[RISCV] Implement support for bf16 truncate/extend on hard FP targets

[asb]

Summary

For the same reasons as D151284, this requires custom lowering of the truncate libcall on hard float ABIs (the normal libcall code path is used on soft ABIs).

The extend operation is implemented by a shift just as in the standard legalisation, but needs to be custom lowered because i32 isn't a legal type on RV64.

This patch aims to make the minimal changes that result in correct codegen for the bfloat.ll tests.

[craig.topper]

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

4946

I think XLenVT here here should be Op.getOperand(0).getValueType() which probably is XLenVT, but it looks better to use the same VT for both.

[joshua-arch1]

Are you still working on this patch? I have implemented a similar one early in May 24th.
https://reviews.llvm.org/D151313

[joshua-arch1]

This comment was removed by joshua-arch1.

[asb]

Address review comments, rebase.

@joshua-arch1: I'm sorry our work in this area has overlapped - as I mentioned in the first version of your patch, I was already working on a series in this area and the original form of D151313 was quite different to how it ended up.

[craig.topper]

LGTM

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

if (Subtarget.hasStdExtFOrZfinx()) {
     setCondCodeAction(FPCCToExpand, MVT::f32, Expand);
     setOperationAction(ISD::SELECT_CC, MVT::f32, Expand);
     setOperationAction(ISD::SELECT, MVT::f32, Custom);
     setOperationAction(ISD::BR_CC, MVT::f32, Expand);
     setOperationAction(FPOpToExpand, MVT::f32, Expand);
     setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::f16, Expand);
     setTruncStoreAction(MVT::f32, MVT::f16, Expand);
     setOperationAction(ISD::IS_FPCLASS, MVT::f32, Custom);
+    setOperationAction(ISD::BF16_TO_FP, MVT::f32, Custom);
+    setOperationAction(ISD::FP_TO_BF16, MVT::f32,
+                       Subtarget.isSoftFPABI() ? LibCall : Custom);
 
     if (Subtarget.hasStdExtZfa())
       setOperationAction(ISD::FNEARBYINT, MVT::f32, Legal);
   }
 
   if (Subtarget.hasStdExtFOrZfinx() && Subtarget.is64Bit())
     setOperationAction(ISD::BITCAST, MVT::i32, Custom);
 
if (Subtarget.hasStdExtDOrZdinx()) {
     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);
     setOperationAction(FPOpToExpand, MVT::f64, Expand);
     setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f16, Expand);
     setTruncStoreAction(MVT::f64, MVT::f16, Expand);
     setOperationAction(ISD::IS_FPCLASS, MVT::f64, Custom);
+    setOperationAction(ISD::BF16_TO_FP, MVT::f64, Custom);
+    setOperationAction(ISD::FP_TO_BF16, MVT::f64,
+                       Subtarget.isSoftFPABI() ? LibCall : Custom);
   }
 
   if (Subtarget.is64Bit()) {
     setOperationAction({ISD::FP_TO_UINT, ISD::FP_TO_SINT,
                         ISD::STRICT_FP_TO_UINT, ISD::STRICT_FP_TO_SINT},
                        MVT::i32, Custom);
     setOperationAction(ISD::LROUND, MVT::i32, Custom);
   }
if (IsStrict) {
       return DAG.getMergeValues({SubVec, Src.getValue(1)}, DL);
     }
     Src = DAG.getNode(RVVOpc, DL, ContainerVT, Src, Mask, VL);
     return convertFromScalableVector(VT, Src, DAG, Subtarget);
   }
   case ISD::FP_TO_SINT_SAT:
   case ISD::FP_TO_UINT_SAT:
     return lowerFP_TO_INT_SAT(Op, DAG, Subtarget);
+  case ISD::FP_TO_BF16: {
+    // Custom lower to ensure the libcall return is passed in an FPR on hard
+    // float ABIs.
+    assert(!Subtarget.isSoftFPABI() && "Unexpected custom legalization");
+    SDLoc DL(Op);
+    MakeLibCallOptions CallOptions;
+    RTLIB::Libcall LC =
+        RTLIB::getFPROUND(Op.getOperand(0).getValueType(), MVT::bf16);
+    SDValue Res =
+        makeLibCall(DAG, LC, MVT::f32, Op.getOperand(0), CallOptions, DL).first;
+    if (Subtarget.is64Bit())
+      return DAG.getNode(RISCVISD::FMV_X_ANYEXTW_RV64, DL, MVT::i64, Res);
+    return DAG.getBitcast(MVT::i32, Res);
+  }
+  case ISD::BF16_TO_FP: {
+    assert(Subtarget.hasStdExtFOrZfinx() && "Unexpected custom legalization");
+    MVT VT = Op.getSimpleValueType();
+    SDLoc DL(Op);
+    Op = DAG.getNode(
+        ISD::SHL, DL, Op.getOperand(0).getValueType(), Op.getOperand(0),
+        DAG.getShiftAmountConstant(16, Op.getOperand(0).getValueType(), DL));
+    SDValue Res = Subtarget.is64Bit()
+                      ? DAG.getNode(RISCVISD::FMV_W_X_RV64, DL, MVT::f32, Op)
+                      : DAG.getBitcast(MVT::f32, Op);
+    // fp_extend if the target VT is bigger than f32.
+    if (VT != MVT::f32)
+      return DAG.getNode(ISD::FP_EXTEND, DL, VT, Res);
+    return Res;
+  }
   case ISD::FTRUNC:
   case ISD::FCEIL:
   case ISD::FFLOOR:
   case ISD::FNEARBYINT:
   case ISD::FRINT:
   case ISD::FROUND:
   case ISD::FROUNDEVEN:
     return lowerFTRUNC_FCEIL_FFLOOR_FROUND(Op, DAG, Subtarget);
bool RISCVTargetLowering::allowsMisalignedMemoryAccesses(
   return Subtarget.enableUnalignedVectorMem();
 }
 
