[RISCV] Expand test coverage of bf16 operations with Zfbfmin and fix gaps

[asb]

Summary

This doesn't bring us to parity with the test/CodeGen/RISCV/half-* test cases, it simply picks off an initial set that can be supported especially easy. In order to make the review more manageable, I'll follow up with other cases.

There is zero innovation in the test cases - they simply take the existing half/float cases and replace f16->bf16 and half->bfloat.

[craig.topper]

llvm/test/CodeGen/RISCV/bfloat-arith.ll

99

Missing CHECK lines

[asb]

Fix issue with missing check lines for two functions (thanks Craig, good catch - now sure why update_llc_test_checks didn't give the usual error about a mismatch).

[craig.topper]

LGTM

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

static const unsigned FPOpToExpand[] = {
       ISD::FREM};
 
   static const unsigned FPRndMode[] = {
       ISD::FCEIL, ISD::FFLOOR, ISD::FTRUNC, ISD::FRINT, ISD::FROUND,
       ISD::FROUNDEVEN};
 
   if (Subtarget.hasStdExtZfhOrZfhminOrZhinxOrZhinxmin())
     setOperationAction(ISD::BITCAST, MVT::i16, Custom);
-  
+  static const unsigned ZfhminZfbfminPromoteOps[] = {
+      ISD::FMINNUM,      ISD::FMAXNUM,       ISD::FADD,
+      ISD::FSUB,         ISD::FMUL,          ISD::FMA,
+      ISD::FDIV,         ISD::FSQRT,         ISD::FABS,
+      ISD::FNEG,         ISD::STRICT_FMA,    ISD::STRICT_FADD,
+      ISD::STRICT_FSUB,  ISD::STRICT_FMUL,   ISD::STRICT_FDIV,
+      ISD::STRICT_FSQRT, ISD::STRICT_FSETCC, ISD::STRICT_FSETCCS,
+      ISD::SETCC,        ISD::FCEIL,         ISD::FFLOOR,
+      ISD::FTRUNC,       ISD::FRINT,         ISD::FROUND,
+      ISD::FROUNDEVEN,   ISD::SELECT};
+
   if (Subtarget.hasStdExtZfbfmin()) {
     setOperationAction(ISD::BITCAST, MVT::i16, Custom);
     setOperationAction(ISD::BITCAST, MVT::bf16, Custom);
     setOperationAction(ISD::FP_ROUND, MVT::bf16, Custom);
     setOperationAction(ISD::FP_EXTEND, MVT::f32, Custom);
     setOperationAction(ISD::FP_EXTEND, MVT::f64, Custom);
     setOperationAction(ISD::ConstantFP, MVT::bf16, Expand);
     setOperationAction(ISD::SELECT_CC, MVT::bf16, Expand);
-    setOperationAction(ISD::SELECT, MVT::bf16, Promote);
-    setOperationAction(ISD::SETCC, MVT::bf16, Promote);
+    setOperationAction(ISD::BR_CC, MVT::bf16, Expand);
+    setOperationAction(ZfhminZfbfminPromoteOps, MVT::bf16, Promote);
+    setOperationAction(ISD::FREM, MVT::bf16, Promote);
+    // FIXME: Need to promote bf16 FCOPYSIGN to f32, but the
+    // DAGCombiner::visitFP_ROUND probably needs improvements first.
+    setOperationAction(ISD::FCOPYSIGN, MVT::bf16, Expand);
   }
 
   if (Subtarget.hasStdExtZfhOrZfhminOrZhinxOrZhinxmin()) {
     if (Subtarget.hasStdExtZfhOrZhinx()) {
       setOperationAction(FPLegalNodeTypes, MVT::f16, Legal);
       setOperationAction(FPRndMode, MVT::f16,
                          Subtarget.hasStdExtZfa() ? Legal : Custom);
       setOperationAction(ISD::SELECT, MVT::f16, Custom);
       setOperationAction(ISD::IS_FPCLASS, MVT::f16, Custom);
     } else {
-      static const unsigned ZfhminPromoteOps[] = {
-          ISD::FMINNUM,      ISD::FMAXNUM,       ISD::FADD,
-          ISD::FSUB,         ISD::FMUL,          ISD::FMA,
-          ISD::FDIV,         ISD::FSQRT,         ISD::FABS,
-          ISD::FNEG,         ISD::STRICT_FMA,    ISD::STRICT_FADD,
-          ISD::STRICT_FSUB,  ISD::STRICT_FMUL,   ISD::STRICT_FDIV,
-          ISD::STRICT_FSQRT, ISD::STRICT_FSETCC, ISD::STRICT_FSETCCS,
-          ISD::SETCC,        ISD::FCEIL,         ISD::FFLOOR,
-          ISD::FTRUNC,       ISD::FRINT,         ISD::FROUND,
-          ISD::FROUNDEVEN,   ISD::SELECT};
-
-      setOperationAction(ZfhminPromoteOps, MVT::f16, Promote);
+      setOperationAction(ZfhminZfbfminPromoteOps, MVT::f16, Promote);
       setOperationAction({ISD::STRICT_LRINT, ISD::STRICT_LLRINT,
                           ISD::STRICT_LROUND, ISD::STRICT_LLROUND},
                          MVT::f16, Legal);
       // FIXME: Need to promote f16 FCOPYSIGN to f32, but the
       // DAGCombiner::visitFP_ROUND probably needs improvements first.
       setOperationAction(ISD::FCOPYSIGN, MVT::f16, Expand);
     }
 

