64
+                                : RISCVISD::STRICT_FCVT_WU_RTZ_RV64;
+        SDVTList VTs = DAG.getVTList(MVT::i64, MVT::Other);
+        SDValue Res = DAG.getNode(Opc, DL, VTs, N->getOperand(0), Op0);
+        Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Res));
+        Results.push_back(Res.getValue(1));
+        return;
+      }
       unsigned Opc =
           IsSigned ? RISCVISD::FCVT_W_RTZ_RV64 : RISCVISD::FCVT_WU_RTZ_RV64;
       SDValue Res = DAG.getNode(Opc, DL, MVT::i64, Op0);
       Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Res));
       return;
     }
     // If the FP type needs to be softened, emit a library call using the 'si'
     // version. If we left it to default legalization we'd end up with 'di'. If
SDValue RISCVTargetLowering::PerformDAGCombine(SDNode *N,
   case ISD::XOR:
     return performXORCombine(N, DAG);
   case ISD::ANY_EXTEND:
     return performANY_EXTENDCombine(N, DCI, Subtarget);
   case ISD::ZERO_EXTEND:
     // Fold (zero_extend (fp_to_uint X)) to prevent forming fcvt+zexti32 during
     // type legalization. This is safe because fp_to_uint produces poison if
     // it overflows.
-    if (N->getValueType(0) == MVT::i64 && Subtarget.is64Bit() &&
-        N->getOperand(0).getOpcode() == ISD::FP_TO_UINT &&
-        isTypeLegal(N->getOperand(0).getOperand(0).getValueType()))
+    if (N->getValueType(0) == MVT::i64 && Subtarget.is64Bit()) {
+      SDValue Src = N->getOperand(0);
+      if (Src.getOpcode() == ISD::FP_TO_UINT &&
+          isTypeLegal(Src.getOperand(0).getValueType()))
         return DAG.getNode(ISD::FP_TO_UINT, SDLoc(N), MVT::i64,
-                         N->getOperand(0).getOperand(0));
+                           Src.getOperand(0));
+      if (Src.getOpcode() == ISD::STRICT_FP_TO_UINT && Src.hasOneUse() &&
+          isTypeLegal(Src.getOperand(1).getValueType())) {
+        SDVTList VTs = DAG.getVTList(MVT::i64, MVT::Other);
+        SDValue Res = DAG.getNode(ISD::STRICT_FP_TO_UINT, SDLoc(N), VTs,
+                                  Src.getOperand(0), Src.getOperand(1));
+        DCI.CombineTo(N, Res);
+        DAG.ReplaceAllUsesOfValueWith(Src.getValue(1), Res.getValue(1));
+        DCI.recursivelyDeleteUnusedNodes(Src.getNode());
+        return SDValue(N, 0); // Return N so it doesn't get rechecked.
+      }
+    }
     return SDValue();
   case RISCVISD::SELECT_CC: {
     // Transform
     SDValue LHS = N->getOperand(0);
     SDValue RHS = N->getOperand(1);
     SDValue TrueV = N->getOperand(3);
     SDValue FalseV = N->getOperand(4);
 
unsigned RISCVTargetLowering::ComputeNumSignBitsForTargetNode(
   case RISCVISD::FSLW:
   case RISCVISD::FSRW:
   case RISCVISD::SHFLW:
   case RISCVISD::UNSHFLW:
   case RISCVISD::BCOMPRESSW:
   case RISCVISD::BDECOMPRESSW:
   case RISCVISD::FCVT_W_RTZ_RV64:
   case RISCVISD::FCVT_WU_RTZ_RV64:
+  case RISCVISD::STRICT_FCVT_W_RTZ_RV64:
+  case RISCVISD::STRICT_FCVT_WU_RTZ_RV64:
     // TODO: As the result is sign-extended, this is conservatively correct. A
     // more precise answer could be calculated for SRAW depending on known
     // bits in the shift amount.
     return 33;
   case RISCVISD::SHFL:
   case RISCVISD::UNSHFL: {
     // There is no SHFLIW, but a i64 SHFLI with bit 4 of the control word
     // cleared doesn't affect bit 31. The upper 32 bits will be shuffled, but
#define NODE_NAME_CASE(NODE)                                                   \
   NODE_NAME_CASE(FMV_H_X)
   NODE_NAME_CASE(FMV_X_ANYEXTH)
   NODE_NAME_CASE(FMV_W_X_RV64)
   NODE_NAME_CASE(FMV_X_ANYEXTW_RV64)
   NODE_NAME_CASE(FCVT_X_RTZ)
   NODE_NAME_CASE(FCVT_XU_RTZ)
   NODE_NAME_CASE(FCVT_W_RTZ_RV64)
   NODE_NAME_CASE(FCVT_WU_RTZ_RV64)
+  NODE_NAME_CASE(STRICT_FCVT_W_RTZ_RV64)
+  NODE_NAME_CASE(STRICT_FCVT_WU_RTZ_RV64)
   NODE_NAME_CASE(READ_CYCLE_WIDE)
   NODE_NAME_CASE(GREV)
   NODE_NAME_CASE(GREVW)
   NODE_NAME_CASE(GORC)
   NODE_NAME_CASE(GORCW)
   NODE_NAME_CASE(SHFL)
   NODE_NAME_CASE(SHFLW)
   NODE_NAME_CASE(UNSHFL)

llvm/lib/Target/RISCV/RISCVInstrInfoD.td

 class PatFpr64Fpr64DynFrm<SDPatternOperator OpNode, RVInstRFrm Inst>
     : Pat<(OpNode FPR64:$rs1, FPR64:$rs2), (Inst $rs1, $rs2, 0b111)>;
 
 let Predicates = [HasStdExtD] in {
 
 /// Float conversion operations
 
 // f64 -> f32, f32 -> f64
-def : Pat<(fpround FPR64:$rs1), (FCVT_S_D FPR64:$rs1, 0b111)>;
-def : Pat<(fpextend FPR32:$rs1), (FCVT_D_S FPR32:$rs1)>;
+def : Pat<(any_fpround FPR64:$rs1), (FCVT_S_D FPR64:$rs1, 0b111)>;
+def : Pat<(any_fpextend FPR32:$rs1), (FCVT_D_S FPR32:$rs1)>;
 
 // [u]int<->double conversion patterns must be gated on IsRV32 or IsRV64, so
 // are defined later.
 
 /// Float arithmetic operations
 
 def : PatFpr64Fpr64DynFrm<any_fadd, FADD_D>;
 def : PatFpr64Fpr64DynFrm<any_fsub, FSUB_D>;

 } // Predicates = [HasStdExtD]
 
 let Predicates = [HasStdExtD, IsRV32] in {
 
 /// Float constants
 def : Pat<(f64 (fpimm0)), (FCVT_D_W (i32 X0))>;
 
 // double->[u]int. Round-to-zero must be used.
-def : Pat<(i32 (fp_to_sint FPR64:$rs1)), (FCVT_W_D FPR64:$rs1, 0b001)>;
-def : Pat<(i32 (fp_to_uint FPR64:$rs1)), (FCVT_WU_D FPR64:$rs1, 0b001)>;
+def : Pat<(i32 (any_fp_to_sint FPR64:$rs1)), (FCVT_W_D FPR64:$rs1, 0b001)>;
+def : Pat<(i32 (any_fp_to_uint FPR64:$rs1)), (FCVT_WU_D FPR64:$rs1, 0b001)>;
 
 // Saturating double->[u]int32.
 def : Pat<(i32 (riscv_fcvt_x_rtz FPR64:$rs1)), (FCVT_W_D $rs1, 0b001)>;
 def : Pat<(i32 (riscv_fcvt_xu_rtz FPR64:$rs1)), (FCVT_WU_D $rs1, 0b001)>;
 
 // float->int32 with current rounding mode.
 def : Pat<(i32 (lrint FPR64:$rs1)), (FCVT_W_D $rs1, 0b111)>;
 
 // float->int32 rounded to nearest with ties rounded away from zero.
 def : Pat<(i32 (lround FPR64:$rs1)), (FCVT_W_D $rs1, 0b100)>;
 
 // [u]int->double.
-def : Pat<(sint_to_fp (i32 GPR:$rs1)), (FCVT_D_W GPR:$rs1)>;
-def : Pat<(uint_to_fp (i32 GPR:$rs1)), (FCVT_D_WU GPR:$rs1)>;
+def : Pat<(any_sint_to_fp (i32 GPR:$rs1)), (FCVT_D_W GPR:$rs1)>;
+def : Pat<(any_uint_to_fp (i32 GPR:$rs1)), (FCVT_D_WU GPR:$rs1)>;
 } // Predicates = [HasStdExtD, IsRV32]
 
 let Predicates = [HasStdExtD, IsRV64] in {
 
 /// Float constants
 def : Pat<(f64 (fpimm0)), (FMV_D_X (i64 X0))>;
 
 // Moves (no conversion)
 def : Pat<(bitconvert (i64 GPR:$rs1)), (FMV_D_X GPR:$rs1)>;
 def : Pat<(i64 (bitconvert FPR64:$rs1)), (FMV_X_D FPR64:$rs1)>;
 
 // Use target specific isd nodes to help us remember the result is sign
 // extended. Matching sext_inreg+fptoui/fptosi may cause the conversion to be
 // duplicated if it has another user that didn't need the sign_extend.
-def : Pat<(riscv_fcvt_w_rtz_rv64 FPR64:$rs1),  (FCVT_W_D $rs1, 0b001)>;
-def : Pat<(riscv_fcvt_wu_rtz_rv64 FPR64:$rs1), (FCVT_WU_D $rs1, 0b001)>;
+def : Pat<(riscv_any_fcvt_w_rtz_rv64 FPR64:$rs1),  (FCVT_W_D $rs1, 0b001)>;
+def : Pat<(riscv_any_fcvt_wu_rtz_rv64 FPR64:$rs1), (FCVT_WU_D $rs1, 0b001)>;
 
 // [u]int32->fp
-def : Pat<(sint_to_fp (i64 (sexti32 (i64 GPR:$rs1)))), (FCVT_D_W $rs1)>;
-def : Pat<(uint_to_fp (i64 (zexti32 (i64 GPR:$rs1)))), (FCVT_D_WU $rs1)>;
+def : Pat<(any_sint_to_fp (i64 (sexti32 (i64 GPR:$rs1)))), (FCVT_D_W $rs1)>;
+def : Pat<(any_uint_to_fp (i64 (zexti32 (i64 GPR:$rs1)))), (FCVT_D_WU $rs1)>;
 
 // Saturating double->[u]int64.
 def : Pat<(i64 (riscv_fcvt_x_rtz FPR64:$rs1)), (FCVT_L_D $rs1, 0b001)>;
 def : Pat<(i64 (riscv_fcvt_xu_rtz FPR64:$rs1)), (FCVT_LU_D $rs1, 0b001)>;
 
 // double->[u]int64. Round-to-zero must be used.
-def : Pat<(i64 (fp_to_sint FPR64:$rs1)), (FCVT_L_D FPR64:$rs1, 0b001)>;
-def : Pat<(i64 (fp_to_uint FPR64:$rs1)), (FCVT_LU_D FPR64:$rs1, 0b001)>;
+def : Pat<(i64 (any_fp_to_sint FPR64:$rs1)), (FCVT_L_D FPR64:$rs1, 0b001)>;
+def : Pat<(i64 (any_fp_to_uint FPR64:$rs1)), (FCVT_LU_D FPR64:$rs1, 0b001)>;
 
 // double->int64 with current rounding mode.
 def : Pat<(i64 (lrint FPR64:$rs1)), (FCVT_L_D $rs1, 0b111)>;
 def : Pat<(i64 (llrint FPR64:$rs1)), (FCVT_L_D $rs1, 0b111)>;
 
 // double->int64 rounded to nearest with ties rounded away from zero.
 def : Pat<(i64 (lround FPR64:$rs1)), (FCVT_L_D $rs1, 0b100)>;
 def : Pat<(i64 (llround FPR64:$rs1)), (FCVT_L_D $rs1, 0b100)>;
 
