[RISCV] Add rvv codegen support for vp.fptrunc.

[jacquesguan]

Summary

This patch adds rvv codegen support for vp.fptrunc. The lowering of fp_round and vp.fptrunc share most code so use a common lowering function to handle those two, similar to vp.trunc.

[frasercrmck]

The incongruousness of lowering vp.fptrunc to VP_FPTRUNC but calling a "RoundLike" function was a bit confusing. I opened D123847 to try and make them more consistent.

[frasercrmck]

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

4438

We're returning here after already creating the VL ops. Ideally we'd only create nodes once we know we're going to commit to returning something. This is a smaller issue in the original code but I don't think it needs to be like this. The early-exit check isn't dependent on any previous computations.

[jacquesguan]

address comment.

[jacquesguan]

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

4438

Done, I moved it to the begin of the function.

[craig.topper]

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

3284

This check wasn't in the original code. Why is it needed now?

[jacquesguan]

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

3284

There is a branch that will just return Op when Op is not a vector in the original code.

if (!VT.isVector() || VT.getVectorElementType() != MVT::f16 ||
        SrcVT.getVectorElementType() != MVT::f64) {
      // For scalable vectors, we only need to close the gap between
      // vXf64<->vXf16.
      if (!VT.isFixedLengthVector())
        return Op;

...

}

These two ifs contain the no vector type situation, I split it out because this new lowering function only focus the vector FP round.

[jacquesguan]

rebase

[jacquesguan]

fix test.

[craig.topper]

LGTM

llvm/lib/Target/RISCV/RISCVISelLowering.h

private:
   SDValue lowerShiftLeftParts(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerShiftRightParts(SDValue Op, SelectionDAG &DAG, bool IsSRA) const;
   SDValue lowerSPLAT_VECTOR_PARTS(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerVectorMaskSplat(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerVectorMaskExt(SDValue Op, SelectionDAG &DAG,
                              int64_t ExtTrueVal) const;
   SDValue lowerVectorMaskTruncLike(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerVectorTruncLike(SDValue Op, SelectionDAG &DAG) const;
+  SDValue lowerVectorFPRoundLike(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerINTRINSIC_VOID(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerVPREDUCE(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerVECREDUCE(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerVectorMaskVecReduction(SDValue Op, SelectionDAG &DAG,

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

if (Subtarget.hasVInstructions()) {
     static const unsigned FloatingPointVPOps[] = {
         ISD::VP_FADD,        ISD::VP_FSUB,
         ISD::VP_FMUL,        ISD::VP_FDIV,
         ISD::VP_FNEG,        ISD::VP_FMA,
         ISD::VP_REDUCE_FADD, ISD::VP_REDUCE_SEQ_FADD,
         ISD::VP_REDUCE_FMIN, ISD::VP_REDUCE_FMAX,
         ISD::VP_MERGE,       ISD::VP_SELECT,
         ISD::VP_SITOFP,      ISD::VP_UITOFP,
-        ISD::VP_SETCC};
+        ISD::VP_SETCC,       ISD::VP_FPTRUNC};
 
     if (!Subtarget.is64Bit()) {
       // We must custom-lower certain vXi64 operations on RV32 due to the vector
       // element type being illegal.
       setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::i64, Custom);
       setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::i64, Custom);
 
       setOperationAction(ISD::VECREDUCE_ADD, MVT::i64, Custom);
SDValue IntermediateExtend = getRVVFPExtendOrRound(
         Src, InterVT, InterContainerVT, DL, DAG, Subtarget);
 