llvm/test/CodeGen/RISCV/bfloat-arith.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-zfbfmin -verify-machineinstrs \
+; RUN:   -target-abi ilp32f < %s | FileCheck -check-prefixes=CHECK,RV32IZFBFMIN %s
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-zfbfmin -verify-machineinstrs \
+; RUN:   -target-abi lp64f < %s | FileCheck -check-prefixes=CHECK,RV64IZFBFMIN %s
+
+; These tests descend from float-arith.ll, where each function was targeted at
+; a particular RISC-V FPU instruction.
+
+define bfloat @fadd_s(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fadd_s:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    fadd.s fa5, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %1 = fadd bfloat %a, %b
+  ret bfloat %1
+}
+
+define bfloat @fsub_s(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fsub_s:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    fsub.s fa5, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %1 = fsub bfloat %a, %b
+  ret bfloat %1
+}
+
+define bfloat @fmul_s(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fmul_s:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    fmul.s fa5, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %1 = fmul bfloat %a, %b
+  ret bfloat %1
+}
+
+define bfloat @fdiv_s(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fdiv_s:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    fdiv.s fa5, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %1 = fdiv bfloat %a, %b
+  ret bfloat %1
+}
+
+declare bfloat @llvm.sqrt.bf16(bfloat)
+
+define bfloat @fsqrt_s(bfloat %a) nounwind {
+; CHECK-LABEL: fsqrt_s:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa0
+; CHECK-NEXT:    fsqrt.s fa5, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %1 = call bfloat @llvm.sqrt.bf16(bfloat %a)
+  ret bfloat %1
+}
+
+declare bfloat @llvm.copysign.bf16(bfloat, bfloat)
+
+define bfloat @fsgnj_s(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: fsgnj_s:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    fsh fa1, 12(sp)
+; RV32IZFBFMIN-NEXT:    fsh fa0, 8(sp)
+; RV32IZFBFMIN-NEXT:    lbu a0, 13(sp)
+; RV32IZFBFMIN-NEXT:    lbu a1, 9(sp)
+; RV32IZFBFMIN-NEXT:    andi a0, a0, 128
+; RV32IZFBFMIN-NEXT:    andi a1, a1, 127
+; RV32IZFBFMIN-NEXT:    or a0, a1, a0
+; RV32IZFBFMIN-NEXT:    sb a0, 9(sp)
+; RV32IZFBFMIN-NEXT:    flh fa0, 8(sp)
+; RV32IZFBFMIN-NEXT:    addi sp, sp, 16
+; RV32IZFBFMIN-NEXT:    ret
+;
+; RV64IZFBFMIN-LABEL: fsgnj_s:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    fsh fa1, 8(sp)
+; RV64IZFBFMIN-NEXT:    fsh fa0, 0(sp)
+; RV64IZFBFMIN-NEXT:    lbu a0, 9(sp)
+; RV64IZFBFMIN-NEXT:    lbu a1, 1(sp)
+; RV64IZFBFMIN-NEXT:    andi a0, a0, 128
+; RV64IZFBFMIN-NEXT:    andi a1, a1, 127
+; RV64IZFBFMIN-NEXT:    or a0, a1, a0
+; RV64IZFBFMIN-NEXT:    sb a0, 1(sp)
+; RV64IZFBFMIN-NEXT:    flh fa0, 0(sp)
+; RV64IZFBFMIN-NEXT:    addi sp, sp, 16
+; RV64IZFBFMIN-NEXT:    ret
+  %1 = call bfloat @llvm.copysign.bf16(bfloat %a, bfloat %b)
+  ret bfloat %1
+}
+
+define i32 @fneg_s(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fneg_s:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa0
+; CHECK-NEXT:    fadd.s fa5, fa5, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fneg.s fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa4, fa4
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa4
+; CHECK-NEXT:    feq.s a0, fa5, fa4
+; CHECK-NEXT:    ret
+  %1 = fadd bfloat %a, %a
+  %2 = fneg bfloat %1
+  %3 = fcmp oeq bfloat %1, %2
+  %4 = zext i1 %3 to i32
+  ret i32 %4
+}
+
+define bfloat @fsgnjn_s(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: fsgnjn_s:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV32IZFBFMIN-NEXT:    fadd.s fa5, fa4, fa5
+; RV32IZFBFMIN-NEXT:    fcvt.bf16.s fa5, fa5
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa5
+; RV32IZFBFMIN-NEXT:    fneg.s fa5, fa5
+; RV32IZFBFMIN-NEXT:    fcvt.bf16.s fa5, fa5
+; RV32IZFBFMIN-NEXT:    fsh fa0, 8(sp)
+; RV32IZFBFMIN-NEXT:    fsh fa5, 12(sp)
+; RV32IZFBFMIN-NEXT:    lbu a0, 9(sp)
+; RV32IZFBFMIN-NEXT:    lbu a1, 13(sp)
+; RV32IZFBFMIN-NEXT:    andi a0, a0, 127
+; RV32IZFBFMIN-NEXT:    andi a1, a1, 128
+; RV32IZFBFMIN-NEXT:    or a0, a0, a1
+; RV32IZFBFMIN-NEXT:    sb a0, 9(sp)
+; RV32IZFBFMIN-NEXT:    flh fa0, 8(sp)
+; RV32IZFBFMIN-NEXT:    addi sp, sp, 16
+; RV32IZFBFMIN-NEXT:    ret
+;
+; RV64IZFBFMIN-LABEL: fsgnjn_s:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV64IZFBFMIN-NEXT:    fadd.s fa5, fa4, fa5
+; RV64IZFBFMIN-NEXT:    fcvt.bf16.s fa5, fa5
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa5
+; RV64IZFBFMIN-NEXT:    fneg.s fa5, fa5
+; RV64IZFBFMIN-NEXT:    fcvt.bf16.s fa5, fa5
+; RV64IZFBFMIN-NEXT:    fsh fa0, 0(sp)
+; RV64IZFBFMIN-NEXT:    fsh fa5, 8(sp)
+; RV64IZFBFMIN-NEXT:    lbu a0, 1(sp)
+; RV64IZFBFMIN-NEXT:    lbu a1, 9(sp)
+; RV64IZFBFMIN-NEXT:    andi a0, a0, 127
+; RV64IZFBFMIN-NEXT:    andi a1, a1, 128
+; RV64IZFBFMIN-NEXT:    or a0, a0, a1
+; RV64IZFBFMIN-NEXT:    sb a0, 1(sp)
+; RV64IZFBFMIN-NEXT:    flh fa0, 0(sp)
+; RV64IZFBFMIN-NEXT:    addi sp, sp, 16
+; RV64IZFBFMIN-NEXT:    ret
+  %1 = fadd bfloat %a, %b
+  %2 = fneg bfloat %1
+  %3 = call bfloat @llvm.copysign.bf16(bfloat %a, bfloat %2)
+  ret bfloat %3
+}
+
+declare bfloat @llvm.fabs.bf16(bfloat)
+
+define bfloat @fabs_s(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fabs_s:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    fadd.s fa5, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fabs.s fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa4, fa4
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa4
+; CHECK-NEXT:    fadd.s fa5, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %1 = fadd bfloat %a, %b
+  %2 = call bfloat @llvm.fabs.bf16(bfloat %1)
+  %3 = fadd bfloat %2, %1
+  ret bfloat %3
+}
+
+declare bfloat @llvm.minnum.bf16(bfloat, bfloat)
+
+define bfloat @fmin_s(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fmin_s:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    fmin.s fa5, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %1 = call bfloat @llvm.minnum.bf16(bfloat %a, bfloat %b)
+  ret bfloat %1
+}
+
+declare bfloat @llvm.maxnum.bf16(bfloat, bfloat)
+
+define bfloat @fmax_s(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fmax_s:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    fmax.s fa5, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %1 = call bfloat @llvm.maxnum.bf16(bfloat %a, bfloat %b)
+  ret bfloat %1
+}
+
+declare bfloat @llvm.fma.bf16(bfloat, bfloat, bfloat)
+
+define bfloat @fmadd_s(bfloat %a, bfloat %b, bfloat %c) nounwind {
+; CHECK-LABEL: fmadd_s:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa2
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa3, fa0
+; CHECK-NEXT:    fmadd.s fa5, fa3, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %1 = call bfloat @llvm.fma.bf16(bfloat %a, bfloat %b, bfloat %c)
+  ret bfloat %1
+}
+
+define bfloat @fmsub_s(bfloat %a, bfloat %b, bfloat %c) nounwind {
+; CHECK-LABEL: fmsub_s:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa2
+; CHECK-NEXT:    fmv.w.x fa4, zero
+; CHECK-NEXT:    fadd.s fa5, fa5, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fneg.s fa5, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa3, fa0
+; CHECK-NEXT:    fmadd.s fa5, fa3, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %c_ = fadd bfloat 0.0, %c ; avoid negation using xor
+  %negc = fsub bfloat -0.0, %c_