 // [u]int64->fp. Match GCC and default to using dynamic rounding mode.
-def : Pat<(sint_to_fp (i64 GPR:$rs1)), (FCVT_D_L GPR:$rs1, 0b111)>;
-def : Pat<(uint_to_fp (i64 GPR:$rs1)), (FCVT_D_LU GPR:$rs1, 0b111)>;
+def : Pat<(any_sint_to_fp (i64 GPR:$rs1)), (FCVT_D_L GPR:$rs1, 0b111)>;
+def : Pat<(any_uint_to_fp (i64 GPR:$rs1)), (FCVT_D_LU GPR:$rs1, 0b111)>;
 } // Predicates = [HasStdExtD, IsRV64]

llvm/lib/Target/RISCV/RISCVInstrInfoF.td

 //===----------------------------------------------------------------------===//
 // RISC-V specific DAG Nodes.
 //===----------------------------------------------------------------------===//
 
 def SDT_RISCVFMV_W_X_RV64
     : SDTypeProfile<1, 1, [SDTCisVT<0, f32>, SDTCisVT<1, i64>]>;
 def SDT_RISCVFMV_X_ANYEXTW_RV64
     : SDTypeProfile<1, 1, [SDTCisVT<0, i64>, SDTCisVT<1, f32>]>;
-def STD_RISCVFCVT_W_RV64
+def SDT_RISCVFCVT_W_RV64
     : SDTypeProfile<1, 1, [SDTCisVT<0, i64>, SDTCisFP<1>]>;
-def STD_RISCVFCVT_X
+def SDT_RISCVFCVT_X
     : SDTypeProfile<1, 1, [SDTCisVT<0, XLenVT>, SDTCisFP<1>]>;
 
 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_rtz_rv64
-    : SDNode<"RISCVISD::FCVT_W_RTZ_RV64", STD_RISCVFCVT_W_RV64>;
+    : SDNode<"RISCVISD::FCVT_W_RTZ_RV64", SDT_RISCVFCVT_W_RV64>;
 def riscv_fcvt_wu_rtz_rv64
-    : SDNode<"RISCVISD::FCVT_WU_RTZ_RV64", STD_RISCVFCVT_W_RV64>;
+    : SDNode<"RISCVISD::FCVT_WU_RTZ_RV64", SDT_RISCVFCVT_W_RV64>;
 def riscv_fcvt_x_rtz
-    : SDNode<"RISCVISD::FCVT_X_RTZ", STD_RISCVFCVT_X>;
+    : SDNode<"RISCVISD::FCVT_X_RTZ", SDT_RISCVFCVT_X>;
 def riscv_fcvt_xu_rtz
-    : SDNode<"RISCVISD::FCVT_XU_RTZ", STD_RISCVFCVT_X>;
+    : SDNode<"RISCVISD::FCVT_XU_RTZ", SDT_RISCVFCVT_X>;
+
+def riscv_strict_fcvt_w_rtz_rv64
+    : SDNode<"RISCVISD::STRICT_FCVT_W_RTZ_RV64", SDT_RISCVFCVT_W_RV64,
+             [SDNPHasChain]>;
+def riscv_strict_fcvt_wu_rtz_rv64
+    : SDNode<"RISCVISD::STRICT_FCVT_WU_RTZ_RV64", SDT_RISCVFCVT_W_RV64,
+             [SDNPHasChain]>;
+
+def riscv_any_fcvt_w_rtz_rv64 : PatFrags<(ops node:$src),
+                                         [(riscv_strict_fcvt_w_rtz_rv64 node:$src),
+                                          (riscv_fcvt_w_rtz_rv64 node:$src)]>;
+def riscv_any_fcvt_wu_rtz_rv64 : PatFrags<(ops node:$src),
+                                          [(riscv_strict_fcvt_wu_rtz_rv64 node:$src),
+                                           (riscv_fcvt_wu_rtz_rv64 node:$src)]>;
 
 //===----------------------------------------------------------------------===//
 // Operand and SDNode transformation definitions.
 //===----------------------------------------------------------------------===//
 
 // Floating-point rounding mode
 
 def FRMArg : AsmOperandClass {

 } // Predicates = [HasStdExtF]
 
 let Predicates = [HasStdExtF, IsRV32] in {
 // Moves (no conversion)
 def : Pat<(bitconvert (i32 GPR:$rs1)), (FMV_W_X GPR:$rs1)>;
 def : Pat<(i32 (bitconvert FPR32:$rs1)), (FMV_X_W FPR32:$rs1)>;
 
 // float->[u]int. Round-to-zero must be used.
-def : Pat<(i32 (fp_to_sint FPR32:$rs1)), (FCVT_W_S $rs1, 0b001)>;
-def : Pat<(i32 (fp_to_uint FPR32:$rs1)), (FCVT_WU_S $rs1, 0b001)>;
+def : Pat<(i32 (any_fp_to_sint FPR32:$rs1)), (FCVT_W_S $rs1, 0b001)>;
+def : Pat<(i32 (any_fp_to_uint FPR32:$rs1)), (FCVT_WU_S $rs1, 0b001)>;
 
 // Saturating float->[u]int32.
 def : Pat<(i32 (riscv_fcvt_x_rtz FPR32:$rs1)), (FCVT_W_S $rs1, 0b001)>;
 def : Pat<(i32 (riscv_fcvt_xu_rtz FPR32:$rs1)), (FCVT_WU_S $rs1, 0b001)>;
 
 // float->int32 with current rounding mode.
 def : Pat<(i32 (lrint FPR32:$rs1)), (FCVT_W_S $rs1, 0b111)>;
 
 // float->int32 rounded to nearest with ties rounded away from zero.
 def : Pat<(i32 (lround FPR32:$rs1)), (FCVT_W_S $rs1, 0b100)>;
 
 // [u]int->float. Match GCC and default to using dynamic rounding mode.
-def : Pat<(sint_to_fp (i32 GPR:$rs1)), (FCVT_S_W $rs1, 0b111)>;
-def : Pat<(uint_to_fp (i32 GPR:$rs1)), (FCVT_S_WU $rs1, 0b111)>;
+def : Pat<(any_sint_to_fp (i32 GPR:$rs1)), (FCVT_S_W $rs1, 0b111)>;
+def : Pat<(any_uint_to_fp (i32 GPR:$rs1)), (FCVT_S_WU $rs1, 0b111)>;
 } // Predicates = [HasStdExtF, IsRV32]
 
 let Predicates = [HasStdExtF, IsRV64] in {
 // Moves (no conversion)
 def : Pat<(riscv_fmv_w_x_rv64 GPR:$src), (FMV_W_X GPR:$src)>;
 def : Pat<(riscv_fmv_x_anyextw_rv64 FPR32:$src), (FMV_X_W FPR32:$src)>;
 def : Pat<(sext_inreg (riscv_fmv_x_anyextw_rv64 FPR32:$src), i32),
           (FMV_X_W FPR32:$src)>;
 
 // Use target specific isd nodes to help us remember the result is sign
 // extended. Matching sext_inreg+fptoui/fptosi may cause the conversion to be
 // duplicated if it has another user that didn't need the sign_extend.
-def : Pat<(riscv_fcvt_w_rtz_rv64 FPR32:$rs1),  (FCVT_W_S $rs1, 0b001)>;
-def : Pat<(riscv_fcvt_wu_rtz_rv64 FPR32:$rs1), (FCVT_WU_S $rs1, 0b001)>;
+def : Pat<(riscv_any_fcvt_w_rtz_rv64 FPR32:$rs1),  (FCVT_W_S $rs1, 0b001)>;
+def : Pat<(riscv_any_fcvt_wu_rtz_rv64 FPR32:$rs1), (FCVT_WU_S $rs1, 0b001)>;
 
 // float->[u]int64. Round-to-zero must be used.
-def : Pat<(i64 (fp_to_sint FPR32:$rs1)), (FCVT_L_S $rs1, 0b001)>;
-def : Pat<(i64 (fp_to_uint FPR32:$rs1)), (FCVT_LU_S $rs1, 0b001)>;
+def : Pat<(i64 (any_fp_to_sint FPR32:$rs1)), (FCVT_L_S $rs1, 0b001)>;
+def : Pat<(i64 (any_fp_to_uint FPR32:$rs1)), (FCVT_LU_S $rs1, 0b001)>;
 
 // Saturating float->[u]int64.
 def : Pat<(i64 (riscv_fcvt_x_rtz FPR32:$rs1)), (FCVT_L_S $rs1, 0b001)>;
 def : Pat<(i64 (riscv_fcvt_xu_rtz FPR32:$rs1)), (FCVT_LU_S $rs1, 0b001)>;
 
 // float->int64 with current rounding mode.
 def : Pat<(i64 (lrint FPR32:$rs1)), (FCVT_L_S $rs1, 0b111)>;
 def : Pat<(i64 (llrint FPR32:$rs1)), (FCVT_L_S $rs1, 0b111)>;
 
 // float->int64 rounded to neartest with ties rounded away from zero.
 def : Pat<(i64 (lround FPR32:$rs1)), (FCVT_L_S $rs1, 0b100)>;
 def : Pat<(i64 (llround FPR32:$rs1)), (FCVT_L_S $rs1, 0b100)>;
 
 // [u]int->fp. Match GCC and default to using dynamic rounding mode.
-def : Pat<(sint_to_fp (i64 (sexti32 (i64 GPR:$rs1)))), (FCVT_S_W $rs1, 0b111)>;
-def : Pat<(uint_to_fp (i64 (zexti32 (i64 GPR:$rs1)))), (FCVT_S_WU $rs1, 0b111)>;
-def : Pat<(sint_to_fp (i64 GPR:$rs1)), (FCVT_S_L $rs1, 0b111)>;
-def : Pat<(uint_to_fp (i64 GPR:$rs1)), (FCVT_S_LU $rs1, 0b111)>;
+def : Pat<(any_sint_to_fp (i64 (sexti32 (i64 GPR:$rs1)))), (FCVT_S_W $rs1, 0b111)>;
+def : Pat<(any_uint_to_fp (i64 (zexti32 (i64 GPR:$rs1)))), (FCVT_S_WU $rs1, 0b111)>;
+def : Pat<(any_sint_to_fp (i64 GPR:$rs1)), (FCVT_S_L $rs1, 0b111)>;
+def : Pat<(any_uint_to_fp (i64 GPR:$rs1)), (FCVT_S_LU $rs1, 0b111)>;
 } // Predicates = [HasStdExtF, IsRV64]

llvm/lib/Target/RISCV/RISCVInstrInfoZfh.td

 
 /// Stores
 
 defm : StPat<store, FSH, FPR16, f16>;
 
 /// Float conversion operations
 
 // f32 -> f16, f16 -> f32
-def : Pat<(fpround FPR32:$rs1), (FCVT_H_S FPR32:$rs1, 0b111)>;
-def : Pat<(fpextend FPR16:$rs1), (FCVT_S_H FPR16:$rs1)>;
+def : Pat<(any_fpround FPR32:$rs1), (FCVT_H_S FPR32:$rs1, 0b111)>;
+def : Pat<(any_fpextend FPR16:$rs1), (FCVT_S_H FPR16:$rs1)>;
 
 // Moves (no conversion)
 def : Pat<(riscv_fmv_h_x GPR:$src), (FMV_H_X GPR:$src)>;
 def : Pat<(riscv_fmv_x_anyexth FPR16:$src), (FMV_X_H FPR16:$src)>;
 } // Predicates = [HasStdExtZfhmin]
 
 let Predicates = [HasStdExtZfh, IsRV32] in {
 // half->[u]int. Round-to-zero must be used.
-def : Pat<(i32 (fp_to_sint FPR16:$rs1)), (FCVT_W_H $rs1, 0b001)>;
-def : Pat<(i32 (fp_to_uint FPR16:$rs1)), (FCVT_WU_H $rs1, 0b001)>;
+def : Pat<(i32 (any_fp_to_sint FPR16:$rs1)), (FCVT_W_H $rs1, 0b001)>;
+def : Pat<(i32 (any_fp_to_uint FPR16:$rs1)), (FCVT_WU_H $rs1, 0b001)>;
 
 // Saturating float->[u]int32.
 def : Pat<(i32 (riscv_fcvt_x_rtz FPR16:$rs1)), (FCVT_W_H $rs1, 0b001)>;
 def : Pat<(i32 (riscv_fcvt_xu_rtz FPR16:$rs1)), (FCVT_WU_H $rs1, 0b001)>;
 
 // half->int32 with current rounding mode.
 def : Pat<(i32 (lrint FPR16:$rs1)), (FCVT_W_H $rs1, 0b111)>;
 
 // half->int32 rounded to nearest with ties rounded away from zero.
 def : Pat<(i32 (lround FPR16:$rs1)), (FCVT_W_H $rs1, 0b100)>;
 