     SDValue Extend = getRVVFPExtendOrRound(IntermediateExtend, VT, ContainerVT,
                                            DL, DAG, Subtarget);
     if (VT.isFixedLengthVector())
       return convertFromScalableVector(VT, Extend, DAG, Subtarget);
     return Extend;
   }
-  case ISD::FP_ROUND: {
-    // RVV can only do fp_round to types half the size as the source. We
-    // custom-lower f64->f16 rounds via RVV's round-to-odd float
-    // conversion instruction.
-    SDLoc DL(Op);
-    MVT VT = Op.getSimpleValueType();
-    SDValue Src = Op.getOperand(0);
-    MVT SrcVT = Src.getSimpleValueType();
-
-    // Prepare any fixed-length vector operands.
-    MVT ContainerVT = VT;
-    if (VT.isFixedLengthVector()) {
-      MVT SrcContainerVT = getContainerForFixedLengthVector(SrcVT);
-      ContainerVT =
-          SrcContainerVT.changeVectorElementType(VT.getVectorElementType());
-      Src = convertToScalableVector(SrcContainerVT, Src, DAG, Subtarget);
-    }
-
-    if (!VT.isVector() || VT.getVectorElementType() != MVT::f16 ||
-        SrcVT.getVectorElementType() != MVT::f64) {
-      // For scalable vectors, we only need to close the gap between
-      // vXf64<->vXf16.
-      if (!VT.isFixedLengthVector())
-        return Op;
-      // For fixed-length vectors, lower the FP_ROUND to a custom "VL" version.
-      Src = getRVVFPExtendOrRound(Src, VT, ContainerVT, DL, DAG, Subtarget);
-      return convertFromScalableVector(VT, Src, DAG, Subtarget);
-    }
-
-    SDValue Mask, VL;
-    std::tie(Mask, VL) = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
-
-    MVT InterVT = ContainerVT.changeVectorElementType(MVT::f32);
-    SDValue IntermediateRound =
-        DAG.getNode(RISCVISD::VFNCVT_ROD_VL, DL, InterVT, Src, Mask, VL);
-    SDValue Round = getRVVFPExtendOrRound(IntermediateRound, VT, ContainerVT,
-                                          DL, DAG, Subtarget);
-
-    if (VT.isFixedLengthVector())
-      return convertFromScalableVector(VT, Round, DAG, Subtarget);
-    return Round;
-  }
+  case ISD::FP_ROUND:
+    if (!Op.getValueType().isVector())
+    return lowerVectorFPRoundLike(Op, DAG);
   case ISD::FP_TO_SINT:
   case ISD::FP_TO_UINT:
   case ISD::SINT_TO_FP:
   case ISD::UINT_TO_FP: {
     // RVV can only do fp<->int conversions to types half/double the size as
     // the source. We custom-lower any conversions that do two hops into
     // sequences.
     MVT VT = Op.getSimpleValueType();
SDValue RISCVTargetLowering::LowerOperation(SDValue Op,
   case ISD::VP_ZEXT:
     if (Op.getOperand(0).getSimpleValueType().getVectorElementType() == MVT::i1)
       return lowerVPExtMaskOp(Op, DAG);
     return lowerVPOp(Op, DAG,
                      Op.getOpcode() == ISD::VP_SEXT ? RISCVISD::VSEXT_VL
                                                     : RISCVISD::VZEXT_VL);
   case ISD::VP_TRUNC:
     return lowerVectorTruncLike(Op, DAG);
+  case ISD::VP_FPTRUNC:
+    return lowerVectorFPRoundLike(Op, DAG);
   case ISD::VP_FPTOSI:
     return lowerVPFPIntConvOp(Op, DAG, RISCVISD::FP_TO_SINT_VL);
   case ISD::VP_FPTOUI:
     return lowerVPFPIntConvOp(Op, DAG, RISCVISD::FP_TO_UINT_VL);
   case ISD::VP_SITOFP:
     return lowerVPFPIntConvOp(Op, DAG, RISCVISD::SINT_TO_FP_VL);
   case ISD::VP_UITOFP:
     return lowerVPFPIntConvOp(Op, DAG, RISCVISD::UINT_TO_FP_VL);
SDValue RISCVTargetLowering::lowerVectorTruncLike(SDValue Op,
   } while (SrcEltVT != DstEltVT);
 
   if (SrcVT.isFixedLengthVector())
     Result = convertFromScalableVector(VT, Result, DAG, Subtarget);
 