+  %1 = call bfloat @llvm.fma.bf16(bfloat %a, bfloat %b, bfloat %negc)
+  ret bfloat %1
+}
+
+define bfloat @fnmadd_s(bfloat %a, bfloat %b, bfloat %c) nounwind {
+; CHECK-LABEL: fnmadd_s:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa0
+; CHECK-NEXT:    fmv.w.x fa4, zero
+; CHECK-NEXT:    fadd.s fa5, fa5, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa3, fa2
+; CHECK-NEXT:    fadd.s fa4, fa3, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa4, fa4
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fneg.s fa5, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa4
+; CHECK-NEXT:    fneg.s fa4, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa4, fa4
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa4
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa3, fa1
+; CHECK-NEXT:    fmadd.s fa5, fa5, fa3, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %a_ = fadd bfloat 0.0, %a
+  %c_ = fadd bfloat 0.0, %c
+  %nega = fsub bfloat -0.0, %a_
+  %negc = fsub bfloat -0.0, %c_
+  %1 = call bfloat @llvm.fma.bf16(bfloat %nega, bfloat %b, bfloat %negc)
+  ret bfloat %1
+}
+
+define bfloat @fnmadd_s_2(bfloat %a, bfloat %b, bfloat %c) nounwind {
+; CHECK-LABEL: fnmadd_s_2:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fmv.w.x fa4, zero
+; CHECK-NEXT:    fadd.s fa5, fa5, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa3, fa2
+; CHECK-NEXT:    fadd.s fa4, fa3, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa4, fa4
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fneg.s fa5, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa4
+; CHECK-NEXT:    fneg.s fa4, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa4, fa4
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa4
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa3, fa0
+; CHECK-NEXT:    fmadd.s fa5, fa3, fa5, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %b_ = fadd bfloat 0.0, %b
+  %c_ = fadd bfloat 0.0, %c
+  %negb = fsub bfloat -0.0, %b_
+  %negc = fsub bfloat -0.0, %c_
+  %1 = call bfloat @llvm.fma.bf16(bfloat %a, bfloat %negb, bfloat %negc)
+  ret bfloat %1
+}
+
+define bfloat @fnmadd_s_3(bfloat %a, bfloat %b, bfloat %c) nounwind {
+; CHECK-LABEL: fnmadd_s_3:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa2
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa3, fa0
+; CHECK-NEXT:    fmadd.s fa5, fa3, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fneg.s fa5, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %1 = call bfloat @llvm.fma.bf16(bfloat %a, bfloat %b, bfloat %c)
+  %neg = fneg bfloat %1
+  ret bfloat %neg
+}
+
+
+define bfloat @fnmadd_nsz(bfloat %a, bfloat %b, bfloat %c) nounwind {
+; CHECK-LABEL: fnmadd_nsz:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa2
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa3, fa0
+; CHECK-NEXT:    fmadd.s fa5, fa3, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fneg.s fa5, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %1 = call nsz bfloat @llvm.fma.bf16(bfloat %a, bfloat %b, bfloat %c)
+  %neg = fneg nsz bfloat %1
+  ret bfloat %neg
+}
+
+define bfloat @fnmsub_s(bfloat %a, bfloat %b, bfloat %c) nounwind {
+; CHECK-LABEL: fnmsub_s:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa0
+; CHECK-NEXT:    fmv.w.x fa4, zero
+; CHECK-NEXT:    fadd.s fa5, fa5, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fneg.s fa5, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa2
+; CHECK-NEXT:    fcvt.s.bf16 fa3, fa1
+; CHECK-NEXT:    fmadd.s fa5, fa5, fa3, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %a_ = fadd bfloat 0.0, %a
+  %nega = fsub bfloat -0.0, %a_
+  %1 = call bfloat @llvm.fma.bf16(bfloat %nega, bfloat %b, bfloat %c)
+  ret bfloat %1
+}
+
+define bfloat @fnmsub_s_2(bfloat %a, bfloat %b, bfloat %c) nounwind {
+; CHECK-LABEL: fnmsub_s_2:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fmv.w.x fa4, zero
+; CHECK-NEXT:    fadd.s fa5, fa5, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fneg.s fa5, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa2
+; CHECK-NEXT:    fcvt.s.bf16 fa3, fa0
+; CHECK-NEXT:    fmadd.s fa5, fa3, fa5, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %b_ = fadd bfloat 0.0, %b
+  %negb = fsub bfloat -0.0, %b_
+  %1 = call bfloat @llvm.fma.bf16(bfloat %a, bfloat %negb, bfloat %c)
+  ret bfloat %1
+}
+
+define bfloat @fmadd_s_contract(bfloat %a, bfloat %b, bfloat %c) nounwind {
+; CHECK-LABEL: fmadd_s_contract:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    fmul.s fa5, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa2
+; CHECK-NEXT:    fadd.s fa5, fa5, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %1 = fmul contract bfloat %a, %b
+  %2 = fadd contract bfloat %1, %c
+  ret bfloat %2
+}
+
+define bfloat @fmsub_s_contract(bfloat %a, bfloat %b, bfloat %c) nounwind {
+; CHECK-LABEL: fmsub_s_contract:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa2
+; CHECK-NEXT:    fmv.w.x fa4, zero
+; CHECK-NEXT:    fadd.s fa5, fa5, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa3, fa0
+; CHECK-NEXT:    fmul.s fa4, fa3, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa4, fa4
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa4
+; CHECK-NEXT:    fsub.s fa5, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %c_ = fadd bfloat 0.0, %c ; avoid negation using xor
+  %1 = fmul contract bfloat %a, %b
+  %2 = fsub contract bfloat %1, %c_
+  ret bfloat %2
+}
+
+define bfloat @fnmadd_s_contract(bfloat %a, bfloat %b, bfloat %c) nounwind {
+; CHECK-LABEL: fnmadd_s_contract:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa0
+; CHECK-NEXT:    fmv.w.x fa4, zero
+; CHECK-NEXT:    fadd.s fa5, fa5, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa3, fa1
+; CHECK-NEXT:    fadd.s fa3, fa3, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa3, fa3
+; CHECK-NEXT:    fcvt.s.bf16 fa2, fa2
+; CHECK-NEXT:    fadd.s fa4, fa2, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa4, fa4
+; CHECK-NEXT:    fcvt.s.bf16 fa3, fa3
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fmul.s fa5, fa5, fa3
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fneg.s fa5, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa4
+; CHECK-NEXT:    fsub.s fa5, fa5, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %a_ = fadd bfloat 0.0, %a ; avoid negation using xor
+  %b_ = fadd bfloat 0.0, %b ; avoid negation using xor
+  %c_ = fadd bfloat 0.0, %c ; avoid negation using xor
+  %1 = fmul contract bfloat %a_, %b_
+  %2 = fneg bfloat %1
+  %3 = fsub contract bfloat %2, %c_
+  ret bfloat %3
+}
+
+define bfloat @fnmsub_s_contract(bfloat %a, bfloat %b, bfloat %c) nounwind {
+; CHECK-LABEL: fnmsub_s_contract:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa0
+; CHECK-NEXT:    fmv.w.x fa4, zero
+; CHECK-NEXT:    fadd.s fa5, fa5, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa3, fa1
+; CHECK-NEXT:    fadd.s fa4, fa3, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa4, fa4
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa4
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fmul.s fa5, fa5, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa2
+; CHECK-NEXT:    fsub.s fa5, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %a_ = fadd bfloat 0.0, %a ; avoid negation using xor
+  %b_ = fadd bfloat 0.0, %b ; avoid negation using xor
+  %1 = fmul contract bfloat %a_, %b_
+  %2 = fsub contract bfloat %c, %1
+  ret bfloat %2
+}