 bool RISCVTargetLowering::splitValueIntoRegisterParts(
     SelectionDAG &DAG, const SDLoc &DL, SDValue Val, SDValue *Parts,
     unsigned NumParts, MVT PartVT, std::optional<CallingConv::ID> CC) const {
   bool IsABIRegCopy = CC.has_value();
   EVT ValueVT = Val.getValueType();
-  if (IsABIRegCopy && ValueVT == MVT::f16 && PartVT == MVT::f32) {
-    // Cast the f16 to i16, extend to i32, pad with ones to make a float nan,
-    // and cast to f32.
+  if (IsABIRegCopy && (ValueVT == MVT::f16 || ValueVT == MVT::bf16) &&
+      PartVT == MVT::f32) {
+    // Cast the [b]f16 to i16, extend to i32, pad with ones to make a float
+    // nan, and cast to f32.
     Val = DAG.getNode(ISD::BITCAST, DL, MVT::i16, Val);
     Val = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i32, Val);
     Val = DAG.getNode(ISD::OR, DL, MVT::i32, Val,
                       DAG.getConstant(0xFFFF0000, DL, MVT::i32));
     Val = DAG.getNode(ISD::BITCAST, DL, MVT::f32, Val);
     Parts[0] = Val;
     return true;
   }
bool RISCVTargetLowering::splitValueIntoRegisterParts(
   }
   return false;
 }
 
 SDValue RISCVTargetLowering::joinRegisterPartsIntoValue(
     SelectionDAG &DAG, const SDLoc &DL, const SDValue *Parts, unsigned NumParts,
     MVT PartVT, EVT ValueVT, std::optional<CallingConv::ID> CC) const {
   bool IsABIRegCopy = CC.has_value();
-  if (IsABIRegCopy && ValueVT == MVT::f16 && PartVT == MVT::f32) {
+  if (IsABIRegCopy && (ValueVT == MVT::f16 || ValueVT == MVT::bf16) &&
+      PartVT == MVT::f32) {
     SDValue Val = Parts[0];
 
-    // Cast the f32 to i32, truncate to i16, and cast back to f16.
+    // Cast the f32 to i32, truncate to i16, and cast back to [b]f16.
     Val = DAG.getNode(ISD::BITCAST, DL, MVT::i32, Val);
     Val = DAG.getNode(ISD::TRUNCATE, DL, MVT::i16, Val);
-    Val = DAG.getNode(ISD::BITCAST, DL, MVT::f16, Val);
+    Val = DAG.getNode(ISD::BITCAST, DL, ValueVT, Val);
     return Val;
   }
 
   if (ValueVT.isScalableVector() && PartVT.isScalableVector()) {
     LLVMContext &Context = *DAG.getContext();
     SDValue Val = Parts[0];
     EVT ValueEltVT = ValueVT.getVectorElementType();
     EVT PartEltVT = PartVT.getVectorElementType();

llvm/test/CodeGen/RISCV/bfloat.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=riscv32 -verify-machineinstrs < %s | FileCheck %s -check-prefix=RV32I-ILP32
 ; RUN: llc -mtriple=riscv64 -verify-machineinstrs < %s | FileCheck %s -check-prefix=RV64I-LP64
-
-; TODO: Enable codegen for hard float.
+; RUN: llc -mtriple=riscv32 -mattr=+d -target-abi=ilp32 -verify-machineinstrs < %s | FileCheck %s -check-prefix=RV32ID-ILP32
+; RUN: llc -mtriple=riscv64 -mattr=+d -target-abi=lp64 -verify-machineinstrs < %s  | FileCheck %s -check-prefix=RV64ID-LP64
+; RUN: llc -mtriple=riscv32 -mattr=+d -target-abi=ilp32d -verify-machineinstrs < %s | FileCheck %s -check-prefix=RV32ID-ILP32D
+; RUN: llc -mtriple=riscv64 -mattr=+d -target-abi=lp64d -verify-machineinstrs < %s  | FileCheck %s -check-prefix=RV64ID-LP64D
 