   return Result;
 }
 
+SDValue RISCVTargetLowering::lowerVectorFPRoundLike(SDValue Op,
+                                                    SelectionDAG &DAG) const {
+  bool IsVPFPTrunc = Op.getOpcode() == ISD::VP_FPTRUNC;
+  // RVV can only do truncate fp to types half the size as the source. We
+  // custom-lower f64->f16 rounds via RVV's round-to-odd float
+  // conversion instruction.
+  SDLoc DL(Op);
+  MVT VT = Op.getSimpleValueType();
+
+  assert(VT.isVector() && "Unexpected type for vector truncate lowering");
+
+  SDValue Src = Op.getOperand(0);
+  MVT SrcVT = Src.getSimpleValueType();
+
+  // Prepare any fixed-length vector operands.
+  MVT ContainerVT = VT;
+  SDValue Mask, VL;
+  if (IsVPFPTrunc) {
+    Mask = Op.getOperand(1);
+    VL = Op.getOperand(2);
+  }
+  if (VT.isFixedLengthVector()) {
+    MVT SrcContainerVT = getContainerForFixedLengthVector(SrcVT);
+    ContainerVT =
+        SrcContainerVT.changeVectorElementType(VT.getVectorElementType());
+    Src = convertToScalableVector(SrcContainerVT, Src, DAG, Subtarget);
+    if (IsVPFPTrunc) {
+      MVT MaskVT =
+          MVT::getVectorVT(MVT::i1, ContainerVT.getVectorElementCount());
+      Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
+    }
+  }
+
+  if (!IsVPFPTrunc)
+    std::tie(Mask, VL) =
+        getDefaultVLOps(SrcVT, ContainerVT, DL, DAG, Subtarget);
+
+  if (VT.getVectorElementType() != MVT::f16 ||
+      SrcVT.getVectorElementType() != MVT::f64) {
+    // For FP_ROUND of scalable vectors, leave it to the pattern.
+    if (!VT.isFixedLengthVector() && !IsVPFPTrunc)
+      return Op;
+    Src = DAG.getNode(RISCVISD::FP_ROUND_VL, DL, ContainerVT, Src, Mask, VL);
+    if (VT.isFixedLengthVector())
+      Src = convertFromScalableVector(VT, Src, DAG, Subtarget);
+    return Src;
+  }
+
+  MVT InterVT = ContainerVT.changeVectorElementType(MVT::f32);
+  SDValue IntermediateRound =
+      DAG.getNode(RISCVISD::VFNCVT_ROD_VL, DL, InterVT, Src, Mask, VL);
+  SDValue Round = DAG.getNode(RISCVISD::FP_ROUND_VL, DL, ContainerVT,
+                              IntermediateRound, Mask, VL);
+  if (VT.isFixedLengthVector())
+    return convertFromScalableVector(VT, Round, DAG, Subtarget);
+  return Round;
+}
+
 // Custom-legalize INSERT_VECTOR_ELT so that the value is inserted into the
 // first position of a vector, and that vector is slid up to the insert index.
 // By limiting the active vector length to index+1 and merging with the
 // original vector (with an undisturbed tail policy for elements >= VL), we
 // achieve the desired result of leaving all elements untouched except the one
 // at VL-1, which is replaced with the desired value.
 SDValue RISCVTargetLowering::lowerINSERT_VECTOR_ELT(SDValue Op,
                                                     SelectionDAG &DAG) const {

llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td

def : Pat<(fvti.Vector (riscv_fpround_vl (fwti.Vector fwti.RegClass:$rs1),
                   (fvti.Vector (IMPLICIT_DEF)), fwti.RegClass:$rs1,
                   (fwti.Mask V0), GPR:$vl, fvti.Log2SEW, TAIL_AGNOSTIC)>;
 
     def : Pat<(fvti.Vector (riscv_fncvt_rod_vl (fwti.Vector fwti.RegClass:$rs1),
                                                (fwti.Mask true_mask),
                                                VLOpFrag)),
               (!cast<Instruction>("PseudoVFNCVT_ROD_F_F_W_"#fvti.LMul.MX)
                   fwti.RegClass:$rs1, GPR:$vl, fvti.Log2SEW)>;
-  }
+
+    def : Pat<(fvti.Vector (riscv_fncvt_rod_vl (fwti.Vector fwti.RegClass:$rs1),
+                                               (fwti.Mask V0),
+                                               VLOpFrag)),
+              (!cast<Instruction>("PseudoVFNCVT_ROD_F_F_W_"#fvti.LMul.MX#"_MASK")
+                  (fvti.Vector (IMPLICIT_DEF)), fwti.RegClass:$rs1,
+                  (fwti.Mask V0), GPR:$vl, fvti.Log2SEW, TAIL_AGNOSTIC)>;  }
 }
 