llvm/test/CodeGen/RISCV/bfloat-br-fcmp.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-zfbfmin -verify-machineinstrs \
+; RUN:   -target-abi ilp32f < %s | FileCheck -check-prefix=RV32IZFBFMIN %s
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-zfbfmin -verify-machineinstrs \
+; RUN:   -target-abi lp64f < %s | FileCheck -check-prefix=RV64IZFBFMIN %s
+
+declare void @abort()
+declare void @exit(i32)
+declare bfloat @dummy(bfloat)
+
+define void @br_fcmp_false(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_false:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    li a0, 1
+; RV32IZFBFMIN-NEXT:    bnez a0, .LBB0_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.then
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB0_2: # %if.else
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_false:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    li a0, 1
+; RV64IZFBFMIN-NEXT:    bnez a0, .LBB0_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.then
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB0_2: # %if.else
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp false bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.then:
+  ret void
+if.else:
+  tail call void @abort()
+  unreachable
+}
+
+define void @br_fcmp_oeq(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_oeq:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV32IZFBFMIN-NEXT:    feq.s a0, fa4, fa5
+; RV32IZFBFMIN-NEXT:    bnez a0, .LBB1_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB1_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_oeq:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV64IZFBFMIN-NEXT:    feq.s a0, fa4, fa5
+; RV64IZFBFMIN-NEXT:    bnez a0, .LBB1_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB1_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp oeq bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.else:
+  ret void
+if.then:
+  tail call void @abort()
+  unreachable
+}
+
+define void @br_fcmp_oeq_alt(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_oeq_alt:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV32IZFBFMIN-NEXT:    feq.s a0, fa4, fa5
+; RV32IZFBFMIN-NEXT:    bnez a0, .LBB2_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB2_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_oeq_alt:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV64IZFBFMIN-NEXT:    feq.s a0, fa4, fa5
+; RV64IZFBFMIN-NEXT:    bnez a0, .LBB2_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB2_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp oeq bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.then:
+  tail call void @abort()
+  unreachable
+if.else:
+  ret void
+}
+
+define void @br_fcmp_ogt(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_ogt:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa0
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa1
+; RV32IZFBFMIN-NEXT:    flt.s a0, fa4, fa5
+; RV32IZFBFMIN-NEXT:    bnez a0, .LBB3_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB3_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_ogt:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa0
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa1
+; RV64IZFBFMIN-NEXT:    flt.s a0, fa4, fa5
+; RV64IZFBFMIN-NEXT:    bnez a0, .LBB3_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB3_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp ogt bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.else:
+  ret void
+if.then:
+  tail call void @abort()
+  unreachable
+}
+
+define void @br_fcmp_oge(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_oge:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa0
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa1
+; RV32IZFBFMIN-NEXT:    fle.s a0, fa4, fa5
+; RV32IZFBFMIN-NEXT:    bnez a0, .LBB4_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB4_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_oge:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa0
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa1
+; RV64IZFBFMIN-NEXT:    fle.s a0, fa4, fa5
+; RV64IZFBFMIN-NEXT:    bnez a0, .LBB4_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB4_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp oge bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.else:
+  ret void
+if.then:
+  tail call void @abort()
+  unreachable
+}
+
+define void @br_fcmp_olt(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_olt:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV32IZFBFMIN-NEXT:    flt.s a0, fa4, fa5
+; RV32IZFBFMIN-NEXT:    bnez a0, .LBB5_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB5_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_olt:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV64IZFBFMIN-NEXT:    flt.s a0, fa4, fa5
+; RV64IZFBFMIN-NEXT:    bnez a0, .LBB5_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB5_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp olt bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.else:
+  ret void
+if.then:
+  tail call void @abort()
+  unreachable
+}
+
+define void @br_fcmp_ole(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_ole:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV32IZFBFMIN-NEXT:    fle.s a0, fa4, fa5
+; RV32IZFBFMIN-NEXT:    bnez a0, .LBB6_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB6_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_ole:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV64IZFBFMIN-NEXT:    fle.s a0, fa4, fa5
+; RV64IZFBFMIN-NEXT:    bnez a0, .LBB6_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB6_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp ole bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.else:
+  ret void
+if.then:
+  tail call void @abort()
+  unreachable
+}
+
+define void @br_fcmp_one(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_one:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV32IZFBFMIN-NEXT:    flt.s a0, fa4, fa5
+; RV32IZFBFMIN-NEXT:    flt.s a1, fa5, fa4
+; RV32IZFBFMIN-NEXT:    or a0, a1, a0
+; RV32IZFBFMIN-NEXT:    bnez a0, .LBB7_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB7_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_one:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV64IZFBFMIN-NEXT:    flt.s a0, fa4, fa5
+; RV64IZFBFMIN-NEXT:    flt.s a1, fa5, fa4
+; RV64IZFBFMIN-NEXT:    or a0, a1, a0
+; RV64IZFBFMIN-NEXT:    bnez a0, .LBB7_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB7_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp one bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.else:
+  ret void
+if.then:
+  tail call void @abort()
+  unreachable
+}
+
+define void @br_fcmp_ord(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_ord:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV32IZFBFMIN-NEXT:    feq.s a0, fa5, fa5
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa0
+; RV32IZFBFMIN-NEXT:    feq.s a1, fa5, fa5
+; RV32IZFBFMIN-NEXT:    and a0, a1, a0
+; RV32IZFBFMIN-NEXT:    bnez a0, .LBB8_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB8_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_ord:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV64IZFBFMIN-NEXT:    feq.s a0, fa5, fa5
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa0
+; RV64IZFBFMIN-NEXT:    feq.s a1, fa5, fa5
+; RV64IZFBFMIN-NEXT:    and a0, a1, a0
+; RV64IZFBFMIN-NEXT:    bnez a0, .LBB8_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB8_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp ord bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.else:
+  ret void
+if.then:
+  tail call void @abort()
+  unreachable
+}
+
+define void @br_fcmp_ueq(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_ueq:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV32IZFBFMIN-NEXT:    flt.s a0, fa4, fa5
+; RV32IZFBFMIN-NEXT:    flt.s a1, fa5, fa4
+; RV32IZFBFMIN-NEXT:    or a0, a1, a0
+; RV32IZFBFMIN-NEXT:    beqz a0, .LBB9_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB9_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_ueq:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV64IZFBFMIN-NEXT:    flt.s a0, fa4, fa5
+; RV64IZFBFMIN-NEXT:    flt.s a1, fa5, fa4
+; RV64IZFBFMIN-NEXT:    or a0, a1, a0
+; RV64IZFBFMIN-NEXT:    beqz a0, .LBB9_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB9_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp ueq bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.else:
+  ret void
+if.then:
+  tail call void @abort()
+  unreachable
+}
+
+define void @br_fcmp_ugt(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_ugt:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV32IZFBFMIN-NEXT:    fle.s a0, fa4, fa5
+; RV32IZFBFMIN-NEXT:    beqz a0, .LBB10_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB10_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_ugt:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV64IZFBFMIN-NEXT:    fle.s a0, fa4, fa5
+; RV64IZFBFMIN-NEXT:    beqz a0, .LBB10_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB10_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp ugt bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.else:
+  ret void
+if.then:
+  tail call void @abort()
+  unreachable
+}
+
+define void @br_fcmp_uge(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_uge:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV32IZFBFMIN-NEXT:    flt.s a0, fa4, fa5
+; RV32IZFBFMIN-NEXT:    beqz a0, .LBB11_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB11_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_uge:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV64IZFBFMIN-NEXT:    flt.s a0, fa4, fa5
+; RV64IZFBFMIN-NEXT:    beqz a0, .LBB11_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB11_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp uge bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.else:
+  ret void
+if.then:
+  tail call void @abort()
+  unreachable
+}
+
+define void @br_fcmp_ult(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_ult:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa0
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa1
+; RV32IZFBFMIN-NEXT:    fle.s a0, fa4, fa5
+; RV32IZFBFMIN-NEXT:    beqz a0, .LBB12_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB12_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_ult:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa0
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa1
+; RV64IZFBFMIN-NEXT:    fle.s a0, fa4, fa5
+; RV64IZFBFMIN-NEXT:    beqz a0, .LBB12_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB12_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp ult bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.else:
+  ret void
+if.then:
+  tail call void @abort()
+  unreachable
+}
+
+define void @br_fcmp_ule(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_ule:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa0
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa1
+; RV32IZFBFMIN-NEXT:    flt.s a0, fa4, fa5
+; RV32IZFBFMIN-NEXT:    beqz a0, .LBB13_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB13_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_ule:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa0
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa1
+; RV64IZFBFMIN-NEXT:    flt.s a0, fa4, fa5
+; RV64IZFBFMIN-NEXT:    beqz a0, .LBB13_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB13_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp ule bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.else:
+  ret void
+if.then:
+  tail call void @abort()
+  unreachable
+}
+
+define void @br_fcmp_une(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_une:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV32IZFBFMIN-NEXT:    feq.s a0, fa4, fa5
+; RV32IZFBFMIN-NEXT:    beqz a0, .LBB14_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB14_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_une:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV64IZFBFMIN-NEXT:    feq.s a0, fa4, fa5
+; RV64IZFBFMIN-NEXT:    beqz a0, .LBB14_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB14_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp une bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.else:
+  ret void
+if.then:
+  tail call void @abort()
+  unreachable
+}
+
+define void @br_fcmp_uno(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_uno:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV32IZFBFMIN-NEXT:    feq.s a0, fa5, fa5
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa0
+; RV32IZFBFMIN-NEXT:    feq.s a1, fa5, fa5
+; RV32IZFBFMIN-NEXT:    and a0, a1, a0
+; RV32IZFBFMIN-NEXT:    beqz a0, .LBB15_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB15_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_uno:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV64IZFBFMIN-NEXT:    feq.s a0, fa5, fa5
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa0
+; RV64IZFBFMIN-NEXT:    feq.s a1, fa5, fa5
+; RV64IZFBFMIN-NEXT:    and a0, a1, a0
+; RV64IZFBFMIN-NEXT:    beqz a0, .LBB15_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB15_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp uno bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.else:
+  ret void
+if.then:
+  tail call void @abort()
+  unreachable
+}
+
+define void @br_fcmp_true(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: br_fcmp_true:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    li a0, 1
+; RV32IZFBFMIN-NEXT:    bnez a0, .LBB16_2
+; RV32IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV32IZFBFMIN-NEXT:    ret
+; RV32IZFBFMIN-NEXT:  .LBB16_2: # %if.then
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    call abort@plt
+;
+; RV64IZFBFMIN-LABEL: br_fcmp_true:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    li a0, 1
+; RV64IZFBFMIN-NEXT:    bnez a0, .LBB16_2
+; RV64IZFBFMIN-NEXT:  # %bb.1: # %if.else
+; RV64IZFBFMIN-NEXT:    ret
+; RV64IZFBFMIN-NEXT:  .LBB16_2: # %if.then
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    call abort@plt
+  %1 = fcmp true bfloat %a, %b
+  br i1 %1, label %if.then, label %if.else
+if.else:
+  ret void
+if.then:
+  tail call void @abort()
+  unreachable
+}