 define bfloat @float_to_bfloat(float %a) nounwind {
 ; RV32I-ILP32-LABEL: float_to_bfloat:
 ; RV32I-ILP32:       # %bb.0:
 ; RV32I-ILP32-NEXT:    addi sp, sp, -16
 ; RV32I-ILP32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-ILP32-NEXT:    call __truncsfbf2@plt
 ; RV32I-ILP32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-ILP32-NEXT:    addi sp, sp, 16
 ; RV32I-ILP32-NEXT:    ret
 ;
 ; RV64I-LP64-LABEL: float_to_bfloat:
 ; RV64I-LP64:       # %bb.0:
 ; RV64I-LP64-NEXT:    addi sp, sp, -16
 ; RV64I-LP64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64I-LP64-NEXT:    call __truncsfbf2@plt
 ; RV64I-LP64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-LP64-NEXT:    addi sp, sp, 16
 ; RV64I-LP64-NEXT:    ret
+;
+; RV32ID-ILP32-LABEL: float_to_bfloat:
+; RV32ID-ILP32:       # %bb.0:
+; RV32ID-ILP32-NEXT:    addi sp, sp, -16
+; RV32ID-ILP32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32ID-ILP32-NEXT:    call __truncsfbf2@plt
+; RV32ID-ILP32-NEXT:    lui a1, 1048560
+; RV32ID-ILP32-NEXT:    or a0, a0, a1
+; RV32ID-ILP32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32ID-ILP32-NEXT:    addi sp, sp, 16
+; RV32ID-ILP32-NEXT:    ret
+;
+; RV64ID-LP64-LABEL: float_to_bfloat:
+; RV64ID-LP64:       # %bb.0:
+; RV64ID-LP64-NEXT:    addi sp, sp, -16
+; RV64ID-LP64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64ID-LP64-NEXT:    call __truncsfbf2@plt
+; RV64ID-LP64-NEXT:    lui a1, 1048560
+; RV64ID-LP64-NEXT:    or a0, a0, a1
+; RV64ID-LP64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64ID-LP64-NEXT:    addi sp, sp, 16
+; RV64ID-LP64-NEXT:    ret
+;
+; RV32ID-ILP32D-LABEL: float_to_bfloat:
+; RV32ID-ILP32D:       # %bb.0:
+; RV32ID-ILP32D-NEXT:    addi sp, sp, -16
+; RV32ID-ILP32D-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32ID-ILP32D-NEXT:    call __truncsfbf2@plt
+; RV32ID-ILP32D-NEXT:    fmv.x.w a0, fa0
+; RV32ID-ILP32D-NEXT:    lui a1, 1048560
+; RV32ID-ILP32D-NEXT:    or a0, a0, a1
+; RV32ID-ILP32D-NEXT:    fmv.w.x fa0, a0
+; RV32ID-ILP32D-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32ID-ILP32D-NEXT:    addi sp, sp, 16
+; RV32ID-ILP32D-NEXT:    ret
+;
+; RV64ID-LP64D-LABEL: float_to_bfloat:
+; RV64ID-LP64D:       # %bb.0:
+; RV64ID-LP64D-NEXT:    addi sp, sp, -16
+; RV64ID-LP64D-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64ID-LP64D-NEXT:    call __truncsfbf2@plt
+; RV64ID-LP64D-NEXT:    fmv.x.w a0, fa0
+; RV64ID-LP64D-NEXT:    lui a1, 1048560
+; RV64ID-LP64D-NEXT:    or a0, a0, a1
+; RV64ID-LP64D-NEXT:    fmv.w.x fa0, a0
+; RV64ID-LP64D-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64ID-LP64D-NEXT:    addi sp, sp, 16
+; RV64ID-LP64D-NEXT:    ret
   %1 = fptrunc float %a to bfloat
   ret bfloat %1
 }
 
 define bfloat @double_to_bfloat(double %a) nounwind {
 ; RV32I-ILP32-LABEL: double_to_bfloat:
 ; RV32I-ILP32:       # %bb.0:
 ; RV32I-ILP32-NEXT:    addi sp, sp, -16
 ; RV32I-ILP32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-ILP32-NEXT:    call __truncdfbf2@plt
 ; RV32I-ILP32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-ILP32-NEXT:    addi sp, sp, 16
 ; RV32I-ILP32-NEXT:    ret
 ;
 ; RV64I-LP64-LABEL: double_to_bfloat:
 ; RV64I-LP64:       # %bb.0:
 ; RV64I-LP64-NEXT:    addi sp, sp, -16
 ; RV64I-LP64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64I-LP64-NEXT:    call __truncdfbf2@plt
 ; RV64I-LP64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-LP64-NEXT:    addi sp, sp, 16
 ; RV64I-LP64-NEXT:    ret
+;
+; RV32ID-ILP32-LABEL: double_to_bfloat:
+; RV32ID-ILP32:       # %bb.0:
+; RV32ID-ILP32-NEXT:    addi sp, sp, -16
+; RV32ID-ILP32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32ID-ILP32-NEXT:    call __truncdfbf2@plt
+; RV32ID-ILP32-NEXT:    lui a1, 1048560
+; RV32ID-ILP32-NEXT:    or a0, a0, a1
+; RV32ID-ILP32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32ID-ILP32-NEXT:    addi sp, sp, 16
+; RV32ID-ILP32-NEXT:    ret
+;
+; RV64ID-LP64-LABEL: double_to_bfloat:
+; RV64ID-LP64:       # %bb.0:
+; RV64ID-LP64-NEXT:    addi sp, sp, -16
+; RV64ID-LP64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64ID-LP64-NEXT:    call __truncdfbf2@plt
+; RV64ID-LP64-NEXT:    lui a1, 1048560
+; RV64ID-LP64-NEXT:    or a0, a0, a1
+; RV64ID-LP64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64ID-LP64-NEXT:    addi sp, sp, 16
+; RV64ID-LP64-NEXT:    ret
+;
+; RV32ID-ILP32D-LABEL: double_to_bfloat:
+; RV32ID-ILP32D:       # %bb.0:
+; RV32ID-ILP32D-NEXT:    addi sp, sp, -16
+; RV32ID-ILP32D-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32ID-ILP32D-NEXT:    call __truncdfbf2@plt
+; RV32ID-ILP32D-NEXT:    fmv.x.w a0, fa0
+; RV32ID-ILP32D-NEXT:    lui a1, 1048560
+; RV32ID-ILP32D-NEXT:    or a0, a0, a1
+; RV32ID-ILP32D-NEXT:    fmv.w.x fa0, a0
+; RV32ID-ILP32D-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32ID-ILP32D-NEXT:    addi sp, sp, 16
+; RV32ID-ILP32D-NEXT:    ret
+;
+; RV64ID-LP64D-LABEL: double_to_bfloat:
+; RV64ID-LP64D:       # %bb.0:
+; RV64ID-LP64D-NEXT:    addi sp, sp, -16
+; RV64ID-LP64D-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64ID-LP64D-NEXT:    call __truncdfbf2@plt
+; RV64ID-LP64D-NEXT:    fmv.x.w a0, fa0
+; RV64ID-LP64D-NEXT:    lui a1, 1048560
+; RV64ID-LP64D-NEXT:    or a0, a0, a1
+; RV64ID-LP64D-NEXT:    fmv.w.x fa0, a0
+; RV64ID-LP64D-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64ID-LP64D-NEXT:    addi sp, sp, 16
+; RV64ID-LP64D-NEXT:    ret
   %1 = fptrunc double %a to bfloat
   ret bfloat %1
 }
 