 // [u]int->half. Match GCC and default to using dynamic rounding mode.
-def : Pat<(sint_to_fp (i32 GPR:$rs1)), (FCVT_H_W $rs1, 0b111)>;
-def : Pat<(uint_to_fp (i32 GPR:$rs1)), (FCVT_H_WU $rs1, 0b111)>;
+def : Pat<(any_sint_to_fp (i32 GPR:$rs1)), (FCVT_H_W $rs1, 0b111)>;
+def : Pat<(any_uint_to_fp (i32 GPR:$rs1)), (FCVT_H_WU $rs1, 0b111)>;
 } // Predicates = [HasStdExtZfh, IsRV32]
 
 let Predicates = [HasStdExtZfh, IsRV64] in {
 // Use target specific isd nodes to help us remember the result is sign
 // extended. Matching sext_inreg+fptoui/fptosi may cause the conversion to be
 // duplicated if it has another user that didn't need the sign_extend.
-def : Pat<(riscv_fcvt_w_rtz_rv64 FPR16:$rs1),  (FCVT_W_H $rs1, 0b001)>;
-def : Pat<(riscv_fcvt_wu_rtz_rv64 FPR16:$rs1), (FCVT_WU_H $rs1, 0b001)>;
+def : Pat<(riscv_any_fcvt_w_rtz_rv64 FPR16:$rs1),  (FCVT_W_H $rs1, 0b001)>;
+def : Pat<(riscv_any_fcvt_wu_rtz_rv64 FPR16:$rs1), (FCVT_WU_H $rs1, 0b001)>;
 
 // half->[u]int64. Round-to-zero must be used.
-def : Pat<(i64 (fp_to_sint FPR16:$rs1)), (FCVT_L_H $rs1, 0b001)>;
-def : Pat<(i64 (fp_to_uint FPR16:$rs1)), (FCVT_LU_H $rs1, 0b001)>;
+def : Pat<(i64 (any_fp_to_sint FPR16:$rs1)), (FCVT_L_H $rs1, 0b001)>;
+def : Pat<(i64 (any_fp_to_uint FPR16:$rs1)), (FCVT_LU_H $rs1, 0b001)>;
 
 // Saturating float->[u]int64.
 def : Pat<(i64 (riscv_fcvt_x_rtz FPR16:$rs1)), (FCVT_L_H $rs1, 0b001)>;
 def : Pat<(i64 (riscv_fcvt_xu_rtz FPR16:$rs1)), (FCVT_LU_H $rs1, 0b001)>;
 
 // half->int64 with current rounding mode.
 def : Pat<(i64 (lrint FPR16:$rs1)), (FCVT_L_H $rs1, 0b111)>;
 def : Pat<(i64 (llrint FPR16:$rs1)), (FCVT_L_H $rs1, 0b111)>;
 
 // half->int64 rounded to nearest with ties rounded away from zero.
 def : Pat<(i64 (lround FPR16:$rs1)), (FCVT_L_H $rs1, 0b100)>;
 def : Pat<(i64 (llround FPR16:$rs1)), (FCVT_L_H $rs1, 0b100)>;
 
 // [u]int->fp. Match GCC and default to using dynamic rounding mode.
-def : Pat<(sint_to_fp (i64 (sexti32 (i64 GPR:$rs1)))), (FCVT_H_W $rs1, 0b111)>;
-def : Pat<(uint_to_fp (i64 (zexti32 (i64 GPR:$rs1)))), (FCVT_H_WU $rs1, 0b111)>;
-def : Pat<(sint_to_fp (i64 GPR:$rs1)), (FCVT_H_L $rs1, 0b111)>;
-def : Pat<(uint_to_fp (i64 GPR:$rs1)), (FCVT_H_LU $rs1, 0b111)>;
+def : Pat<(any_sint_to_fp (i64 (sexti32 (i64 GPR:$rs1)))), (FCVT_H_W $rs1, 0b111)>;
+def : Pat<(any_uint_to_fp (i64 (zexti32 (i64 GPR:$rs1)))), (FCVT_H_WU $rs1, 0b111)>;
+def : Pat<(any_sint_to_fp (i64 GPR:$rs1)), (FCVT_H_L $rs1, 0b111)>;
+def : Pat<(any_uint_to_fp (i64 GPR:$rs1)), (FCVT_H_LU $rs1, 0b111)>;
 } // Predicates = [HasStdExtZfh, IsRV64]
 
 let Predicates = [HasStdExtZfhmin, HasStdExtD] in {
 /// Float conversion operations
 // f64 -> f16, f16 -> f64
-def : Pat<(fpround FPR64:$rs1), (FCVT_H_D FPR64:$rs1, 0b111)>;
-def : Pat<(fpextend FPR16:$rs1), (FCVT_D_H FPR16:$rs1)>;
+def : Pat<(any_fpround FPR64:$rs1), (FCVT_H_D FPR64:$rs1, 0b111)>;
+def : Pat<(any_fpextend FPR16:$rs1), (FCVT_D_H FPR16:$rs1)>;
 
 /// Float arithmetic operations
 def : Pat<(fcopysign FPR16:$rs1, FPR64:$rs2),
           (FSGNJ_H $rs1, (FCVT_H_D $rs2, 0b111))>;
 def : Pat<(fcopysign FPR64:$rs1, FPR16:$rs2), (FSGNJ_D $rs1, (FCVT_D_H $rs2))>;
 } // Predicates = [HasStdExtZfhmin, HasStdExtD]