llvm/test/CodeGen/RISCV/bfloat-fcmp.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-zfbfmin -verify-machineinstrs \
+; RUN:   -target-abi ilp32f < %s | FileCheck %s
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-zfbfmin -verify-machineinstrs \
+; RUN:   -target-abi lp64f < %s | FileCheck %s
+
+define i32 @fcmp_false(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_false:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    li a0, 0
+; CHECK-NEXT:    ret
+  %1 = fcmp false bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_oeq(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_oeq:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    feq.s a0, fa4, fa5
+; CHECK-NEXT:    ret
+  %1 = fcmp oeq bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ogt(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_ogt:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa0
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa1
+; CHECK-NEXT:    flt.s a0, fa4, fa5
+; CHECK-NEXT:    ret
+  %1 = fcmp ogt bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_oge(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_oge:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa0
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa1
+; CHECK-NEXT:    fle.s a0, fa4, fa5
+; CHECK-NEXT:    ret
+  %1 = fcmp oge bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_olt(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_olt:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    flt.s a0, fa4, fa5
+; CHECK-NEXT:    ret
+  %1 = fcmp olt bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ole(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_ole:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    fle.s a0, fa4, fa5
+; CHECK-NEXT:    ret
+  %1 = fcmp ole bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_one(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_one:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    flt.s a0, fa4, fa5
+; CHECK-NEXT:    flt.s a1, fa5, fa4
+; CHECK-NEXT:    or a0, a1, a0
+; CHECK-NEXT:    ret
+  %1 = fcmp one bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ord(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_ord:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    feq.s a0, fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa0
+; CHECK-NEXT:    feq.s a1, fa5, fa5
+; CHECK-NEXT:    and a0, a1, a0
+; CHECK-NEXT:    ret
+  %1 = fcmp ord bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ueq(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_ueq:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    flt.s a0, fa4, fa5
+; CHECK-NEXT:    flt.s a1, fa5, fa4
+; CHECK-NEXT:    or a0, a1, a0
+; CHECK-NEXT:    xori a0, a0, 1
+; CHECK-NEXT:    ret
+  %1 = fcmp ueq bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ugt(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_ugt:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    fle.s a0, fa4, fa5
+; CHECK-NEXT:    xori a0, a0, 1
+; CHECK-NEXT:    ret
+  %1 = fcmp ugt bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_uge(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_uge:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    flt.s a0, fa4, fa5
+; CHECK-NEXT:    xori a0, a0, 1
+; CHECK-NEXT:    ret
+  %1 = fcmp uge bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ult(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_ult:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa0
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa1
+; CHECK-NEXT:    fle.s a0, fa4, fa5
+; CHECK-NEXT:    xori a0, a0, 1
+; CHECK-NEXT:    ret
+  %1 = fcmp ult bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ule(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_ule:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa0
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa1
+; CHECK-NEXT:    flt.s a0, fa4, fa5
+; CHECK-NEXT:    xori a0, a0, 1
+; CHECK-NEXT:    ret
+  %1 = fcmp ule bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_une(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_une:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    feq.s a0, fa4, fa5
+; CHECK-NEXT:    xori a0, a0, 1
+; CHECK-NEXT:    ret
+  %1 = fcmp une bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_uno(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_uno:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    feq.s a0, fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa0
+; CHECK-NEXT:    feq.s a1, fa5, fa5
+; CHECK-NEXT:    and a0, a1, a0
+; CHECK-NEXT:    xori a0, a0, 1
+; CHECK-NEXT:    ret
+  %1 = fcmp uno bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_true(bfloat %a, bfloat %b) nounwind {
+; CHECK-LABEL: fcmp_true:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    li a0, 1
+; CHECK-NEXT:    ret
+  %1 = fcmp true bfloat %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}