 } // Predicates = [HasVInstructionsAnyF]
 
 let Predicates = [HasVInstructions] in {
 
 foreach mti = AllMasks in {
   // 16.1 Vector Mask-Register Logical Instructions

llvm/test/CodeGen/RISCV/rvv/fixed-vector-fptrunc-vp.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+d,+zfh,+experimental-zvfh,+v -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=riscv64 -mattr=+d,+zfh,+experimental-zvfh,+v -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s
+
+declare <2 x half> @llvm.vp.fptrunc.nxv2f16.nxv2f32(<2 x float>, <2 x i1>, i32)
+
+define <2 x half> @vfptrunc_nxv2f16_nxv2f32(<2 x float> %a, <2 x i1> %m, i32 zeroext %vl) {
+; CHECK-LABEL: vfptrunc_nxv2f16_nxv2f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e16, mf4, ta, mu
+; CHECK-NEXT:    vfncvt.f.f.w v9, v8, v0.t
+; CHECK-NEXT:    vmv1r.v v8, v9
+; CHECK-NEXT:    ret
+  %v = call <2 x half> @llvm.vp.fptrunc.nxv2f16.nxv2f32(<2 x float> %a, <2 x i1> %m, i32 %vl)
+  ret <2 x half> %v
+}
+
+define <2 x half> @vfptrunc_nxv2f16_nxv2f32_unmasked(<2 x float> %a, i32 zeroext %vl) {
+; CHECK-LABEL: vfptrunc_nxv2f16_nxv2f32_unmasked:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e16, mf4, ta, mu
+; CHECK-NEXT:    vfncvt.f.f.w v9, v8
+; CHECK-NEXT:    vmv1r.v v8, v9
+; CHECK-NEXT:    ret
+  %v = call <2 x half> @llvm.vp.fptrunc.nxv2f16.nxv2f32(<2 x float> %a, <2 x i1> shufflevector (<2 x i1> insertelement (<2 x i1> undef, i1 true, i32 0), <2 x i1> undef, <2 x i32> zeroinitializer), i32 %vl)
+  ret <2 x half> %v
+}
+
+declare <2 x half> @llvm.vp.fptrunc.nxv2f16.nxv2f64(<2 x double>, <2 x i1>, i32)
+
+define <2 x half> @vfptrunc_nxv2f16_nxv2f64(<2 x double> %a, <2 x i1> %m, i32 zeroext %vl) {
+; CHECK-LABEL: vfptrunc_nxv2f16_nxv2f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e32, mf2, ta, mu
+; CHECK-NEXT:    vfncvt.rod.f.f.w v9, v8, v0.t
+; CHECK-NEXT:    vsetvli zero, zero, e16, mf4, ta, mu
+; CHECK-NEXT:    vfncvt.f.f.w v8, v9, v0.t
+; CHECK-NEXT:    ret
+  %v = call <2 x half> @llvm.vp.fptrunc.nxv2f16.nxv2f64(<2 x double> %a, <2 x i1> %m, i32 %vl)
+  ret <2 x half> %v
+}
+
+define <2 x half> @vfptrunc_nxv2f16_nxv2f64_unmasked(<2 x double> %a, i32 zeroext %vl) {
+; CHECK-LABEL: vfptrunc_nxv2f16_nxv2f64_unmasked:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e32, mf2, ta, mu
+; CHECK-NEXT:    vfncvt.rod.f.f.w v9, v8
+; CHECK-NEXT:    vsetvli zero, zero, e16, mf4, ta, mu
+; CHECK-NEXT:    vfncvt.f.f.w v8, v9
+; CHECK-NEXT:    ret
+  %v = call <2 x half> @llvm.vp.fptrunc.nxv2f16.nxv2f64(<2 x double> %a, <2 x i1> shufflevector (<2 x i1> insertelement (<2 x i1> undef, i1 true, i32 0), <2 x i1> undef, <2 x i32> zeroinitializer), i32 %vl)
+  ret <2 x half> %v
+}
+
+declare <2 x float> @llvm.vp.fptrunc.nxv2f64.nxv2f32(<2 x double>, <2 x i1>, i32)
+
+define <2 x float> @vfptrunc_nxv2f32_nxv2f64(<2 x double> %a, <2 x i1> %m, i32 zeroext %vl) {
+; CHECK-LABEL: vfptrunc_nxv2f32_nxv2f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e32, mf2, ta, mu
+; CHECK-NEXT:    vfncvt.f.f.w v9, v8, v0.t
+; CHECK-NEXT:    vmv1r.v v8, v9
+; CHECK-NEXT:    ret
+  %v = call <2 x float> @llvm.vp.fptrunc.nxv2f64.nxv2f32(<2 x double> %a, <2 x i1> %m, i32 %vl)
+  ret <2 x float> %v
+}
+
+define <2 x float> @vfptrunc_nxv2f32_nxv2f64_unmasked(<2 x double> %a, i32 zeroext %vl) {
+; CHECK-LABEL: vfptrunc_nxv2f32_nxv2f64_unmasked:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e32, mf2, ta, mu
+; CHECK-NEXT:    vfncvt.f.f.w v9, v8
+; CHECK-NEXT:    vmv1r.v v8, v9
+; CHECK-NEXT:    ret
+  %v = call <2 x float> @llvm.vp.fptrunc.nxv2f64.nxv2f32(<2 x double> %a, <2 x i1> shufflevector (<2 x i1> insertelement (<2 x i1> undef, i1 true, i32 0), <2 x i1> undef, <2 x i32> zeroinitializer), i32 %vl)
+  ret <2 x float> %v
+}