 define float @bfloat_to_float(bfloat %a) nounwind {
 ; RV32I-ILP32-LABEL: bfloat_to_float:
 ; RV32I-ILP32:       # %bb.0:
 ; RV32I-ILP32-NEXT:    slli a0, a0, 16
 ; RV32I-ILP32-NEXT:    ret
 ;
 ; RV64I-LP64-LABEL: bfloat_to_float:
 ; RV64I-LP64:       # %bb.0:
 ; RV64I-LP64-NEXT:    slliw a0, a0, 16
 ; RV64I-LP64-NEXT:    ret
+;
+; RV32ID-ILP32-LABEL: bfloat_to_float:
+; RV32ID-ILP32:       # %bb.0:
+; RV32ID-ILP32-NEXT:    slli a0, a0, 16
+; RV32ID-ILP32-NEXT:    ret
+;
+; RV64ID-LP64-LABEL: bfloat_to_float:
+; RV64ID-LP64:       # %bb.0:
+; RV64ID-LP64-NEXT:    slli a0, a0, 48
+; RV64ID-LP64-NEXT:    srli a0, a0, 48
+; RV64ID-LP64-NEXT:    slli a0, a0, 16
+; RV64ID-LP64-NEXT:    ret
+;
+; RV32ID-ILP32D-LABEL: bfloat_to_float:
+; RV32ID-ILP32D:       # %bb.0:
+; RV32ID-ILP32D-NEXT:    fmv.x.w a0, fa0
+; RV32ID-ILP32D-NEXT:    slli a0, a0, 16
+; RV32ID-ILP32D-NEXT:    fmv.w.x fa0, a0
+; RV32ID-ILP32D-NEXT:    ret
+;
+; RV64ID-LP64D-LABEL: bfloat_to_float:
+; RV64ID-LP64D:       # %bb.0:
+; RV64ID-LP64D-NEXT:    fmv.x.w a0, fa0
+; RV64ID-LP64D-NEXT:    slli a0, a0, 48
+; RV64ID-LP64D-NEXT:    srli a0, a0, 48
+; RV64ID-LP64D-NEXT:    slli a0, a0, 16
+; RV64ID-LP64D-NEXT:    fmv.w.x fa0, a0
+; RV64ID-LP64D-NEXT:    ret
   %1 = fpext bfloat %a to float
   ret float %1
 }
 
 define double @bfloat_to_double(bfloat %a) nounwind {
 ; RV32I-ILP32-LABEL: bfloat_to_double:
 ; RV32I-ILP32:       # %bb.0:
 ; RV32I-ILP32-NEXT:    addi sp, sp, -16
 ; RV32I-ILP32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32I-ILP32-NEXT:    slli a0, a0, 16
 ; RV32I-ILP32-NEXT:    call __extendsfdf2@plt
 ; RV32I-ILP32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-ILP32-NEXT:    addi sp, sp, 16
 ; RV32I-ILP32-NEXT:    ret
 ;
 ; RV64I-LP64-LABEL: bfloat_to_double:
 ; RV64I-LP64:       # %bb.0:
 ; RV64I-LP64-NEXT:    addi sp, sp, -16
 ; RV64I-LP64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64I-LP64-NEXT:    slliw a0, a0, 16
 ; RV64I-LP64-NEXT:    call __extendsfdf2@plt
 ; RV64I-LP64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-LP64-NEXT:    addi sp, sp, 16
 ; RV64I-LP64-NEXT:    ret
+;
+; RV32ID-ILP32-LABEL: bfloat_to_double:
+; RV32ID-ILP32:       # %bb.0:
+; RV32ID-ILP32-NEXT:    addi sp, sp, -16
+; RV32ID-ILP32-NEXT:    slli a0, a0, 16
+; RV32ID-ILP32-NEXT:    fmv.w.x fa5, a0
+; RV32ID-ILP32-NEXT:    fcvt.d.s fa5, fa5
+; RV32ID-ILP32-NEXT:    fsd fa5, 8(sp)
+; RV32ID-ILP32-NEXT:    lw a0, 8(sp)
+; RV32ID-ILP32-NEXT:    lw a1, 12(sp)
+; RV32ID-ILP32-NEXT:    addi sp, sp, 16
+; RV32ID-ILP32-NEXT:    ret
+;
+; RV64ID-LP64-LABEL: bfloat_to_double:
+; RV64ID-LP64:       # %bb.0:
+; RV64ID-LP64-NEXT:    slli a0, a0, 48
+; RV64ID-LP64-NEXT:    srli a0, a0, 48
+; RV64ID-LP64-NEXT:    slli a0, a0, 16
+; RV64ID-LP64-NEXT:    fmv.w.x fa5, a0
+; RV64ID-LP64-NEXT:    fcvt.d.s fa5, fa5
+; RV64ID-LP64-NEXT:    fmv.x.d a0, fa5
+; RV64ID-LP64-NEXT:    ret
+;
+; RV32ID-ILP32D-LABEL: bfloat_to_double:
+; RV32ID-ILP32D:       # %bb.0:
+; RV32ID-ILP32D-NEXT:    fmv.x.w a0, fa0
+; RV32ID-ILP32D-NEXT:    slli a0, a0, 16
+; RV32ID-ILP32D-NEXT:    fmv.w.x fa5, a0
+; RV32ID-ILP32D-NEXT:    fcvt.d.s fa0, fa5
+; RV32ID-ILP32D-NEXT:    ret
+;
+; RV64ID-LP64D-LABEL: bfloat_to_double:
+; RV64ID-LP64D:       # %bb.0:
+; RV64ID-LP64D-NEXT:    fmv.x.w a0, fa0
+; RV64ID-LP64D-NEXT:    slli a0, a0, 48
+; RV64ID-LP64D-NEXT:    srli a0, a0, 48
+; RV64ID-LP64D-NEXT:    slli a0, a0, 16
+; RV64ID-LP64D-NEXT:    fmv.w.x fa5, a0
+; RV64ID-LP64D-NEXT:    fcvt.d.s fa0, fa5
+; RV64ID-LP64D-NEXT:    ret
   %1 = fpext bfloat %a to double
   ret double %1
 }
 