llvm/test/CodeGen/RISCV/bfloat-frem.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-zfbfmin -verify-machineinstrs \
+; RUN:   -target-abi ilp32f < %s | FileCheck -check-prefix=RV32IZFBFMIN %s
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-zfbfmin -verify-machineinstrs \
+; RUN:   -target-abi lp64f < %s | FileCheck -check-prefix=RV64IZFBFMIN %s
+
+define bfloat @frem_bf16(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: frem_bf16:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa0, fa0
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa1, fa1
+; RV32IZFBFMIN-NEXT:    call fmodf@plt
+; RV32IZFBFMIN-NEXT:    fcvt.bf16.s fa0, fa0
+; RV32IZFBFMIN-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IZFBFMIN-NEXT:    addi sp, sp, 16
+; RV32IZFBFMIN-NEXT:    ret
+;
+; RV64IZFBFMIN-LABEL: frem_bf16:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa0, fa0
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa1, fa1
+; RV64IZFBFMIN-NEXT:    call fmodf@plt
+; RV64IZFBFMIN-NEXT:    fcvt.bf16.s fa0, fa0
+; RV64IZFBFMIN-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64IZFBFMIN-NEXT:    addi sp, sp, 16
+; RV64IZFBFMIN-NEXT:    ret
+  %1 = frem bfloat %a, %b
+  ret bfloat %1
+}