llvm/test/CodeGen/RISCV/double-convert-strict.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+d -verify-machineinstrs < %s \
+; RUN:   -disable-strictnode-mutation | FileCheck -check-prefix=RV32IFD %s
+; RUN: llc -mtriple=riscv64 -mattr=+d -verify-machineinstrs < %s \
+; RUN:   -disable-strictnode-mutation | FileCheck -check-prefix=RV64IFD %s
+; RUN: llc -mtriple=riscv32 -verify-machineinstrs < %s \
+; RUN:   -disable-strictnode-mutation | FileCheck -check-prefix=RV32I %s
+; RUN: llc -mtriple=riscv64 -verify-machineinstrs < %s \
+; RUN:   -disable-strictnode-mutation | FileCheck -check-prefix=RV64I %s
+
+define float @fcvt_s_d(double %a) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_s_d:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw a0, 8(sp)
+; RV32IFD-NEXT:    sw a1, 12(sp)
+; RV32IFD-NEXT:    fld ft0, 8(sp)
+; RV32IFD-NEXT:    fcvt.s.d ft0, ft0
+; RV32IFD-NEXT:    fmv.x.w a0, ft0
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_s_d:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    fmv.d.x ft0, a0
+; RV64IFD-NEXT:    fcvt.s.d ft0, ft0
+; RV64IFD-NEXT:    fmv.x.w a0, ft0
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_s_d:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __truncdfsf2@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_s_d:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __truncdfsf2@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.fptrunc.f32.f64(double %a, metadata !"round.dynamic", metadata !"fpexcept.strict")
+  ret float %1
+}
+declare float @llvm.experimental.constrained.fptrunc.f32.f64(double, metadata, metadata)
+
+define double @fcvt_d_s(float %a) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_d_s:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    fmv.w.x ft0, a0
+; RV32IFD-NEXT:    fcvt.d.s ft0, ft0
+; RV32IFD-NEXT:    fsd ft0, 8(sp)
+; RV32IFD-NEXT:    lw a0, 8(sp)
+; RV32IFD-NEXT:    lw a1, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_d_s:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    fmv.w.x ft0, a0
+; RV64IFD-NEXT:    fcvt.d.s ft0, ft0
+; RV64IFD-NEXT:    fmv.x.d a0, ft0
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_d_s:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __extendsfdf2@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_d_s:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __extendsfdf2@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.fpext.f64.f32(float %a, metadata !"fpexcept.strict")
+  ret double %1
+}
+declare double @llvm.experimental.constrained.fpext.f64.f32(float, metadata)
+
+; For RV64D, fcvt.l.d is semantically equivalent to fcvt.w.d in this case
+; because fptosi will produce poison if the result doesn't fit into an i32.
+define i32 @fcvt_w_d(double %a) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_w_d:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw a0, 8(sp)
+; RV32IFD-NEXT:    sw a1, 12(sp)
+; RV32IFD-NEXT:    fld ft0, 8(sp)
+; RV32IFD-NEXT:    fcvt.w.d a0, ft0, rtz
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_w_d:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    fmv.d.x ft0, a0
+; RV64IFD-NEXT:    fcvt.w.d a0, ft0, rtz
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_w_d:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __fixdfsi@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_w_d:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __fixdfsi@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptosi.i32.f64(double %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+declare i32 @llvm.experimental.constrained.fptosi.i32.f64(double, metadata)
+
+; For RV64D, fcvt.lu.d is semantically equivalent to fcvt.wu.d in this case
+; because fptosi will produce poison if the result doesn't fit into an i32.
+define i32 @fcvt_wu_d(double %a) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_wu_d:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw a0, 8(sp)
+; RV32IFD-NEXT:    sw a1, 12(sp)
+; RV32IFD-NEXT:    fld ft0, 8(sp)
+; RV32IFD-NEXT:    fcvt.wu.d a0, ft0, rtz
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_wu_d:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    fmv.d.x ft0, a0
+; RV64IFD-NEXT:    fcvt.wu.d a0, ft0, rtz
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_wu_d:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __fixunsdfsi@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_wu_d:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __fixunsdfsi@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptoui.i32.f64(double %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+declare i32 @llvm.experimental.constrained.fptoui.i32.f64(double, metadata)
+
+; Test where the fptoui has multiple uses, one of which causes a sext to be
+; inserted on RV64.
+define i32 @fcvt_wu_d_multiple_use(double %x, i32* %y) {
+; RV32IFD-LABEL: fcvt_wu_d_multiple_use:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    .cfi_def_cfa_offset 16
+; RV32IFD-NEXT:    sw a0, 8(sp)
+; RV32IFD-NEXT:    sw a1, 12(sp)
+; RV32IFD-NEXT:    fld ft0, 8(sp)
+; RV32IFD-NEXT:    fcvt.wu.d a1, ft0, rtz
+; RV32IFD-NEXT:    li a0, 1
+; RV32IFD-NEXT:    beqz a1, .LBB4_2
+; RV32IFD-NEXT:  # %bb.1:
+; RV32IFD-NEXT:    mv a0, a1
+; RV32IFD-NEXT:  .LBB4_2:
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_wu_d_multiple_use:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    fmv.d.x ft0, a0
+; RV64IFD-NEXT:    fcvt.wu.d a1, ft0, rtz
+; RV64IFD-NEXT:    li a0, 1
+; RV64IFD-NEXT:    beqz a1, .LBB4_2
+; RV64IFD-NEXT:  # %bb.1:
+; RV64IFD-NEXT:    mv a0, a1
+; RV64IFD-NEXT:  .LBB4_2:
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_wu_d_multiple_use:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    .cfi_def_cfa_offset 16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    .cfi_offset ra, -4
+; RV32I-NEXT:    call __fixunsdfsi@plt
+; RV32I-NEXT:    mv a1, a0
+; RV32I-NEXT:    li a0, 1
+; RV32I-NEXT:    beqz a1, .LBB4_2
+; RV32I-NEXT:  # %bb.1:
+; RV32I-NEXT:    mv a0, a1
+; RV32I-NEXT:  .LBB4_2:
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_wu_d_multiple_use:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    .cfi_def_cfa_offset 16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    .cfi_offset ra, -8
+; RV64I-NEXT:    call __fixunsdfsi@plt
+; RV64I-NEXT:    mv a1, a0
+; RV64I-NEXT:    li a0, 1
+; RV64I-NEXT:    beqz a1, .LBB4_2
+; RV64I-NEXT:  # %bb.1:
+; RV64I-NEXT:    mv a0, a1
+; RV64I-NEXT:  .LBB4_2:
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %a = call i32 @llvm.experimental.constrained.fptoui.i32.f64(double %x, metadata !"fpexcept.strict") strictfp
+  %b = icmp eq i32 %a, 0
+  %c = select i1 %b, i32 1, i32 %a
+  ret i32 %c
+}
+
+define double @fcvt_d_w(i32 %a) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_d_w:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    fcvt.d.w ft0, a0
+; RV32IFD-NEXT:    fsd ft0, 8(sp)
+; RV32IFD-NEXT:    lw a0, 8(sp)
+; RV32IFD-NEXT:    lw a1, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_d_w:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    fcvt.d.w ft0, a0
+; RV64IFD-NEXT:    fmv.x.d a0, ft0
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_d_w:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatsidf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_d_w:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sext.w a0, a0
+; RV64I-NEXT:    call __floatsidf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.sitofp.f64.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}
+declare double @llvm.experimental.constrained.sitofp.f64.i32(i32, metadata, metadata)
+
+define double @fcvt_d_w_load(i32* %p) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_d_w_load:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    lw a0, 0(a0)
+; RV32IFD-NEXT:    fcvt.d.w ft0, a0
+; RV32IFD-NEXT:    fsd ft0, 8(sp)
+; RV32IFD-NEXT:    lw a0, 8(sp)
+; RV32IFD-NEXT:    lw a1, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_d_w_load:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    lw a0, 0(a0)
+; RV64IFD-NEXT:    fcvt.d.w ft0, a0
+; RV64IFD-NEXT:    fmv.x.d a0, ft0
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_d_w_load:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    lw a0, 0(a0)
+; RV32I-NEXT:    call __floatsidf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_d_w_load:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    lw a0, 0(a0)
+; RV64I-NEXT:    call __floatsidf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %a = load i32, i32* %p
+  %1 = call double @llvm.experimental.constrained.sitofp.f64.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}
+
+define double @fcvt_d_wu(i32 %a) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_d_wu:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    fcvt.d.wu ft0, a0
+; RV32IFD-NEXT:    fsd ft0, 8(sp)
+; RV32IFD-NEXT:    lw a0, 8(sp)
+; RV32IFD-NEXT:    lw a1, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_d_wu:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    fcvt.d.wu ft0, a0
+; RV64IFD-NEXT:    fmv.x.d a0, ft0
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_d_wu:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatunsidf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_d_wu:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sext.w a0, a0
+; RV64I-NEXT:    call __floatunsidf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}
+declare double @llvm.experimental.constrained.uitofp.f64.i32(i32, metadata, metadata)
+
+define double @fcvt_d_wu_load(i32* %p) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_d_wu_load:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    lw a0, 0(a0)
+; RV32IFD-NEXT:    fcvt.d.wu ft0, a0
+; RV32IFD-NEXT:    fsd ft0, 8(sp)
+; RV32IFD-NEXT:    lw a0, 8(sp)
+; RV32IFD-NEXT:    lw a1, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_d_wu_load:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    lwu a0, 0(a0)
+; RV64IFD-NEXT:    fcvt.d.wu ft0, a0
+; RV64IFD-NEXT:    fmv.x.d a0, ft0
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_d_wu_load:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    lw a0, 0(a0)
+; RV32I-NEXT:    call __floatunsidf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_d_wu_load:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    lw a0, 0(a0)
+; RV64I-NEXT:    call __floatunsidf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %a = load i32, i32* %p
+  %1 = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}
+
+define i64 @fcvt_l_d(double %a) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_l_d:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IFD-NEXT:    call __fixdfdi@plt
+; RV32IFD-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_l_d:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    fmv.d.x ft0, a0
+; RV64IFD-NEXT:    fcvt.l.d a0, ft0, rtz
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_l_d:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __fixdfdi@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_l_d:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __fixdfdi@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call i64 @llvm.experimental.constrained.fptosi.i64.f64(double %a, metadata !"fpexcept.strict") strictfp
+  ret i64 %1
+}
+declare i64 @llvm.experimental.constrained.fptosi.i64.f64(double, metadata)
+
+define i64 @fcvt_lu_d(double %a) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_lu_d:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IFD-NEXT:    call __fixunsdfdi@plt
+; RV32IFD-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_lu_d:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    fmv.d.x ft0, a0
+; RV64IFD-NEXT:    fcvt.lu.d a0, ft0, rtz
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_lu_d:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __fixunsdfdi@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_lu_d:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __fixunsdfdi@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call i64 @llvm.experimental.constrained.fptoui.i64.f64(double %a, metadata !"fpexcept.strict") strictfp
+  ret i64 %1
+}
+declare i64 @llvm.experimental.constrained.fptoui.i64.f64(double, metadata)
+
+define double @fcvt_d_l(i64 %a) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_d_l:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IFD-NEXT:    call __floatdidf@plt
+; RV32IFD-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_d_l:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    fcvt.d.l ft0, a0
+; RV64IFD-NEXT:    fmv.x.d a0, ft0
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_d_l:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatdidf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_d_l:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __floatdidf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.sitofp.f64.i64(i64 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}
+declare double @llvm.experimental.constrained.sitofp.f64.i64(i64, metadata, metadata)
+
+define double @fcvt_d_lu(i64 %a) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_d_lu:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IFD-NEXT:    call __floatundidf@plt
+; RV32IFD-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_d_lu:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    fcvt.d.lu ft0, a0
+; RV64IFD-NEXT:    fmv.x.d a0, ft0
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_d_lu:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatundidf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_d_lu:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __floatundidf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.uitofp.f64.i64(i64 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}
+declare double @llvm.experimental.constrained.uitofp.f64.i64(i64, metadata, metadata)
+
+define double @fcvt_d_w_i8(i8 signext %a) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_d_w_i8:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    fcvt.d.w ft0, a0
+; RV32IFD-NEXT:    fsd ft0, 8(sp)
+; RV32IFD-NEXT:    lw a0, 8(sp)
+; RV32IFD-NEXT:    lw a1, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_d_w_i8:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    fcvt.d.w ft0, a0
+; RV64IFD-NEXT:    fmv.x.d a0, ft0
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_d_w_i8:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatsidf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_d_w_i8:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __floatsidf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.sitofp.f64.i8(i8 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}
+declare double @llvm.experimental.constrained.sitofp.f64.i8(i8, metadata, metadata)
+
+define double @fcvt_d_wu_i8(i8 zeroext %a) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_d_wu_i8:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    fcvt.d.wu ft0, a0
+; RV32IFD-NEXT:    fsd ft0, 8(sp)
+; RV32IFD-NEXT:    lw a0, 8(sp)
+; RV32IFD-NEXT:    lw a1, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_d_wu_i8:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    fcvt.d.wu ft0, a0
+; RV64IFD-NEXT:    fmv.x.d a0, ft0
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_d_wu_i8:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatunsidf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_d_wu_i8:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __floatunsidf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.uitofp.f64.i8(i8 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}
+declare double @llvm.experimental.constrained.uitofp.f64.i8(i8, metadata, metadata)
+
+define double @fcvt_d_w_i16(i16 signext %a) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_d_w_i16:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    fcvt.d.w ft0, a0
+; RV32IFD-NEXT:    fsd ft0, 8(sp)
+; RV32IFD-NEXT:    lw a0, 8(sp)
+; RV32IFD-NEXT:    lw a1, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_d_w_i16:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    fcvt.d.w ft0, a0
+; RV64IFD-NEXT:    fmv.x.d a0, ft0
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_d_w_i16:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatsidf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_d_w_i16:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __floatsidf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.sitofp.f64.i16(i16 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}
+declare double @llvm.experimental.constrained.sitofp.f64.i16(i16, metadata, metadata)
+
+define double @fcvt_d_wu_i16(i16 zeroext %a) nounwind strictfp {
+; RV32IFD-LABEL: fcvt_d_wu_i16:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    fcvt.d.wu ft0, a0
+; RV32IFD-NEXT:    fsd ft0, 8(sp)
+; RV32IFD-NEXT:    lw a0, 8(sp)
+; RV32IFD-NEXT:    lw a1, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_d_wu_i16:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    fcvt.d.wu ft0, a0
+; RV64IFD-NEXT:    fmv.x.d a0, ft0
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_d_wu_i16:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatunsidf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_d_wu_i16:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __floatunsidf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.uitofp.f64.i16(i16 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}
+declare double @llvm.experimental.constrained.uitofp.f64.i16(i16, metadata, metadata)
+
+; Make sure we select W version of addi on RV64.
+define signext i32 @fcvt_d_w_demanded_bits(i32 signext %0, double* %1) {
+; RV32IFD-LABEL: fcvt_d_w_demanded_bits:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi a0, a0, 1
+; RV32IFD-NEXT:    fcvt.d.w ft0, a0
+; RV32IFD-NEXT:    fsd ft0, 0(a1)
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_d_w_demanded_bits:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    addiw a0, a0, 1
+; RV64IFD-NEXT:    fcvt.d.w ft0, a0
+; RV64IFD-NEXT:    fsd ft0, 0(a1)
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_d_w_demanded_bits:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    .cfi_def_cfa_offset 16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s1, 4(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    .cfi_offset ra, -4
+; RV32I-NEXT:    .cfi_offset s0, -8
+; RV32I-NEXT:    .cfi_offset s1, -12
+; RV32I-NEXT:    mv s0, a1
+; RV32I-NEXT:    addi s1, a0, 1
+; RV32I-NEXT:    mv a0, s1
+; RV32I-NEXT:    call __floatsidf@plt
+; RV32I-NEXT:    sw a1, 4(s0)
+; RV32I-NEXT:    sw a0, 0(s0)
+; RV32I-NEXT:    mv a0, s1
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s1, 4(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_d_w_demanded_bits:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -32
+; RV64I-NEXT:    .cfi_def_cfa_offset 32
+; RV64I-NEXT:    sd ra, 24(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s0, 16(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s1, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    .cfi_offset ra, -8
+; RV64I-NEXT:    .cfi_offset s0, -16
+; RV64I-NEXT:    .cfi_offset s1, -24
+; RV64I-NEXT:    mv s0, a1
+; RV64I-NEXT:    addiw s1, a0, 1
+; RV64I-NEXT:    mv a0, s1
+; RV64I-NEXT:    call __floatsidf@plt
+; RV64I-NEXT:    sd a0, 0(s0)
+; RV64I-NEXT:    mv a0, s1
+; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 32
+; RV64I-NEXT:    ret
+  %3 = add i32 %0, 1
+  %4 = call double @llvm.experimental.constrained.sitofp.f64.i32(i32 %3, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  store double %4, double* %1, align 8
+  ret i32 %3
+}
+
+; Make sure we select W version of addi on RV64.
+define signext i32 @fcvt_d_wu_demanded_bits(i32 signext %0, double* %1) {
+; RV32IFD-LABEL: fcvt_d_wu_demanded_bits:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi a0, a0, 1
+; RV32IFD-NEXT:    fcvt.d.wu ft0, a0
+; RV32IFD-NEXT:    fsd ft0, 0(a1)
+; RV32IFD-NEXT:    ret
+;
+; RV64IFD-LABEL: fcvt_d_wu_demanded_bits:
+; RV64IFD:       # %bb.0:
+; RV64IFD-NEXT:    addiw a0, a0, 1
+; RV64IFD-NEXT:    fcvt.d.wu ft0, a0
+; RV64IFD-NEXT:    fsd ft0, 0(a1)
+; RV64IFD-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_d_wu_demanded_bits:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    .cfi_def_cfa_offset 16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s1, 4(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    .cfi_offset ra, -4
+; RV32I-NEXT:    .cfi_offset s0, -8
+; RV32I-NEXT:    .cfi_offset s1, -12
+; RV32I-NEXT:    mv s0, a1
+; RV32I-NEXT:    addi s1, a0, 1
+; RV32I-NEXT:    mv a0, s1
+; RV32I-NEXT:    call __floatunsidf@plt
+; RV32I-NEXT:    sw a1, 4(s0)
+; RV32I-NEXT:    sw a0, 0(s0)
+; RV32I-NEXT:    mv a0, s1
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s1, 4(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_d_wu_demanded_bits:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -32
+; RV64I-NEXT:    .cfi_def_cfa_offset 32
+; RV64I-NEXT:    sd ra, 24(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s0, 16(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s1, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    .cfi_offset ra, -8
+; RV64I-NEXT:    .cfi_offset s0, -16
+; RV64I-NEXT:    .cfi_offset s1, -24
+; RV64I-NEXT:    mv s0, a1
+; RV64I-NEXT:    addiw s1, a0, 1
+; RV64I-NEXT:    mv a0, s1
+; RV64I-NEXT:    call __floatunsidf@plt
+; RV64I-NEXT:    sd a0, 0(s0)
+; RV64I-NEXT:    mv a0, s1
+; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 32
+; RV64I-NEXT:    ret
+  %3 = add i32 %0, 1
+  %4 = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 %3, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  store double %4, double* %1, align 8
+  ret i32 %3
+}