llvm/test/CodeGen/RISCV/rvv/vfptrunc-vp.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+d,+zfh,+experimental-zvfh,+v -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=riscv64 -mattr=+d,+zfh,+experimental-zvfh,+v -verify-machineinstrs < %s | FileCheck %s
+
+declare <vscale x 2 x half> @llvm.vp.fptrunc.nxv2f16.nxv2f32(<vscale x 2 x float>, <vscale x 2 x i1>, i32)
+
+define <vscale x 2 x half> @vfptrunc_nxv2f16_nxv2f32(<vscale x 2 x float> %a, <vscale x 2 x i1> %m, i32 zeroext %vl) {
+; CHECK-LABEL: vfptrunc_nxv2f16_nxv2f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e16, mf2, ta, mu
+; CHECK-NEXT:    vfncvt.f.f.w v9, v8, v0.t
+; CHECK-NEXT:    vmv1r.v v8, v9
+; CHECK-NEXT:    ret
+  %v = call <vscale x 2 x half> @llvm.vp.fptrunc.nxv2f16.nxv2f32(<vscale x 2 x float> %a, <vscale x 2 x i1> %m, i32 %vl)
+  ret <vscale x 2 x half> %v
+}
+
+define <vscale x 2 x half> @vfptrunc_nxv2f16_nxv2f32_unmasked(<vscale x 2 x float> %a, i32 zeroext %vl) {
+; CHECK-LABEL: vfptrunc_nxv2f16_nxv2f32_unmasked:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e16, mf2, ta, mu
+; CHECK-NEXT:    vfncvt.f.f.w v9, v8
+; CHECK-NEXT:    vmv1r.v v8, v9
+; CHECK-NEXT:    ret
+  %v = call <vscale x 2 x half> @llvm.vp.fptrunc.nxv2f16.nxv2f32(<vscale x 2 x float> %a, <vscale x 2 x i1> shufflevector (<vscale x 2 x i1> insertelement (<vscale x 2 x i1> undef, i1 true, i32 0), <vscale x 2 x i1> undef, <vscale x 2 x i32> zeroinitializer), i32 %vl)
+  ret <vscale x 2 x half> %v
+}
+
+declare <vscale x 2 x half> @llvm.vp.fptrunc.nxv2f16.nxv2f64(<vscale x 2 x double>, <vscale x 2 x i1>, i32)
+
+define <vscale x 2 x half> @vfptrunc_nxv2f16_nxv2f64(<vscale x 2 x double> %a, <vscale x 2 x i1> %m, i32 zeroext %vl) {
+; CHECK-LABEL: vfptrunc_nxv2f16_nxv2f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, mu
+; CHECK-NEXT:    vfncvt.rod.f.f.w v10, v8, v0.t
+; CHECK-NEXT:    vsetvli zero, zero, e16, mf2, ta, mu
+; CHECK-NEXT:    vfncvt.f.f.w v8, v10, v0.t
+; CHECK-NEXT:    ret