 define bfloat @i16_to_bfloat(i16 %a) nounwind {
 ; RV32I-ILP32-LABEL: i16_to_bfloat:
 ; RV32I-ILP32:       # %bb.0:
 ; RV32I-ILP32-NEXT:    ret
 ;
 ; RV64I-LP64-LABEL: i16_to_bfloat:
 ; RV64I-LP64:       # %bb.0:
 ; RV64I-LP64-NEXT:    ret
+;
+; RV32ID-ILP32-LABEL: i16_to_bfloat:
+; RV32ID-ILP32:       # %bb.0:
+; RV32ID-ILP32-NEXT:    lui a1, 1048560
+; RV32ID-ILP32-NEXT:    or a0, a0, a1
+; RV32ID-ILP32-NEXT:    ret
+;
+; RV64ID-LP64-LABEL: i16_to_bfloat:
+; RV64ID-LP64:       # %bb.0:
+; RV64ID-LP64-NEXT:    lui a1, 1048560
+; RV64ID-LP64-NEXT:    or a0, a0, a1
+; RV64ID-LP64-NEXT:    ret
+;
+; RV32ID-ILP32D-LABEL: i16_to_bfloat:
+; RV32ID-ILP32D:       # %bb.0:
+; RV32ID-ILP32D-NEXT:    lui a1, 1048560
+; RV32ID-ILP32D-NEXT:    or a0, a0, a1
+; RV32ID-ILP32D-NEXT:    fmv.w.x fa0, a0
+; RV32ID-ILP32D-NEXT:    ret
+;
+; RV64ID-LP64D-LABEL: i16_to_bfloat:
+; RV64ID-LP64D:       # %bb.0:
+; RV64ID-LP64D-NEXT:    lui a1, 1048560
+; RV64ID-LP64D-NEXT:    or a0, a0, a1
+; RV64ID-LP64D-NEXT:    fmv.w.x fa0, a0
+; RV64ID-LP64D-NEXT:    ret
   %1 = bitcast i16 %a to bfloat
   ret bfloat %1
 }
 
 define i16 @bfloat_to_i16(bfloat %a) nounwind {
 ; RV32I-ILP32-LABEL: bfloat_to_i16:
 ; RV32I-ILP32:       # %bb.0:
 ; RV32I-ILP32-NEXT:    ret
 ;
 ; RV64I-LP64-LABEL: bfloat_to_i16:
 ; RV64I-LP64:       # %bb.0:
 ; RV64I-LP64-NEXT:    ret
+;
+; RV32ID-ILP32-LABEL: bfloat_to_i16:
+; RV32ID-ILP32:       # %bb.0:
+; RV32ID-ILP32-NEXT:    ret
+;
+; RV64ID-LP64-LABEL: bfloat_to_i16:
+; RV64ID-LP64:       # %bb.0:
+; RV64ID-LP64-NEXT:    ret
+;
+; RV32ID-ILP32D-LABEL: bfloat_to_i16:
+; RV32ID-ILP32D:       # %bb.0:
+; RV32ID-ILP32D-NEXT:    fmv.x.w a0, fa0
+; RV32ID-ILP32D-NEXT:    ret
+;
+; RV64ID-LP64D-LABEL: bfloat_to_i16:
+; RV64ID-LP64D:       # %bb.0:
+; RV64ID-LP64D-NEXT:    fmv.x.w a0, fa0
+; RV64ID-LP64D-NEXT:    ret
   %1 = bitcast bfloat %a to i16
   ret i16 %1
 }
 
 define bfloat @bfloat_add(bfloat %a, bfloat %b) nounwind {
 ; RV32I-ILP32-LABEL: bfloat_add:
 ; RV32I-ILP32:       # %bb.0:
 ; RV32I-ILP32-NEXT:    addi sp, sp, -16