llvm/test/CodeGen/RISCV/float-convert-strict.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+f -verify-machineinstrs < %s \
+; RUN:   -disable-strictnode-mutation | FileCheck -check-prefix=RV32IF %s
+; RUN: llc -mtriple=riscv64 -mattr=+f -verify-machineinstrs < %s \
+; RUN:   -disable-strictnode-mutation | FileCheck -check-prefix=RV64IF %s
+; RUN: llc -mtriple=riscv32 -verify-machineinstrs < %s \
+; RUN:   -disable-strictnode-mutation | FileCheck -check-prefix=RV32I %s
+; RUN: llc -mtriple=riscv64 -verify-machineinstrs < %s \
+; RUN:   -disable-strictnode-mutation | FileCheck -check-prefix=RV64I %s
+
+; For RV64F, fcvt.l.s is semantically equivalent to fcvt.w.s in this case
+; because fptosi will produce poison if the result doesn't fit into an i32.
+define i32 @fcvt_w_s(float %a) nounwind strictfp {
+; RV32IF-LABEL: fcvt_w_s:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    fmv.w.x ft0, a0
+; RV32IF-NEXT:    fcvt.w.s a0, ft0, rtz
+; RV32IF-NEXT:    ret
+;
+; RV64IF-LABEL: fcvt_w_s:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fmv.w.x ft0, a0
+; RV64IF-NEXT:    fcvt.w.s a0, ft0, rtz
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_w_s:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __fixsfsi@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_w_s:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __fixsfsi@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptosi.i32.f32(float %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+declare i32 @llvm.experimental.constrained.fptosi.i32.f32(float, metadata)
+
+define i32 @fcvt_wu_s(float %a) nounwind strictfp {
+; RV32IF-LABEL: fcvt_wu_s:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    fmv.w.x ft0, a0
+; RV32IF-NEXT:    fcvt.wu.s a0, ft0, rtz
+; RV32IF-NEXT:    ret
+;
+; RV64IF-LABEL: fcvt_wu_s:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fmv.w.x ft0, a0
+; RV64IF-NEXT:    fcvt.wu.s a0, ft0, rtz
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_wu_s:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __fixunssfsi@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_wu_s:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __fixunssfsi@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptoui.i32.f32(float %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+declare i32 @llvm.experimental.constrained.fptoui.i32.f32(float, metadata)
+
+; Test where the fptoui has multiple uses, one of which causes a sext to be
+; inserted on RV64.
+define i32 @fcvt_wu_s_multiple_use(float %x, i32* %y) {
+; RV32IF-LABEL: fcvt_wu_s_multiple_use:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    fmv.w.x ft0, a0
+; RV32IF-NEXT:    fcvt.wu.s a1, ft0, rtz
+; RV32IF-NEXT:    li a0, 1
+; RV32IF-NEXT:    beqz a1, .LBB2_2
+; RV32IF-NEXT:  # %bb.1:
+; RV32IF-NEXT:    mv a0, a1
+; RV32IF-NEXT:  .LBB2_2:
+; RV32IF-NEXT:    ret
+;
+; RV64IF-LABEL: fcvt_wu_s_multiple_use:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fmv.w.x ft0, a0
+; RV64IF-NEXT:    fcvt.wu.s a1, ft0, rtz
+; RV64IF-NEXT:    li a0, 1
+; RV64IF-NEXT:    beqz a1, .LBB2_2
+; RV64IF-NEXT:  # %bb.1:
+; RV64IF-NEXT:    mv a0, a1
+; RV64IF-NEXT:  .LBB2_2:
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_wu_s_multiple_use:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    .cfi_def_cfa_offset 16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    .cfi_offset ra, -4
+; RV32I-NEXT:    call __fixunssfsi@plt
+; RV32I-NEXT:    mv a1, a0
+; RV32I-NEXT:    li a0, 1
+; RV32I-NEXT:    beqz a1, .LBB2_2
+; RV32I-NEXT:  # %bb.1:
+; RV32I-NEXT:    mv a0, a1
+; RV32I-NEXT:  .LBB2_2:
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_wu_s_multiple_use:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    .cfi_def_cfa_offset 16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    .cfi_offset ra, -8
+; RV64I-NEXT:    call __fixunssfsi@plt
+; RV64I-NEXT:    mv a1, a0
+; RV64I-NEXT:    li a0, 1
+; RV64I-NEXT:    beqz a1, .LBB2_2
+; RV64I-NEXT:  # %bb.1:
+; RV64I-NEXT:    mv a0, a1
+; RV64I-NEXT:  .LBB2_2:
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %a = call i32 @llvm.experimental.constrained.fptoui.i32.f32(float %x, metadata !"fpexcept.strict") strictfp
+  %b = icmp eq i32 %a, 0
+  %c = select i1 %b, i32 1, i32 %a
+  ret i32 %c
+}
+
+define float @fcvt_s_w(i32 %a) nounwind strictfp {
+; RV32IF-LABEL: fcvt_s_w:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    fcvt.s.w ft0, a0
+; RV32IF-NEXT:    fmv.x.w a0, ft0
+; RV32IF-NEXT:    ret
+;
+; RV64IF-LABEL: fcvt_s_w:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fcvt.s.w ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_s_w:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatsisf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_s_w:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sext.w a0, a0
+; RV64I-NEXT:    call __floatsisf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret float %1
+}
+declare float @llvm.experimental.constrained.sitofp.f32.i32(i32, metadata, metadata)
+
+define float @fcvt_s_w_load(i32* %p) nounwind strictfp {
+; RV32IF-LABEL: fcvt_s_w_load:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    lw a0, 0(a0)
+; RV32IF-NEXT:    fcvt.s.w ft0, a0
+; RV32IF-NEXT:    fmv.x.w a0, ft0
+; RV32IF-NEXT:    ret
+;
+; RV64IF-LABEL: fcvt_s_w_load:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    lw a0, 0(a0)
+; RV64IF-NEXT:    fcvt.s.w ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_s_w_load:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    lw a0, 0(a0)
+; RV32I-NEXT:    call __floatsisf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_s_w_load:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    lw a0, 0(a0)
+; RV64I-NEXT:    call __floatsisf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %a = load i32, i32* %p
+  %1 = call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret float %1
+}
+
+define float @fcvt_s_wu(i32 %a) nounwind strictfp {
+; RV32IF-LABEL: fcvt_s_wu:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    fcvt.s.wu ft0, a0
+; RV32IF-NEXT:    fmv.x.w a0, ft0
+; RV32IF-NEXT:    ret
+;
+; RV64IF-LABEL: fcvt_s_wu:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fcvt.s.wu ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_s_wu:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatunsisf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_s_wu:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sext.w a0, a0
+; RV64I-NEXT:    call __floatunsisf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.uitofp.f32.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret float %1
+}
+declare float @llvm.experimental.constrained.uitofp.f32.i32(i32 %a, metadata, metadata)
+
+define float @fcvt_s_wu_load(i32* %p) nounwind strictfp {
+; RV32IF-LABEL: fcvt_s_wu_load:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    lw a0, 0(a0)
+; RV32IF-NEXT:    fcvt.s.wu ft0, a0
+; RV32IF-NEXT:    fmv.x.w a0, ft0
+; RV32IF-NEXT:    ret
+;
+; RV64IF-LABEL: fcvt_s_wu_load:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    lwu a0, 0(a0)
+; RV64IF-NEXT:    fcvt.s.wu ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_s_wu_load:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    lw a0, 0(a0)
+; RV32I-NEXT:    call __floatunsisf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_s_wu_load:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    lw a0, 0(a0)
+; RV64I-NEXT:    call __floatunsisf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %a = load i32, i32* %p
+  %1 = call float @llvm.experimental.constrained.uitofp.f32.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret float %1
+}
+
+define i64 @fcvt_l_s(float %a) nounwind strictfp {
+; RV32IF-LABEL: fcvt_l_s:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; 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: fcvt_l_s:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fmv.w.x ft0, a0
+; RV64IF-NEXT:    fcvt.l.s a0, ft0, rtz
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_l_s:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __fixsfdi@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_l_s:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __fixsfdi@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call i64 @llvm.experimental.constrained.fptosi.i64.f32(float %a, metadata !"fpexcept.strict") strictfp
+  ret i64 %1
+}
+declare i64 @llvm.experimental.constrained.fptosi.i64.f32(float, metadata)
+
+define i64 @fcvt_lu_s(float %a) nounwind strictfp {
+; RV32IF-LABEL: fcvt_lu_s:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; 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: fcvt_lu_s:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fmv.w.x ft0, a0
+; RV64IF-NEXT:    fcvt.lu.s a0, ft0, rtz
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_lu_s:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __fixunssfdi@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_lu_s:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __fixunssfdi@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call i64 @llvm.experimental.constrained.fptoui.i64.f32(float %a, metadata !"fpexcept.strict") strictfp
+  ret i64 %1
+}
+declare i64 @llvm.experimental.constrained.fptoui.i64.f32(float, metadata)
+
+define float @fcvt_s_l(i64 %a) nounwind strictfp {
+; RV32IF-LABEL: fcvt_s_l:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IF-NEXT:    call __floatdisf@plt
+; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+;
+; RV64IF-LABEL: fcvt_s_l:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fcvt.s.l ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_s_l:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatdisf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_s_l:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __floatdisf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.sitofp.f32.i64(i64 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret float %1
+}
+declare float @llvm.experimental.constrained.sitofp.f32.i64(i64, metadata, metadata)
+
+define float @fcvt_s_lu(i64 %a) nounwind strictfp {
+; RV32IF-LABEL: fcvt_s_lu:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IF-NEXT:    call __floatundisf@plt
+; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+;
+; RV64IF-LABEL: fcvt_s_lu:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fcvt.s.lu ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_s_lu:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatundisf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_s_lu:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __floatundisf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.uitofp.f32.i64(i64 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret float %1
+}
+declare float @llvm.experimental.constrained.uitofp.f32.i64(i64, metadata, metadata)
+
+define float @fcvt_s_w_i8(i8 signext %a) nounwind strictfp {
+; RV32IF-LABEL: fcvt_s_w_i8:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    fcvt.s.w ft0, a0
+; RV32IF-NEXT:    fmv.x.w a0, ft0
+; RV32IF-NEXT:    ret
+;
+; RV64IF-LABEL: fcvt_s_w_i8:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fcvt.s.w ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_s_w_i8:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatsisf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_s_w_i8:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __floatsisf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.sitofp.f32.i8(i8 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret float %1
+}
+declare float @llvm.experimental.constrained.sitofp.f32.i8(i8, metadata, metadata)
+
+define float @fcvt_s_wu_i8(i8 zeroext %a) nounwind strictfp {
+; RV32IF-LABEL: fcvt_s_wu_i8:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    fcvt.s.wu ft0, a0
+; RV32IF-NEXT:    fmv.x.w a0, ft0
+; RV32IF-NEXT:    ret
+;
+; RV64IF-LABEL: fcvt_s_wu_i8:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fcvt.s.wu ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_s_wu_i8:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatunsisf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_s_wu_i8:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __floatunsisf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.uitofp.f32.i8(i8 %a, metadata !"round.dynamic", metadata !"fpexcept.strict")
+  ret float %1
+}
+declare float @llvm.experimental.constrained.uitofp.f32.i8(i8, metadata, metadata)
+
+define float @fcvt_s_w_i16(i16 signext %a) nounwind strictfp {
+; RV32IF-LABEL: fcvt_s_w_i16:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    fcvt.s.w ft0, a0
+; RV32IF-NEXT:    fmv.x.w a0, ft0
+; RV32IF-NEXT:    ret
+;
+; RV64IF-LABEL: fcvt_s_w_i16:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fcvt.s.w ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_s_w_i16:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatsisf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_s_w_i16:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __floatsisf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.sitofp.f32.i16(i16 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret float %1
+}
+declare float @llvm.experimental.constrained.sitofp.f32.i16(i16, metadata, metadata)
+
+define float @fcvt_s_wu_i16(i16 zeroext %a) nounwind strictfp {
+; RV32IF-LABEL: fcvt_s_wu_i16:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    fcvt.s.wu ft0, a0
+; RV32IF-NEXT:    fmv.x.w a0, ft0
+; RV32IF-NEXT:    ret
+;
+; RV64IF-LABEL: fcvt_s_wu_i16:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fcvt.s.wu ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_s_wu_i16:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __floatunsisf@plt
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_s_wu_i16:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __floatunsisf@plt
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.uitofp.f32.i16(i16 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret float %1
+}
+declare float @llvm.experimental.constrained.uitofp.f32.i16(i16, metadata, metadata)
+
+; Make sure we select W version of addi on RV64.
+define signext i32 @fcvt_s_w_demanded_bits(i32 signext %0, float* %1) {
+; RV32IF-LABEL: fcvt_s_w_demanded_bits:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi a0, a0, 1
+; RV32IF-NEXT:    fcvt.s.w ft0, a0
+; RV32IF-NEXT:    fsw ft0, 0(a1)
+; RV32IF-NEXT:    ret
+;
+; RV64IF-LABEL: fcvt_s_w_demanded_bits:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    addiw a0, a0, 1
+; RV64IF-NEXT:    fcvt.s.w ft0, a0
+; RV64IF-NEXT:    fsw ft0, 0(a1)
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_s_w_demanded_bits:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    .cfi_def_cfa_offset 16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s1, 4(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    .cfi_offset ra, -4
+; RV32I-NEXT:    .cfi_offset s0, -8
+; RV32I-NEXT:    .cfi_offset s1, -12
+; RV32I-NEXT:    mv s0, a1
+; RV32I-NEXT:    addi s1, a0, 1
+; RV32I-NEXT:    mv a0, s1
+; RV32I-NEXT:    call __floatsisf@plt
+; RV32I-NEXT:    sw a0, 0(s0)
+; RV32I-NEXT:    mv a0, s1
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s1, 4(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_s_w_demanded_bits:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -32
+; RV64I-NEXT:    .cfi_def_cfa_offset 32
+; RV64I-NEXT:    sd ra, 24(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s0, 16(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s1, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    .cfi_offset ra, -8
+; RV64I-NEXT:    .cfi_offset s0, -16
+; RV64I-NEXT:    .cfi_offset s1, -24
+; RV64I-NEXT:    mv s0, a1
+; RV64I-NEXT:    addiw s1, a0, 1
+; RV64I-NEXT:    mv a0, s1
+; RV64I-NEXT:    call __floatsisf@plt
+; RV64I-NEXT:    sw a0, 0(s0)
+; RV64I-NEXT:    mv a0, s1
+; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 32
+; RV64I-NEXT:    ret
+  %3 = add i32 %0, 1
+  %4 = call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %3, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  store float %4, float* %1, align 4
+  ret i32 %3
+}
+
+; Make sure we select W version of addi on RV64.
+define signext i32 @fcvt_s_wu_demanded_bits(i32 signext %0, float* %1) {
+; RV32IF-LABEL: fcvt_s_wu_demanded_bits:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi a0, a0, 1
+; RV32IF-NEXT:    fcvt.s.wu ft0, a0
+; RV32IF-NEXT:    fsw ft0, 0(a1)
+; RV32IF-NEXT:    ret
+;
+; RV64IF-LABEL: fcvt_s_wu_demanded_bits:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    addiw a0, a0, 1
+; RV64IF-NEXT:    fcvt.s.wu ft0, a0
+; RV64IF-NEXT:    fsw ft0, 0(a1)
+; RV64IF-NEXT:    ret
+;
+; RV32I-LABEL: fcvt_s_wu_demanded_bits:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    .cfi_def_cfa_offset 16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s1, 4(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    .cfi_offset ra, -4
+; RV32I-NEXT:    .cfi_offset s0, -8
+; RV32I-NEXT:    .cfi_offset s1, -12
+; RV32I-NEXT:    mv s0, a1
+; RV32I-NEXT:    addi s1, a0, 1
+; RV32I-NEXT:    mv a0, s1
+; RV32I-NEXT:    call __floatunsisf@plt
+; RV32I-NEXT:    sw a0, 0(s0)
+; RV32I-NEXT:    mv a0, s1
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s1, 4(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcvt_s_wu_demanded_bits:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -32
+; RV64I-NEXT:    .cfi_def_cfa_offset 32
+; RV64I-NEXT:    sd ra, 24(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s0, 16(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s1, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    .cfi_offset ra, -8
+; RV64I-NEXT:    .cfi_offset s0, -16
+; RV64I-NEXT:    .cfi_offset s1, -24
+; RV64I-NEXT:    mv s0, a1
+; RV64I-NEXT:    addiw s1, a0, 1
+; RV64I-NEXT:    mv a0, s1
+; RV64I-NEXT:    call __floatunsisf@plt
+; RV64I-NEXT:    sw a0, 0(s0)
+; RV64I-NEXT:    mv a0, s1
+; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 32
+; RV64I-NEXT:    ret
+  %3 = add i32 %0, 1
+  %4 = call float @llvm.experimental.constrained.uitofp.f32.i32(i32 %3, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  store float %4, float* %1, align 4
+  ret i32 %3
+}