+  %v = call <vscale x 2 x half> @llvm.vp.fptrunc.nxv2f16.nxv2f64(<vscale x 2 x double> %a, <vscale x 2 x i1> %m, i32 %vl)
+  ret <vscale x 2 x half> %v
+}
+
+define <vscale x 2 x half> @vfptrunc_nxv2f16_nxv2f64_unmasked(<vscale x 2 x double> %a, i32 zeroext %vl) {
+; CHECK-LABEL: vfptrunc_nxv2f16_nxv2f64_unmasked:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, mu
+; CHECK-NEXT:    vfncvt.rod.f.f.w v10, v8
+; CHECK-NEXT:    vsetvli zero, zero, e16, mf2, ta, mu
+; CHECK-NEXT:    vfncvt.f.f.w v8, v10
+; CHECK-NEXT:    ret
+  %v = call <vscale x 2 x half> @llvm.vp.fptrunc.nxv2f16.nxv2f64(<vscale x 2 x double> %a, <vscale x 2 x i1> shufflevector (<vscale x 2 x i1> insertelement (<vscale x 2 x i1> undef, i1 true, i32 0), <vscale x 2 x i1> undef, <vscale x 2 x i32> zeroinitializer), i32 %vl)
+  ret <vscale x 2 x half> %v
+}
+
+declare <vscale x 2 x float> @llvm.vp.fptrunc.nxv2f64.nxv2f32(<vscale x 2 x double>, <vscale x 2 x i1>, i32)
+
+define <vscale x 2 x float> @vfptrunc_nxv2f32_nxv2f64(<vscale x 2 x double> %a, <vscale x 2 x i1> %m, i32 zeroext %vl) {
+; CHECK-LABEL: vfptrunc_nxv2f32_nxv2f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, mu
+; CHECK-NEXT:    vfncvt.f.f.w v10, v8, v0.t
+; CHECK-NEXT:    vmv.v.v v8, v10
+; CHECK-NEXT:    ret
+  %v = call <vscale x 2 x float> @llvm.vp.fptrunc.nxv2f64.nxv2f32(<vscale x 2 x double> %a, <vscale x 2 x i1> %m, i32 %vl)
+  ret <vscale x 2 x float> %v
+}
+
+define <vscale x 2 x float> @vfptrunc_nxv2f32_nxv2f64_unmasked(<vscale x 2 x double> %a, i32 zeroext %vl) {
+; CHECK-LABEL: vfptrunc_nxv2f32_nxv2f64_unmasked:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, mu
+; CHECK-NEXT:    vfncvt.f.f.w v10, v8
+; CHECK-NEXT:    vmv.v.v v8, v10
+; CHECK-NEXT:    ret
+  %v = call <vscale x 2 x float> @llvm.vp.fptrunc.nxv2f64.nxv2f32(<vscale x 2 x double> %a, <vscale x 2 x i1> shufflevector (<vscale x 2 x i1> insertelement (<vscale x 2 x i1> undef, i1 true, i32 0), <vscale x 2 x i1> undef, <vscale x 2 x i32> zeroinitializer), i32 %vl)
+  ret <vscale x 2 x float> %v
+}