 ; RV64I-LP64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64I-LP64-NEXT:    slliw a0, a0, 16
 ; RV64I-LP64-NEXT:    slliw a1, a1, 16
 ; RV64I-LP64-NEXT:    call __addsf3@plt
 ; RV64I-LP64-NEXT:    call __truncsfbf2@plt
 ; RV64I-LP64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-LP64-NEXT:    addi sp, sp, 16
 ; RV64I-LP64-NEXT:    ret
+;
+; RV32ID-ILP32-LABEL: bfloat_add:
+; RV32ID-ILP32:       # %bb.0:
+; RV32ID-ILP32-NEXT:    addi sp, sp, -16
+; RV32ID-ILP32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32ID-ILP32-NEXT:    slli a1, a1, 16
+; RV32ID-ILP32-NEXT:    fmv.w.x fa5, a1
+; RV32ID-ILP32-NEXT:    slli a0, a0, 16
+; RV32ID-ILP32-NEXT:    fmv.w.x fa4, a0
+; RV32ID-ILP32-NEXT:    fadd.s fa5, fa4, fa5
+; RV32ID-ILP32-NEXT:    fmv.x.w a0, fa5
+; RV32ID-ILP32-NEXT:    call __truncsfbf2@plt
+; RV32ID-ILP32-NEXT:    lui a1, 1048560
+; RV32ID-ILP32-NEXT:    or a0, a0, a1
+; RV32ID-ILP32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32ID-ILP32-NEXT:    addi sp, sp, 16
+; RV32ID-ILP32-NEXT:    ret
+;
+; RV64ID-LP64-LABEL: bfloat_add:
+; RV64ID-LP64:       # %bb.0:
+; RV64ID-LP64-NEXT:    addi sp, sp, -16
+; RV64ID-LP64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64ID-LP64-NEXT:    lui a2, 16
+; RV64ID-LP64-NEXT:    addiw a2, a2, -1
+; RV64ID-LP64-NEXT:    and a0, a0, a2
+; RV64ID-LP64-NEXT:    and a1, a1, a2
+; RV64ID-LP64-NEXT:    slli a1, a1, 16
+; RV64ID-LP64-NEXT:    fmv.w.x fa5, a1
+; RV64ID-LP64-NEXT:    slli a0, a0, 16
+; RV64ID-LP64-NEXT:    fmv.w.x fa4, a0
+; RV64ID-LP64-NEXT:    fadd.s fa5, fa4, fa5
+; RV64ID-LP64-NEXT:    fmv.x.w a0, fa5
+; RV64ID-LP64-NEXT:    call __truncsfbf2@plt
+; RV64ID-LP64-NEXT:    lui a1, 1048560
+; RV64ID-LP64-NEXT:    or a0, a0, a1
+; RV64ID-LP64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64ID-LP64-NEXT:    addi sp, sp, 16
+; RV64ID-LP64-NEXT:    ret
+;
+; RV32ID-ILP32D-LABEL: bfloat_add:
+; RV32ID-ILP32D:       # %bb.0:
+; RV32ID-ILP32D-NEXT:    addi sp, sp, -16
+; RV32ID-ILP32D-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32ID-ILP32D-NEXT:    fmv.x.w a0, fa0
+; RV32ID-ILP32D-NEXT:    fmv.x.w a1, fa1
+; RV32ID-ILP32D-NEXT:    slli a1, a1, 16
+; RV32ID-ILP32D-NEXT:    fmv.w.x fa5, a1
+; RV32ID-ILP32D-NEXT:    slli a0, a0, 16
+; RV32ID-ILP32D-NEXT:    fmv.w.x fa4, a0
+; RV32ID-ILP32D-NEXT:    fadd.s fa0, fa4, fa5
+; RV32ID-ILP32D-NEXT:    call __truncsfbf2@plt
+; RV32ID-ILP32D-NEXT:    fmv.x.w a0, fa0
+; RV32ID-ILP32D-NEXT:    lui a1, 1048560
+; RV32ID-ILP32D-NEXT:    or a0, a0, a1
+; RV32ID-ILP32D-NEXT:    fmv.w.x fa0, a0
+; RV32ID-ILP32D-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32ID-ILP32D-NEXT:    addi sp, sp, 16
+; RV32ID-ILP32D-NEXT:    ret
+;
+; RV64ID-LP64D-LABEL: bfloat_add:
+; RV64ID-LP64D:       # %bb.0:
+; RV64ID-LP64D-NEXT:    addi sp, sp, -16
+; RV64ID-LP64D-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64ID-LP64D-NEXT:    fmv.x.w a0, fa0
+; RV64ID-LP64D-NEXT:    lui a1, 16
+; RV64ID-LP64D-NEXT:    addiw a1, a1, -1
+; RV64ID-LP64D-NEXT:    and a0, a0, a1
+; RV64ID-LP64D-NEXT:    fmv.x.w a2, fa1
+; RV64ID-LP64D-NEXT:    and a1, a2, a1
+; RV64ID-LP64D-NEXT:    slli a1, a1, 16
+; RV64ID-LP64D-NEXT:    fmv.w.x fa5, a1
+; RV64ID-LP64D-NEXT:    slli a0, a0, 16
+; RV64ID-LP64D-NEXT:    fmv.w.x fa4, a0
+; RV64ID-LP64D-NEXT:    fadd.s fa0, fa4, fa5
+; RV64ID-LP64D-NEXT:    call __truncsfbf2@plt
+; RV64ID-LP64D-NEXT:    fmv.x.w a0, fa0
+; RV64ID-LP64D-NEXT:    lui a1, 1048560
+; RV64ID-LP64D-NEXT:    or a0, a0, a1
+; RV64ID-LP64D-NEXT:    fmv.w.x fa0, a0
+; RV64ID-LP64D-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64ID-LP64D-NEXT:    addi sp, sp, 16
+; RV64ID-LP64D-NEXT:    ret
   %1 = fadd bfloat %a, %b
   ret bfloat %1
 }
 
 define bfloat @bfloat_load(ptr %a) nounwind {
 ; RV32I-ILP32-LABEL: bfloat_load:
 ; RV32I-ILP32:       # %bb.0:
 ; RV32I-ILP32-NEXT:    addi sp, sp, -16