llvm/test/CodeGen/RISCV/half-convert-strict.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-zfh -verify-machineinstrs \
+; RUN:   -target-abi ilp32f -disable-strictnode-mutation < %s \
+; RUN:   | FileCheck -check-prefix=RV32IZFH %s
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-zfh -verify-machineinstrs \
+; RUN:   -target-abi lp64f -disable-strictnode-mutation < %s \
+; RUN:   | FileCheck -check-prefix=RV64IZFH %s
+; RUN: llc -mtriple=riscv32 -mattr=+d,+experimental-zfh -verify-machineinstrs \
+; RUN:   -target-abi ilp32d -disable-strictnode-mutation < %s \
+; RUN:   | FileCheck -check-prefix=RV32IDZFH %s
+; RUN: llc -mtriple=riscv64 -mattr=+d,+experimental-zfh -verify-machineinstrs \
+; RUN:   -target-abi lp64d -disable-strictnode-mutation < %s \
+; RUN:   | FileCheck -check-prefix=RV64IDZFH %s
+
+define i16 @fcvt_si_h(half %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_si_h:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rtz
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_si_h:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.l.h a0, fa0, rtz
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_si_h:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    fcvt.w.h a0, fa0, rtz
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_si_h:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.l.h a0, fa0, rtz
+; RV64IDZFH-NEXT:    ret
+  %1 = call i16 @llvm.experimental.constrained.fptosi.i16.f16(half %a, metadata !"fpexcept.strict") strictfp
+  ret i16 %1
+}
+declare i16 @llvm.experimental.constrained.fptosi.i16.f16(half, metadata)
+
+define i16 @fcvt_ui_h(half %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_ui_h:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    fcvt.wu.h a0, fa0, rtz
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_ui_h:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.lu.h a0, fa0, rtz
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_ui_h:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    fcvt.wu.h a0, fa0, rtz
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_ui_h:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.lu.h a0, fa0, rtz
+; RV64IDZFH-NEXT:    ret
+  %1 = call i16 @llvm.experimental.constrained.fptoui.i16.f16(half %a, metadata !"fpexcept.strict") strictfp
+  ret i16 %1
+}
+declare i16 @llvm.experimental.constrained.fptoui.i16.f16(half, metadata)
+
+define i32 @fcvt_w_h(half %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_w_h:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    fcvt.w.h a0, fa0, rtz
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_w_h:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.w.h a0, fa0, rtz
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_w_h:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    fcvt.w.h a0, fa0, rtz
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_w_h:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.w.h a0, fa0, rtz
+; RV64IDZFH-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptosi.i32.f16(half %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+declare i32 @llvm.experimental.constrained.fptosi.i32.f16(half, metadata)
+
+define i32 @fcvt_wu_h(half %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_wu_h:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    fcvt.wu.h a0, fa0, rtz
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_wu_h:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.wu.h a0, fa0, rtz
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_wu_h:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    fcvt.wu.h a0, fa0, rtz
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_wu_h:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.wu.h a0, fa0, rtz
+; RV64IDZFH-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptoui.i32.f16(half %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+declare i32 @llvm.experimental.constrained.fptoui.i32.f16(half, metadata)
+
+; Test where the fptoui has multiple uses, one of which causes a sext to be
+; inserted on RV64.
+; FIXME: We should not have an fcvt.wu.h and an fcvt.lu.h.
+define i32 @fcvt_wu_h_multiple_use(half %x, i32* %y) {
+; RV32IZFH-LABEL: fcvt_wu_h_multiple_use:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    fcvt.wu.h a1, fa0, rtz
+; RV32IZFH-NEXT:    li a0, 1
+; RV32IZFH-NEXT:    beqz a1, .LBB4_2
+; RV32IZFH-NEXT:  # %bb.1:
+; RV32IZFH-NEXT:    mv a0, a1
+; RV32IZFH-NEXT:  .LBB4_2:
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_wu_h_multiple_use:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.wu.h a1, fa0, rtz
+; RV64IZFH-NEXT:    li a0, 1
+; RV64IZFH-NEXT:    beqz a1, .LBB4_2
+; RV64IZFH-NEXT:  # %bb.1:
+; RV64IZFH-NEXT:    mv a0, a1
+; RV64IZFH-NEXT:  .LBB4_2:
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_wu_h_multiple_use:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    fcvt.wu.h a1, fa0, rtz
+; RV32IDZFH-NEXT:    li a0, 1
+; RV32IDZFH-NEXT:    beqz a1, .LBB4_2
+; RV32IDZFH-NEXT:  # %bb.1:
+; RV32IDZFH-NEXT:    mv a0, a1
+; RV32IDZFH-NEXT:  .LBB4_2:
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_wu_h_multiple_use:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.wu.h a1, fa0, rtz
+; RV64IDZFH-NEXT:    li a0, 1
+; RV64IDZFH-NEXT:    beqz a1, .LBB4_2
+; RV64IDZFH-NEXT:  # %bb.1:
+; RV64IDZFH-NEXT:    mv a0, a1
+; RV64IDZFH-NEXT:  .LBB4_2:
+; RV64IDZFH-NEXT:    ret
+  %a = call i32 @llvm.experimental.constrained.fptoui.i32.f16(half %x, metadata !"fpexcept.strict") strictfp
+  %b = icmp eq i32 %a, 0
+  %c = select i1 %b, i32 1, i32 %a
+  ret i32 %c
+}
+
+define i64 @fcvt_l_h(half %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_l_h:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    addi sp, sp, -16
+; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; 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: fcvt_l_h:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.l.h a0, fa0, rtz
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_l_h:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    addi sp, sp, -16
+; RV32IDZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IDZFH-NEXT:    call __fixhfdi@plt
+; RV32IDZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IDZFH-NEXT:    addi sp, sp, 16
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_l_h:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.l.h a0, fa0, rtz
+; RV64IDZFH-NEXT:    ret
+  %1 = call i64 @llvm.experimental.constrained.fptosi.i64.f16(half %a, metadata !"fpexcept.strict") strictfp
+  ret i64 %1
+}
+declare i64 @llvm.experimental.constrained.fptosi.i64.f16(half, metadata)
+
+define i64 @fcvt_lu_h(half %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_lu_h:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    addi sp, sp, -16
+; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; 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: fcvt_lu_h:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.lu.h a0, fa0, rtz
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_lu_h:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    addi sp, sp, -16
+; RV32IDZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IDZFH-NEXT:    call __fixunshfdi@plt
+; RV32IDZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IDZFH-NEXT:    addi sp, sp, 16
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_lu_h:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.lu.h a0, fa0, rtz
+; RV64IDZFH-NEXT:    ret
+  %1 = call i64 @llvm.experimental.constrained.fptoui.i64.f16(half %a, metadata !"fpexcept.strict") strictfp
+  ret i64 %1
+}
+declare i64 @llvm.experimental.constrained.fptoui.i64.f16(half, metadata)
+
+define half @fcvt_h_si(i16 %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_h_si:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    slli a0, a0, 16
+; RV32IZFH-NEXT:    srai a0, a0, 16
+; RV32IZFH-NEXT:    fcvt.h.w fa0, a0
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_h_si:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    slli a0, a0, 48
+; RV64IZFH-NEXT:    srai a0, a0, 48
+; RV64IZFH-NEXT:    fcvt.h.w fa0, a0
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_h_si:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    slli a0, a0, 16
+; RV32IDZFH-NEXT:    srai a0, a0, 16
+; RV32IDZFH-NEXT:    fcvt.h.w fa0, a0
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_h_si:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    slli a0, a0, 48
+; RV64IDZFH-NEXT:    srai a0, a0, 48
+; RV64IDZFH-NEXT:    fcvt.h.w fa0, a0
+; RV64IDZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.sitofp.f16.i16(i16 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}
+declare half @llvm.experimental.constrained.sitofp.f16.i16(i16, metadata, metadata)
+
+define half @fcvt_h_si_signext(i16 signext %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_h_si_signext:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    fcvt.h.w fa0, a0
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_h_si_signext:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.h.w fa0, a0
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_h_si_signext:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    fcvt.h.w fa0, a0
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_h_si_signext:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.h.w fa0, a0
+; RV64IDZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.sitofp.f16.i16(i16 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}
+
+define half @fcvt_h_ui(i16 %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_h_ui:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    lui a1, 16
+; RV32IZFH-NEXT:    addi a1, a1, -1
+; RV32IZFH-NEXT:    and a0, a0, a1
+; RV32IZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_h_ui:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    lui a1, 16
+; RV64IZFH-NEXT:    addiw a1, a1, -1
+; RV64IZFH-NEXT:    and a0, a0, a1
+; RV64IZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_h_ui:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    lui a1, 16
+; RV32IDZFH-NEXT:    addi a1, a1, -1
+; RV32IDZFH-NEXT:    and a0, a0, a1
+; RV32IDZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_h_ui:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    lui a1, 16
+; RV64IDZFH-NEXT:    addiw a1, a1, -1
+; RV64IDZFH-NEXT:    and a0, a0, a1
+; RV64IDZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV64IDZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.uitofp.f16.i16(i16 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}
+declare half @llvm.experimental.constrained.uitofp.f16.i16(i16, metadata, metadata)
+
+define half @fcvt_h_ui_zeroext(i16 zeroext %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_h_ui_zeroext:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_h_ui_zeroext:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_h_ui_zeroext:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_h_ui_zeroext:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV64IDZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.uitofp.f16.i16(i16 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}
+
+define half @fcvt_h_w(i32 %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_h_w:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    fcvt.h.w fa0, a0
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_h_w:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.h.w fa0, a0
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_h_w:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    fcvt.h.w fa0, a0
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_h_w:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.h.w fa0, a0
+; RV64IDZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}
+declare half @llvm.experimental.constrained.sitofp.f16.i32(i32, metadata, metadata)
+
+define half @fcvt_h_w_load(i32* %p) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_h_w_load:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    lw a0, 0(a0)
+; RV32IZFH-NEXT:    fcvt.h.w fa0, a0
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_h_w_load:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    lw a0, 0(a0)
+; RV64IZFH-NEXT:    fcvt.h.w fa0, a0
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_h_w_load:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    lw a0, 0(a0)
+; RV32IDZFH-NEXT:    fcvt.h.w fa0, a0
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_h_w_load:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    lw a0, 0(a0)
+; RV64IDZFH-NEXT:    fcvt.h.w fa0, a0
+; RV64IDZFH-NEXT:    ret
+  %a = load i32, i32* %p
+  %1 = call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}
+
+define half @fcvt_h_wu(i32 %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_h_wu:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_h_wu:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_h_wu:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_h_wu:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV64IDZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.uitofp.f16.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}
+declare half @llvm.experimental.constrained.uitofp.f16.i32(i32, metadata, metadata)
+
+define half @fcvt_h_wu_load(i32* %p) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_h_wu_load:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    lw a0, 0(a0)
+; RV32IZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_h_wu_load:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    lwu a0, 0(a0)
+; RV64IZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_h_wu_load:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    lw a0, 0(a0)
+; RV32IDZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_h_wu_load:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    lwu a0, 0(a0)
+; RV64IDZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV64IDZFH-NEXT:    ret
+  %a = load i32, i32* %p
+  %1 = call half @llvm.experimental.constrained.uitofp.f16.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}
+
+define half @fcvt_h_l(i64 %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_h_l:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    addi sp, sp, -16
+; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFH-NEXT:    call __floatdihf@plt
+; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IZFH-NEXT:    addi sp, sp, 16
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_h_l:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.h.l fa0, a0
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_h_l:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    addi sp, sp, -16
+; RV32IDZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IDZFH-NEXT:    call __floatdihf@plt
+; RV32IDZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IDZFH-NEXT:    addi sp, sp, 16
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_h_l:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.h.l fa0, a0
+; RV64IDZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.sitofp.f16.i64(i64 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}
+declare half @llvm.experimental.constrained.sitofp.f16.i64(i64, metadata, metadata)
+
+define half @fcvt_h_lu(i64 %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_h_lu:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    addi sp, sp, -16
+; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFH-NEXT:    call __floatundihf@plt
+; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IZFH-NEXT:    addi sp, sp, 16
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_h_lu:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.h.lu fa0, a0
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_h_lu:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    addi sp, sp, -16
+; RV32IDZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IDZFH-NEXT:    call __floatundihf@plt
+; RV32IDZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IDZFH-NEXT:    addi sp, sp, 16
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_h_lu:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.h.lu fa0, a0
+; RV64IDZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.uitofp.f16.i64(i64 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}
+declare half @llvm.experimental.constrained.uitofp.f16.i64(i64, metadata, metadata)
+
+define half @fcvt_h_s(float %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_h_s:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    fcvt.h.s fa0, fa0
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_h_s:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.h.s fa0, fa0
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_h_s:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    fcvt.h.s fa0, fa0
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_h_s:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.h.s fa0, fa0
+; RV64IDZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.fptrunc.f16.f32(float %a, metadata !"round.dynamic", metadata !"fpexcept.strict")
+  ret half %1
+}
+declare half @llvm.experimental.constrained.fptrunc.f16.f32(float, metadata, metadata)
+
+define float @fcvt_s_h(half %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_s_h:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    fcvt.s.h fa0, fa0
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_s_h:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.s.h fa0, fa0
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_s_h:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    fcvt.s.h fa0, fa0
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_s_h:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.s.h fa0, fa0
+; RV64IDZFH-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.fpext.f32.f16(half %a, metadata !"fpexcept.strict")
+  ret float %1
+}
+declare float @llvm.experimental.constrained.fpext.f32.f16(half, metadata)
+
+define half @fcvt_h_d(double %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_h_d:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    addi sp, sp, -16
+; RV32IZFH-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFH-NEXT:    call __truncdfhf2@plt
+; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IZFH-NEXT:    addi sp, sp, 16
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_h_d:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    addi sp, sp, -16
+; RV64IZFH-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFH-NEXT:    call __truncdfhf2@plt
+; RV64IZFH-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64IZFH-NEXT:    addi sp, sp, 16
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_h_d:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    fcvt.h.d fa0, fa0
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_h_d:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.h.d fa0, fa0
+; RV64IDZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.fptrunc.f16.f64(double %a, metadata !"round.dynamic", metadata !"fpexcept.strict")
+  ret half %1
+}
+declare half @llvm.experimental.constrained.fptrunc.f16.f64(double, metadata, metadata)
+
+define double @fcvt_d_h(half %a) nounwind strictfp {
+; RV32IZFH-LABEL: fcvt_d_h:
+; 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 __extendsfdf2@plt
+; RV32IZFH-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IZFH-NEXT:    addi sp, sp, 16
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_d_h:
+; 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 __extendsfdf2@plt
+; RV64IZFH-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64IZFH-NEXT:    addi sp, sp, 16
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_d_h:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    fcvt.d.h fa0, fa0
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_d_h:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    fcvt.d.h fa0, fa0
+; RV64IDZFH-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.fpext.f64.f16(half %a, metadata !"fpexcept.strict")
+  ret double %1
+}
+declare double @llvm.experimental.constrained.fpext.f64.f16(half, metadata)
+
+; Make sure we select W version of addi on RV64.
+define signext i32 @fcvt_h_w_demanded_bits(i32 signext %0, half* %1) {
+; RV32IZFH-LABEL: fcvt_h_w_demanded_bits:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    addi a0, a0, 1
+; RV32IZFH-NEXT:    fcvt.h.w ft0, a0
+; RV32IZFH-NEXT:    fsh ft0, 0(a1)
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_h_w_demanded_bits:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    addiw a0, a0, 1
+; RV64IZFH-NEXT:    fcvt.h.w ft0, a0
+; RV64IZFH-NEXT:    fsh ft0, 0(a1)
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_h_w_demanded_bits:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    addi a0, a0, 1
+; RV32IDZFH-NEXT:    fcvt.h.w ft0, a0
+; RV32IDZFH-NEXT:    fsh ft0, 0(a1)
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_h_w_demanded_bits:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    addiw a0, a0, 1
+; RV64IDZFH-NEXT:    fcvt.h.w ft0, a0
+; RV64IDZFH-NEXT:    fsh ft0, 0(a1)
+; RV64IDZFH-NEXT:    ret
+  %3 = add i32 %0, 1
+  %4 = call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %3, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  store half %4, half* %1, align 2
+  ret i32 %3
+}
+
+; Make sure we select W version of addi on RV64.
+define signext i32 @fcvt_h_wu_demanded_bits(i32 signext %0, half* %1) {
+; RV32IZFH-LABEL: fcvt_h_wu_demanded_bits:
+; RV32IZFH:       # %bb.0:
+; RV32IZFH-NEXT:    addi a0, a0, 1
+; RV32IZFH-NEXT:    fcvt.h.wu ft0, a0
+; RV32IZFH-NEXT:    fsh ft0, 0(a1)
+; RV32IZFH-NEXT:    ret
+;
+; RV64IZFH-LABEL: fcvt_h_wu_demanded_bits:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    addiw a0, a0, 1
+; RV64IZFH-NEXT:    fcvt.h.wu ft0, a0
+; RV64IZFH-NEXT:    fsh ft0, 0(a1)
+; RV64IZFH-NEXT:    ret
+;
+; RV32IDZFH-LABEL: fcvt_h_wu_demanded_bits:
+; RV32IDZFH:       # %bb.0:
+; RV32IDZFH-NEXT:    addi a0, a0, 1
+; RV32IDZFH-NEXT:    fcvt.h.wu ft0, a0
+; RV32IDZFH-NEXT:    fsh ft0, 0(a1)
+; RV32IDZFH-NEXT:    ret
+;
+; RV64IDZFH-LABEL: fcvt_h_wu_demanded_bits:
+; RV64IDZFH:       # %bb.0:
+; RV64IDZFH-NEXT:    addiw a0, a0, 1
+; RV64IDZFH-NEXT:    fcvt.h.wu ft0, a0
+; RV64IDZFH-NEXT:    fsh ft0, 0(a1)
+; RV64IDZFH-NEXT:    ret
+  %3 = add i32 %0, 1
+  %4 = call half @llvm.experimental.constrained.uitofp.f16.i32(i32 %3, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  store half %4, half* %1, align 2
+  ret i32 %3
+}