llvm/test/CodeGen/RISCV/bfloat-imm.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-zfbfmin -verify-machineinstrs \
+; RUN:   -target-abi ilp32f < %s | FileCheck %s
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-zfbfmin -verify-machineinstrs \
+; RUN:   -target-abi lp64f < %s | FileCheck %s
+
+define bfloat @bfloat_imm() nounwind {
+; CHECK-LABEL: bfloat_imm:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lui a0, %hi(.LCPI0_0)
+; CHECK-NEXT:    flh fa0, %lo(.LCPI0_0)(a0)
+; CHECK-NEXT:    ret
+  ret bfloat 3.0
+}
+
+define bfloat @bfloat_imm_op(bfloat %a) nounwind {
+; CHECK-LABEL: bfloat_imm_op:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa0
+; CHECK-NEXT:    lui a0, 260096
+; CHECK-NEXT:    fmv.w.x fa4, a0
+; CHECK-NEXT:    fadd.s fa5, fa5, fa4
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %1 = fadd bfloat %a, 1.0
+  ret bfloat %1
+}
+
+define bfloat @bfloat_zero() nounwind {
+; CHECK-LABEL: bfloat_zero:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lui a0, %hi(.LCPI2_0)
+; CHECK-NEXT:    flh fa0, %lo(.LCPI2_0)(a0)
+; CHECK-NEXT:    ret
+  ret bfloat 0.0
+}
+
+define bfloat @bfloat_negative_zero() nounwind {
+; CHECK-LABEL: bfloat_negative_zero:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lui a0, %hi(.LCPI3_0)
+; CHECK-NEXT:    flh fa0, %lo(.LCPI3_0)(a0)
+; CHECK-NEXT:    ret
+  ret bfloat -0.0
+}