 ; RV64I-LP64-NEXT:    lh a2, 6(a0)
 ; RV64I-LP64-NEXT:    slliw a0, a1, 16
 ; RV64I-LP64-NEXT:    slliw a1, a2, 16
 ; RV64I-LP64-NEXT:    call __addsf3@plt
 ; RV64I-LP64-NEXT:    call __truncsfbf2@plt
 ; RV64I-LP64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-LP64-NEXT:    addi sp, sp, 16
 ; RV64I-LP64-NEXT:    ret
+;
+; RV32ID-ILP32-LABEL: bfloat_load:
+; RV32ID-ILP32:       # %bb.0:
+; RV32ID-ILP32-NEXT:    addi sp, sp, -16
+; RV32ID-ILP32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32ID-ILP32-NEXT:    lhu a1, 6(a0)
+; RV32ID-ILP32-NEXT:    lhu a0, 0(a0)
+; RV32ID-ILP32-NEXT:    slli a1, a1, 16
+; RV32ID-ILP32-NEXT:    fmv.w.x fa5, a1
+; RV32ID-ILP32-NEXT:    slli a0, a0, 16
+; RV32ID-ILP32-NEXT:    fmv.w.x fa4, a0
+; RV32ID-ILP32-NEXT:    fadd.s fa5, fa4, fa5
+; RV32ID-ILP32-NEXT:    fmv.x.w a0, fa5
+; RV32ID-ILP32-NEXT:    call __truncsfbf2@plt
+; RV32ID-ILP32-NEXT:    lui a1, 1048560
+; RV32ID-ILP32-NEXT:    or a0, a0, a1
+; RV32ID-ILP32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32ID-ILP32-NEXT:    addi sp, sp, 16
+; RV32ID-ILP32-NEXT:    ret
+;
+; RV64ID-LP64-LABEL: bfloat_load:
+; RV64ID-LP64:       # %bb.0:
+; RV64ID-LP64-NEXT:    addi sp, sp, -16
+; RV64ID-LP64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64ID-LP64-NEXT:    lhu a1, 6(a0)
+; RV64ID-LP64-NEXT:    lhu a0, 0(a0)
+; RV64ID-LP64-NEXT:    slli a1, a1, 16
+; RV64ID-LP64-NEXT:    fmv.w.x fa5, a1
+; RV64ID-LP64-NEXT:    slli a0, a0, 16
+; RV64ID-LP64-NEXT:    fmv.w.x fa4, a0
+; RV64ID-LP64-NEXT:    fadd.s fa5, fa4, fa5
+; RV64ID-LP64-NEXT:    fmv.x.w a0, fa5
+; RV64ID-LP64-NEXT:    call __truncsfbf2@plt
+; RV64ID-LP64-NEXT:    lui a1, 1048560
+; RV64ID-LP64-NEXT:    or a0, a0, a1
+; RV64ID-LP64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64ID-LP64-NEXT:    addi sp, sp, 16
+; RV64ID-LP64-NEXT:    ret
+;
+; RV32ID-ILP32D-LABEL: bfloat_load:
+; RV32ID-ILP32D:       # %bb.0:
+; RV32ID-ILP32D-NEXT:    addi sp, sp, -16
+; RV32ID-ILP32D-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32ID-ILP32D-NEXT:    lhu a1, 6(a0)
+; RV32ID-ILP32D-NEXT:    lhu a0, 0(a0)
+; RV32ID-ILP32D-NEXT:    slli a1, a1, 16
+; RV32ID-ILP32D-NEXT:    fmv.w.x fa5, a1
+; RV32ID-ILP32D-NEXT:    slli a0, a0, 16
+; RV32ID-ILP32D-NEXT:    fmv.w.x fa4, a0
+; RV32ID-ILP32D-NEXT:    fadd.s fa0, fa4, fa5
+; RV32ID-ILP32D-NEXT:    call __truncsfbf2@plt
+; RV32ID-ILP32D-NEXT:    fmv.x.w a0, fa0
+; RV32ID-ILP32D-NEXT:    lui a1, 1048560
+; RV32ID-ILP32D-NEXT:    or a0, a0, a1
+; RV32ID-ILP32D-NEXT:    fmv.w.x fa0, a0
+; RV32ID-ILP32D-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32ID-ILP32D-NEXT:    addi sp, sp, 16
+; RV32ID-ILP32D-NEXT:    ret
+;
+; RV64ID-LP64D-LABEL: bfloat_load:
+; RV64ID-LP64D:       # %bb.0:
+; RV64ID-LP64D-NEXT:    addi sp, sp, -16
+; RV64ID-LP64D-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64ID-LP64D-NEXT:    lhu a1, 6(a0)
+; RV64ID-LP64D-NEXT:    lhu a0, 0(a0)
+; RV64ID-LP64D-NEXT:    slli a1, a1, 16
+; RV64ID-LP64D-NEXT:    fmv.w.x fa5, a1
+; RV64ID-LP64D-NEXT:    slli a0, a0, 16
+; RV64ID-LP64D-NEXT:    fmv.w.x fa4, a0
+; RV64ID-LP64D-NEXT:    fadd.s fa0, fa4, fa5
+; RV64ID-LP64D-NEXT:    call __truncsfbf2@plt
+; RV64ID-LP64D-NEXT:    fmv.x.w a0, fa0
+; RV64ID-LP64D-NEXT:    lui a1, 1048560
+; RV64ID-LP64D-NEXT:    or a0, a0, a1
+; RV64ID-LP64D-NEXT:    fmv.w.x fa0, a0
+; RV64ID-LP64D-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64ID-LP64D-NEXT:    addi sp, sp, 16
+; RV64ID-LP64D-NEXT:    ret
   %1 = load bfloat, ptr %a
   %2 = getelementptr bfloat, ptr %a, i32 3
   %3 = load bfloat, ptr %2
   %4 = fadd bfloat %1, %3
   ret bfloat %4
 }
 