llvm/test/CodeGen/RISCV/rv64d-double-convert-strict.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv64 -mattr=+d -verify-machineinstrs < %s \
+; RUN:   -disable-strictnode-mutation | FileCheck %s -check-prefix=RV64ID
+
+; This file exhaustively checks double<->i32 conversions. In general,
+; fcvt.l[u].d can be selected instead of fcvt.w[u].d because poison is
+; generated for an fpto[s|u]i conversion if the result doesn't fit in the
+; target type.
+
+define i32 @aext_fptosi(double %a) nounwind strictfp {
+; RV64ID-LABEL: aext_fptosi:
+; RV64ID:       # %bb.0:
+; RV64ID-NEXT:    fmv.d.x ft0, a0
+; RV64ID-NEXT:    fcvt.w.d a0, ft0, rtz
+; RV64ID-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptosi.i32.f64(double %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+declare i32 @llvm.experimental.constrained.fptosi.i32.f64(double, metadata)
+
+define signext i32 @sext_fptosi(double %a) nounwind strictfp {
+; RV64ID-LABEL: sext_fptosi:
+; RV64ID:       # %bb.0:
+; RV64ID-NEXT:    fmv.d.x ft0, a0
+; RV64ID-NEXT:    fcvt.w.d a0, ft0, rtz
+; RV64ID-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptosi.i32.f64(double %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+
+define zeroext i32 @zext_fptosi(double %a) nounwind strictfp {
+; RV64ID-LABEL: zext_fptosi:
+; RV64ID:       # %bb.0:
+; RV64ID-NEXT:    fmv.d.x ft0, a0
+; RV64ID-NEXT:    fcvt.w.d a0, ft0, rtz
+; RV64ID-NEXT:    slli a0, a0, 32
+; RV64ID-NEXT:    srli a0, a0, 32
+; RV64ID-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptosi.i32.f64(double %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+
+define i32 @aext_fptoui(double %a) nounwind strictfp {
+; RV64ID-LABEL: aext_fptoui:
+; RV64ID:       # %bb.0:
+; RV64ID-NEXT:    fmv.d.x ft0, a0
+; RV64ID-NEXT:    fcvt.wu.d a0, ft0, rtz
+; RV64ID-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptoui.i32.f64(double %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+declare i32 @llvm.experimental.constrained.fptoui.i32.f64(double, metadata)
+
+define signext i32 @sext_fptoui(double %a) nounwind strictfp {
+; RV64ID-LABEL: sext_fptoui:
+; RV64ID:       # %bb.0:
+; RV64ID-NEXT:    fmv.d.x ft0, a0
+; RV64ID-NEXT:    fcvt.wu.d a0, ft0, rtz
+; RV64ID-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptoui.i32.f64(double %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+
+define zeroext i32 @zext_fptoui(double %a) nounwind strictfp {
+; RV64ID-LABEL: zext_fptoui:
+; RV64ID:       # %bb.0:
+; RV64ID-NEXT:    fmv.d.x ft0, a0
+; RV64ID-NEXT:    fcvt.lu.d a0, ft0, rtz
+; RV64ID-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptoui.i32.f64(double %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+
+define double @uitofp_aext_i32_to_f64(i32 %a) nounwind strictfp {
+; RV64ID-LABEL: uitofp_aext_i32_to_f64:
+; RV64ID:       # %bb.0:
+; RV64ID-NEXT:    fcvt.d.wu ft0, a0
+; RV64ID-NEXT:    fmv.x.d a0, ft0
+; RV64ID-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}
+declare double @llvm.experimental.constrained.uitofp.f64.i32(i32 %a, metadata, metadata)
+
+define double @uitofp_sext_i32_to_f64(i32 signext %a) nounwind strictfp {
+; RV64ID-LABEL: uitofp_sext_i32_to_f64:
+; RV64ID:       # %bb.0:
+; RV64ID-NEXT:    fcvt.d.wu ft0, a0
+; RV64ID-NEXT:    fmv.x.d a0, ft0
+; RV64ID-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}
+
+define double @uitofp_zext_i32_to_f64(i32 zeroext %a) nounwind strictfp {
+; RV64ID-LABEL: uitofp_zext_i32_to_f64:
+; RV64ID:       # %bb.0:
+; RV64ID-NEXT:    fcvt.d.wu ft0, a0
+; RV64ID-NEXT:    fmv.x.d a0, ft0
+; RV64ID-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}
+
+define double @sitofp_aext_i32_to_f64(i32 %a) nounwind strictfp {
+; RV64ID-LABEL: sitofp_aext_i32_to_f64:
+; RV64ID:       # %bb.0:
+; RV64ID-NEXT:    fcvt.d.w ft0, a0
+; RV64ID-NEXT:    fmv.x.d a0, ft0
+; RV64ID-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.sitofp.f64.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}
+declare double @llvm.experimental.constrained.sitofp.f64.i32(i32 %a, metadata, metadata)
+
+define double @sitofp_sext_i32_to_f64(i32 signext %a) nounwind strictfp {
+; RV64ID-LABEL: sitofp_sext_i32_to_f64:
+; RV64ID:       # %bb.0:
+; RV64ID-NEXT:    fcvt.d.w ft0, a0
+; RV64ID-NEXT:    fmv.x.d a0, ft0
+; RV64ID-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.sitofp.f64.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}
+
+define double @sitofp_zext_i32_to_f64(i32 zeroext %a) nounwind strictfp {
+; RV64ID-LABEL: sitofp_zext_i32_to_f64:
+; RV64ID:       # %bb.0:
+; RV64ID-NEXT:    fcvt.d.w ft0, a0
+; RV64ID-NEXT:    fmv.x.d a0, ft0
+; RV64ID-NEXT:    ret
+  %1 = call double @llvm.experimental.constrained.sitofp.f64.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret double %1
+}