llvm/test/CodeGen/RISCV/bfloat-isnan.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-zfbfmin -verify-machineinstrs \
+; RUN:   -target-abi ilp32f < %s | FileCheck %s
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-zfbfmin -verify-machineinstrs \
+; RUN:   -target-abi lp64f < %s | FileCheck %s
+
+define zeroext i1 @bfloat_is_nan(bfloat %a) nounwind {
+; CHECK-LABEL: bfloat_is_nan:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa0
+; CHECK-NEXT:    feq.s a0, fa5, fa5
+; CHECK-NEXT:    xori a0, a0, 1
+; CHECK-NEXT:    ret
+  %1 = fcmp uno bfloat %a, 0.000000e+00
+  ret i1 %1
+}
+
+define zeroext i1 @bfloat_not_nan(bfloat %a) nounwind {
+; CHECK-LABEL: bfloat_not_nan:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa0
+; CHECK-NEXT:    feq.s a0, fa5, fa5
+; CHECK-NEXT:    ret
+  %1 = fcmp ord bfloat %a, 0.000000e+00
+  ret i1 %1
+}

llvm/test/CodeGen/RISCV/bfloat-mem.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-zfbfmin -verify-machineinstrs \
+; RUN:   -target-abi ilp32f < %s | FileCheck -check-prefixes=CHECK,RV32IZFBFMIN %s
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-zfbfmin -verify-machineinstrs \
+; RUN:   -target-abi lp64f < %s | FileCheck -check-prefixes=CHECK,RV64IZFBFMIN %s
+
+define bfloat @flh(ptr %a) nounwind {
+; CHECK-LABEL: flh:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    flh fa5, 6(a0)
+; CHECK-NEXT:    flh fa4, 0(a0)
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa5
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa4
+; CHECK-NEXT:    fadd.s fa5, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    ret
+  %1 = load bfloat, ptr %a
+  %2 = getelementptr bfloat, ptr %a, i32 3
+  %3 = load bfloat, ptr %2
+; Use both loaded values in an FP op to ensure an flh is used, even for the
+; soft bfloat ABI
+  %4 = fadd bfloat %1, %3
+  ret bfloat %4
+}
+
+define dso_local void @fsh(ptr %a, bfloat %b, bfloat %c) nounwind {
+; CHECK-LABEL: fsh:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    fadd.s fa5, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa5, fa5
+; CHECK-NEXT:    fsh fa5, 0(a0)
+; CHECK-NEXT:    fsh fa5, 16(a0)
+; CHECK-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
+}
+
+; Check load and store to a global
+@G = dso_local global bfloat 0.0
+
+define bfloat @flh_fsh_global(bfloat %a, bfloat %b) nounwind {
+; Use %a and %b in an FP op to ensure bfloat precision floating point registers
+; are used, even for the soft bfloat ABI
+; CHECK-LABEL: flh_fsh_global:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fcvt.s.bf16 fa5, fa1
+; CHECK-NEXT:    fcvt.s.bf16 fa4, fa0
+; CHECK-NEXT:    fadd.s fa5, fa4, fa5
+; CHECK-NEXT:    fcvt.bf16.s fa0, fa5
+; CHECK-NEXT:    lui a0, %hi(G)
+; CHECK-NEXT:    flh fa5, %lo(G)(a0)
+; CHECK-NEXT:    addi a1, a0, %lo(G)
+; CHECK-NEXT:    fsh fa0, %lo(G)(a0)
+; CHECK-NEXT:    flh fa5, 18(a1)
+; CHECK-NEXT:    fsh fa0, 18(a1)
+; CHECK-NEXT:    ret
+  %1 = fadd bfloat %a, %b
+  %2 = load volatile bfloat, ptr @G
+  store bfloat %1, ptr @G
+  %3 = getelementptr bfloat, ptr @G, i32 9
+  %4 = load volatile bfloat, ptr %3
+  store bfloat %1, ptr %3
+  ret bfloat %1
+}
+
+; Ensure that 1 is added to the high 20 bits if bit 11 of the low part is 1
+define bfloat @flh_fsh_constant(bfloat %a) nounwind {
+; RV32IZFBFMIN-LABEL: flh_fsh_constant:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    lui a0, 912092
+; RV32IZFBFMIN-NEXT:    flh fa5, -273(a0)
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa5
+; RV32IZFBFMIN-NEXT:    fadd.s fa5, fa4, fa5
+; RV32IZFBFMIN-NEXT:    fcvt.bf16.s fa0, fa5
+; RV32IZFBFMIN-NEXT:    fsh fa0, -273(a0)
+; RV32IZFBFMIN-NEXT:    ret
+;
+; RV64IZFBFMIN-LABEL: flh_fsh_constant:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    lui a0, 228023
+; RV64IZFBFMIN-NEXT:    slli a0, a0, 2
+; RV64IZFBFMIN-NEXT:    flh fa5, -273(a0)
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa5
+; RV64IZFBFMIN-NEXT:    fadd.s fa5, fa4, fa5
+; RV64IZFBFMIN-NEXT:    fcvt.bf16.s fa0, fa5
+; RV64IZFBFMIN-NEXT:    fsh fa0, -273(a0)
+; RV64IZFBFMIN-NEXT:    ret
+  %1 = inttoptr i32 3735928559 to ptr
+  %2 = load volatile bfloat, ptr %1
+  %3 = fadd bfloat %a, %2
+  store bfloat %3, ptr %1
+  ret bfloat %3
+}
+
+declare void @notdead(ptr)
+
+define bfloat @flh_stack(bfloat %a) nounwind {
+; RV32IZFBFMIN-LABEL: flh_stack:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    fsw fs0, 8(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    fmv.s fs0, fa0
+; RV32IZFBFMIN-NEXT:    addi a0, sp, 4
+; RV32IZFBFMIN-NEXT:    call notdead@plt
+; RV32IZFBFMIN-NEXT:    flh fa5, 4(sp)
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fs0
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa5
+; RV32IZFBFMIN-NEXT:    fadd.s fa5, fa5, fa4
+; RV32IZFBFMIN-NEXT:    fcvt.bf16.s fa0, fa5
+; RV32IZFBFMIN-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IZFBFMIN-NEXT:    flw fs0, 8(sp) # 4-byte Folded Reload
+; RV32IZFBFMIN-NEXT:    addi sp, sp, 16
+; RV32IZFBFMIN-NEXT:    ret
+;
+; RV64IZFBFMIN-LABEL: flh_stack:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    fmv.s fs0, fa0
+; RV64IZFBFMIN-NEXT:    mv a0, sp
+; RV64IZFBFMIN-NEXT:    call notdead@plt
+; RV64IZFBFMIN-NEXT:    flh fa5, 0(sp)
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fs0
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa5
+; RV64IZFBFMIN-NEXT:    fadd.s fa5, fa5, fa4
+; RV64IZFBFMIN-NEXT:    fcvt.bf16.s fa0, fa5
+; RV64IZFBFMIN-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64IZFBFMIN-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
+; RV64IZFBFMIN-NEXT:    addi sp, sp, 16
+; RV64IZFBFMIN-NEXT:    ret
+  %1 = alloca bfloat, align 4
+  call void @notdead(ptr %1)
+  %2 = load bfloat, ptr %1
+  %3 = fadd bfloat %2, %a ; force load in to FPR16
+  ret bfloat %3
+}
+
+define dso_local void @fsh_stack(bfloat %a, bfloat %b) nounwind {
+; RV32IZFBFMIN-LABEL: fsh_stack:
+; RV32IZFBFMIN:       # %bb.0:
+; RV32IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV32IZFBFMIN-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV32IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV32IZFBFMIN-NEXT:    fadd.s fa5, fa4, fa5
+; RV32IZFBFMIN-NEXT:    fcvt.bf16.s fa5, fa5
+; RV32IZFBFMIN-NEXT:    fsh fa5, 8(sp)
+; RV32IZFBFMIN-NEXT:    addi a0, sp, 8
+; RV32IZFBFMIN-NEXT:    call notdead@plt
+; RV32IZFBFMIN-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IZFBFMIN-NEXT:    addi sp, sp, 16
+; RV32IZFBFMIN-NEXT:    ret
+;
+; RV64IZFBFMIN-LABEL: fsh_stack:
+; RV64IZFBFMIN:       # %bb.0:
+; RV64IZFBFMIN-NEXT:    addi sp, sp, -16
+; RV64IZFBFMIN-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa5, fa1
+; RV64IZFBFMIN-NEXT:    fcvt.s.bf16 fa4, fa0
+; RV64IZFBFMIN-NEXT:    fadd.s fa5, fa4, fa5
+; RV64IZFBFMIN-NEXT:    fcvt.bf16.s fa5, fa5
+; RV64IZFBFMIN-NEXT:    fsh fa5, 4(sp)
+; RV64IZFBFMIN-NEXT:    addi a0, sp, 4
+; RV64IZFBFMIN-NEXT:    call notdead@plt
+; RV64IZFBFMIN-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64IZFBFMIN-NEXT:    addi sp, sp, 16
+; RV64IZFBFMIN-NEXT:    ret
+  %1 = fadd bfloat %a, %b ; force store from FPR16
+  %2 = alloca bfloat, align 4
+  store bfloat %1, ptr %2
+  call void @notdead(ptr %2)
+  ret void
+}