 define void @bfloat_store(ptr %a, bfloat %b, bfloat %c) nounwind {

 ; RV64I-LP64-NEXT:    call __addsf3@plt
 ; RV64I-LP64-NEXT:    call __truncsfbf2@plt
 ; RV64I-LP64-NEXT:    sh a0, 0(s0)
 ; RV64I-LP64-NEXT:    sh a0, 16(s0)
 ; RV64I-LP64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-LP64-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
 ; RV64I-LP64-NEXT:    addi sp, sp, 16
 ; RV64I-LP64-NEXT:    ret
+;
+; RV32ID-ILP32-LABEL: bfloat_store:
+; RV32ID-ILP32:       # %bb.0:
+; RV32ID-ILP32-NEXT:    addi sp, sp, -16
+; RV32ID-ILP32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32ID-ILP32-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
+; RV32ID-ILP32-NEXT:    mv s0, a0
+; RV32ID-ILP32-NEXT:    slli a2, a2, 16
+; RV32ID-ILP32-NEXT:    fmv.w.x fa5, a2
+; RV32ID-ILP32-NEXT:    slli a1, a1, 16
+; RV32ID-ILP32-NEXT:    fmv.w.x fa4, a1
+; RV32ID-ILP32-NEXT:    fadd.s fa5, fa4, fa5
+; RV32ID-ILP32-NEXT:    fmv.x.w a0, fa5
+; RV32ID-ILP32-NEXT:    call __truncsfbf2@plt
+; RV32ID-ILP32-NEXT:    sh a0, 0(s0)
+; RV32ID-ILP32-NEXT:    sh a0, 16(s0)
+; RV32ID-ILP32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32ID-ILP32-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
+; RV32ID-ILP32-NEXT:    addi sp, sp, 16
+; RV32ID-ILP32-NEXT:    ret
+;
+; RV64ID-LP64-LABEL: bfloat_store:
+; RV64ID-LP64:       # %bb.0:
+; RV64ID-LP64-NEXT:    addi sp, sp, -16
+; RV64ID-LP64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64ID-LP64-NEXT:    sd s0, 0(sp) # 8-byte Folded Spill
+; RV64ID-LP64-NEXT:    mv s0, a0
+; RV64ID-LP64-NEXT:    lui a0, 16
+; RV64ID-LP64-NEXT:    addiw a0, a0, -1
+; RV64ID-LP64-NEXT:    and a1, a1, a0
+; RV64ID-LP64-NEXT:    and a0, a2, a0
+; RV64ID-LP64-NEXT:    slli a0, a0, 16
+; RV64ID-LP64-NEXT:    fmv.w.x fa5, a0
+; RV64ID-LP64-NEXT:    slli a1, a1, 16
+; RV64ID-LP64-NEXT:    fmv.w.x fa4, a1
+; RV64ID-LP64-NEXT:    fadd.s fa5, fa4, fa5
+; RV64ID-LP64-NEXT:    fmv.x.w a0, fa5
+; RV64ID-LP64-NEXT:    call __truncsfbf2@plt
+; RV64ID-LP64-NEXT:    sh a0, 0(s0)
+; RV64ID-LP64-NEXT:    sh a0, 16(s0)
+; RV64ID-LP64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64ID-LP64-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
+; RV64ID-LP64-NEXT:    addi sp, sp, 16
+; RV64ID-LP64-NEXT:    ret
+;
+; RV32ID-ILP32D-LABEL: bfloat_store:
+; RV32ID-ILP32D:       # %bb.0:
+; RV32ID-ILP32D-NEXT:    addi sp, sp, -16
+; RV32ID-ILP32D-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32ID-ILP32D-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
+; RV32ID-ILP32D-NEXT:    mv s0, a0
+; RV32ID-ILP32D-NEXT:    fmv.x.w a0, fa0
+; RV32ID-ILP32D-NEXT:    fmv.x.w a1, fa1
+; RV32ID-ILP32D-NEXT:    slli a1, a1, 16
+; RV32ID-ILP32D-NEXT:    fmv.w.x fa5, a1
+; RV32ID-ILP32D-NEXT:    slli a0, a0, 16
+; RV32ID-ILP32D-NEXT:    fmv.w.x fa4, a0
+; RV32ID-ILP32D-NEXT:    fadd.s fa0, fa4, fa5
+; RV32ID-ILP32D-NEXT:    call __truncsfbf2@plt
+; RV32ID-ILP32D-NEXT:    fmv.x.w a0, fa0
+; RV32ID-ILP32D-NEXT:    sh a0, 0(s0)
+; RV32ID-ILP32D-NEXT:    sh a0, 16(s0)
+; RV32ID-ILP32D-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32ID-ILP32D-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
+; RV32ID-ILP32D-NEXT:    addi sp, sp, 16
+; RV32ID-ILP32D-NEXT:    ret
+;
+; RV64ID-LP64D-LABEL: bfloat_store:
+; RV64ID-LP64D:       # %bb.0:
+; RV64ID-LP64D-NEXT:    addi sp, sp, -16
+; RV64ID-LP64D-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64ID-LP64D-NEXT:    sd s0, 0(sp) # 8-byte Folded Spill
+; RV64ID-LP64D-NEXT:    mv s0, a0
+; RV64ID-LP64D-NEXT:    fmv.x.w a0, fa0
+; RV64ID-LP64D-NEXT:    lui a1, 16
+; RV64ID-LP64D-NEXT:    addiw a1, a1, -1
+; RV64ID-LP64D-NEXT:    and a0, a0, a1
+; RV64ID-LP64D-NEXT:    fmv.x.w a2, fa1
+; RV64ID-LP64D-NEXT:    and a1, a2, a1
+; RV64ID-LP64D-NEXT:    slli a1, a1, 16
+; RV64ID-LP64D-NEXT:    fmv.w.x fa5, a1
+; RV64ID-LP64D-NEXT:    slli a0, a0, 16
+; RV64ID-LP64D-NEXT:    fmv.w.x fa4, a0
+; RV64ID-LP64D-NEXT:    fadd.s fa0, fa4, fa5
+; RV64ID-LP64D-NEXT:    call __truncsfbf2@plt
+; RV64ID-LP64D-NEXT:    fmv.x.w a0, fa0
+; RV64ID-LP64D-NEXT:    sh a0, 0(s0)
+; RV64ID-LP64D-NEXT:    sh a0, 16(s0)
+; RV64ID-LP64D-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64ID-LP64D-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
+; RV64ID-LP64D-NEXT:    addi sp, sp, 16
+; RV64ID-LP64D-NEXT:    ret
   %1 = fadd bfloat %b, %c
   store bfloat %1, ptr %a
   %2 = getelementptr bfloat, ptr %a, i32 8
   store bfloat %1, ptr %2
   ret void
 }