llvm/test/CodeGen/RISCV/rv64f-float-convert-strict.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv64 -mattr=+f -verify-machineinstrs < %s \
+; RUN:   -disable-strictnode-mutation | FileCheck %s -check-prefix=RV64IF
+
+; This file exhaustively checks float<->i32 conversions. In general,
+; fcvt.l[u].s can be selected instead of fcvt.w[u].s because poison is
+; generated for an fpto[s|u]i conversion if the result doesn't fit in the
+; target type.
+
+define i32 @aext_fptosi(float %a) nounwind strictfp {
+; RV64IF-LABEL: aext_fptosi:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fmv.w.x ft0, a0
+; RV64IF-NEXT:    fcvt.w.s a0, ft0, rtz
+; RV64IF-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptosi.i32.f32(float %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+declare i32 @llvm.experimental.constrained.fptosi.i32.f32(float, metadata)
+
+define signext i32 @sext_fptosi(float %a) nounwind strictfp {
+; RV64IF-LABEL: sext_fptosi:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fmv.w.x ft0, a0
+; RV64IF-NEXT:    fcvt.w.s a0, ft0, rtz
+; RV64IF-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptosi.i32.f32(float %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+
+define zeroext i32 @zext_fptosi(float %a) nounwind strictfp {
+; RV64IF-LABEL: zext_fptosi:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fmv.w.x ft0, a0
+; RV64IF-NEXT:    fcvt.w.s a0, ft0, rtz
+; RV64IF-NEXT:    slli a0, a0, 32
+; RV64IF-NEXT:    srli a0, a0, 32
+; RV64IF-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptosi.i32.f32(float %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+
+define i32 @aext_fptoui(float %a) nounwind strictfp {
+; RV64IF-LABEL: aext_fptoui:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fmv.w.x ft0, a0
+; RV64IF-NEXT:    fcvt.wu.s a0, ft0, rtz
+; RV64IF-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptoui.i32.f32(float %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+declare i32 @llvm.experimental.constrained.fptoui.i32.f32(float, metadata)
+
+define signext i32 @sext_fptoui(float %a) nounwind strictfp {
+; RV64IF-LABEL: sext_fptoui:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fmv.w.x ft0, a0
+; RV64IF-NEXT:    fcvt.wu.s a0, ft0, rtz
+; RV64IF-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptoui.i32.f32(float %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+
+define zeroext i32 @zext_fptoui(float %a) nounwind strictfp {
+; RV64IF-LABEL: zext_fptoui:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fmv.w.x ft0, a0
+; RV64IF-NEXT:    fcvt.lu.s a0, ft0, rtz
+; RV64IF-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptoui.i32.f32(float %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+
+define float @uitofp_aext_i32_to_f32(i32 %a) nounwind strictfp {
+; RV64IF-LABEL: uitofp_aext_i32_to_f32:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fcvt.s.wu ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.uitofp.f32.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret float %1
+}
+declare float @llvm.experimental.constrained.uitofp.f32.i32(i32 %a, metadata, metadata)
+
+define float @uitofp_sext_i32_to_f32(i32 signext %a) nounwind strictfp {
+; RV64IF-LABEL: uitofp_sext_i32_to_f32:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fcvt.s.wu ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.uitofp.f32.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret float %1
+}
+
+define float @uitofp_zext_i32_to_f32(i32 zeroext %a) nounwind strictfp {
+; RV64IF-LABEL: uitofp_zext_i32_to_f32:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fcvt.s.wu ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.uitofp.f32.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret float %1
+}
+
+define float @sitofp_aext_i32_to_f32(i32 %a) nounwind strictfp {
+; RV64IF-LABEL: sitofp_aext_i32_to_f32:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fcvt.s.w ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret float %1
+}
+declare float @llvm.experimental.constrained.sitofp.f32.i32(i32, metadata, metadata)
+
+define float @sitofp_sext_i32_to_f32(i32 signext %a) nounwind strictfp {
+; RV64IF-LABEL: sitofp_sext_i32_to_f32:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fcvt.s.w ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret float %1
+}
+
+define float @sitofp_zext_i32_to_f32(i32 zeroext %a) nounwind strictfp {
+; RV64IF-LABEL: sitofp_zext_i32_to_f32:
+; RV64IF:       # %bb.0:
+; RV64IF-NEXT:    fcvt.s.w ft0, a0
+; RV64IF-NEXT:    fmv.x.w a0, ft0
+; RV64IF-NEXT:    ret
+  %1 = call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret float %1
+}

llvm/test/CodeGen/RISCV/rv64f-half-convert-strict.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-zfh -verify-machineinstrs \
+; RUN:   -target-abi lp64f -disable-strictnode-mutation < %s | \
+; RUN:   FileCheck %s -check-prefix=RV64IZFH
+
+; This file exhaustively checks half<->i32 conversions. In general,
+; fcvt.l[u].h can be selected instead of fcvt.w[u].h because poison is
+; generated for an fpto[s|u]i conversion if the result doesn't fit in the
+; target type.
+
+define i32 @aext_fptosi(half %a) nounwind {
+; RV64IZFH-LABEL: aext_fptosi:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.w.h a0, fa0, rtz
+; RV64IZFH-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptosi.i32.f16(half %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+declare i32 @llvm.experimental.constrained.fptosi.i32.f16(half, metadata)
+
+define signext i32 @sext_fptosi(half %a) nounwind {
+; RV64IZFH-LABEL: sext_fptosi:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.w.h a0, fa0, rtz
+; RV64IZFH-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptosi.i32.f16(half %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+
+define zeroext i32 @zext_fptosi(half %a) nounwind {
+; RV64IZFH-LABEL: zext_fptosi:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.w.h a0, fa0, rtz
+; RV64IZFH-NEXT:    slli a0, a0, 32
+; RV64IZFH-NEXT:    srli a0, a0, 32
+; RV64IZFH-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptosi.i32.f16(half %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+
+define i32 @aext_fptoui(half %a) nounwind {
+; RV64IZFH-LABEL: aext_fptoui:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.wu.h a0, fa0, rtz
+; RV64IZFH-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptoui.i32.f16(half %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+declare i32 @llvm.experimental.constrained.fptoui.i32.f16(half, metadata)
+
+define signext i32 @sext_fptoui(half %a) nounwind {
+; RV64IZFH-LABEL: sext_fptoui:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.wu.h a0, fa0, rtz
+; RV64IZFH-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptoui.i32.f16(half %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+
+define zeroext i32 @zext_fptoui(half %a) nounwind {
+; RV64IZFH-LABEL: zext_fptoui:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.lu.h a0, fa0, rtz
+; RV64IZFH-NEXT:    ret
+  %1 = call i32 @llvm.experimental.constrained.fptoui.i32.f16(half %a, metadata !"fpexcept.strict") strictfp
+  ret i32 %1
+}
+
+define half @uitofp_aext_i32_to_f16(i32 %a) nounwind {
+; RV64IZFH-LABEL: uitofp_aext_i32_to_f16:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV64IZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.uitofp.f16.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}
+declare half @llvm.experimental.constrained.uitofp.f16.i32(i32 %a, metadata, metadata)
+
+define half @uitofp_sext_i32_to_f16(i32 signext %a) nounwind {
+; RV64IZFH-LABEL: uitofp_sext_i32_to_f16:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV64IZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.uitofp.f16.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}
+
+define half @uitofp_zext_i32_to_f16(i32 zeroext %a) nounwind {
+; RV64IZFH-LABEL: uitofp_zext_i32_to_f16:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.h.wu fa0, a0
+; RV64IZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.uitofp.f16.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}
+
+define half @sitofp_aext_i32_to_f16(i32 %a) nounwind {
+; RV64IZFH-LABEL: sitofp_aext_i32_to_f16:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.h.w fa0, a0
+; RV64IZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}
+declare half @llvm.experimental.constrained.sitofp.f16.i32(i32 %a, metadata, metadata)
+
+define half @sitofp_sext_i32_to_f16(i32 signext %a) nounwind {
+; RV64IZFH-LABEL: sitofp_sext_i32_to_f16:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.h.w fa0, a0
+; RV64IZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}
+
+define half @sitofp_zext_i32_to_f16(i32 zeroext %a) nounwind {
+; RV64IZFH-LABEL: sitofp_zext_i32_to_f16:
+; RV64IZFH:       # %bb.0:
+; RV64IZFH-NEXT:    fcvt.h.w fa0, a0
+; RV64IZFH-NEXT:    ret
+  %1 = call half @llvm.experimental.constrained.sitofp.f16.i32(i32 %a, metadata !"round.dynamic", metadata !"fpexcept.strict") strictfp
+  ret half %1
+}