[RISCV] Add expansions for floating-point intrinsics

[luismarques]

Summary

A number of intrinsics, such as llvm.sin.f32, would result in a failure to select. This patch adds expansions for the relevant selection DAG nodes, as well as exhaustive testing for all f32 and f64 intrinsics.

The codegen for FMA remains a TODO item, pending support for the various RISC-V FMA instruction variants.

The llvm.minimum.f32.* and llvm.maximum.* tests are commented-out, pending upstream support for target-independent expansion, as discussed in http://lists.llvm.org/pipermail/llvm-dev/2018-November/127408.html.

[asb]

Thanks Luis, this looks good to me. Just a minor nitpick re OP -> Op.

Expanding FMA to a libcall is better than a compiler assertion, so it makes sense to add this as a minimal incremental improvement and then follow-up with a patch adding patterns for FMAs.

lib/Target/RISCV/RISCVISelLowering.cpp

116

Nitpick: should be FPOpToExtend

127

OP -> Op

141

OP -> Op

[luismarques]

Fixes some capitalization issues, as suggested by Alex Bradbury.

lib/Target/RISCV/RISCVISelLowering.cpp

RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM,
   setOperationAction(ISD::CTLZ, XLenVT, Expand);
   setOperationAction(ISD::CTPOP, XLenVT, Expand);
 
   ISD::CondCode FPCCToExtend[] = {
       ISD::SETOGT, ISD::SETOGE, ISD::SETONE, ISD::SETO,   ISD::SETUEQ,
       ISD::SETUGT, ISD::SETUGE, ISD::SETULT, ISD::SETULE, ISD::SETUNE,
       ISD::SETGT,  ISD::SETGE,  ISD::SETNE};
 
+  // TODO: add proper support for the various FMA variants
+  // (FMADD.S, FMSUB.S, FNMSUB.S, FNMADD.S).
+  ISD::NodeType FPOPToExtend[] = {
+      ISD::FSIN, ISD::FCOS, ISD::FSINCOS, ISD::FPOW, ISD::FMA};
+
   if (Subtarget.hasStdExtF()) {
     setOperationAction(ISD::FMINNUM, MVT::f32, Legal);
     setOperationAction(ISD::FMAXNUM, MVT::f32, Legal);
     for (auto CC : FPCCToExtend)
       setCondCodeAction(CC, MVT::f32, Expand);
     setOperationAction(ISD::SELECT_CC, MVT::f32, Expand);
     setOperationAction(ISD::SELECT, MVT::f32, Custom);
     setOperationAction(ISD::BR_CC, MVT::f32, Expand);
+    for (auto OP : FPOPToExtend)
+      setOperationAction(OP, MVT::f32, Expand);
   }
 
   if (Subtarget.hasStdExtD()) {
     setOperationAction(ISD::FMINNUM, MVT::f64, Legal);
     setOperationAction(ISD::FMAXNUM, MVT::f64, Legal);
     for (auto CC : FPCCToExtend)
       setCondCodeAction(CC, MVT::f64, Expand);
     setOperationAction(ISD::SELECT_CC, MVT::f64, Expand);
     setOperationAction(ISD::SELECT, MVT::f64, Custom);
     setOperationAction(ISD::BR_CC, MVT::f64, Expand);
     setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f32, Expand);
     setTruncStoreAction(MVT::f64, MVT::f32, Expand);
+    for (auto OP : FPOPToExtend)
+      setOperationAction(OP, MVT::f64, Expand);
   }
 
   setOperationAction(ISD::GlobalAddress, XLenVT, Custom);
   setOperationAction(ISD::BlockAddress, XLenVT, Custom);
   setOperationAction(ISD::ConstantPool, XLenVT, Custom);
 
   if (Subtarget.hasStdExtA()) {
     setMaxAtomicSizeInBitsSupported(Subtarget.getXLen());

test/CodeGen/RISCV/double-intrinsics.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=riscv32 -mattr=+d -verify-machineinstrs < %s \
 ; RUN:   | FileCheck -check-prefix=RV32IFD %s
 
-declare double @llvm.floor.f64(double)
+declare double @llvm.sqrt.f64(double)
+
+define double @sqrt_f64(double %a) {
+; RV32IFD-LABEL: sqrt_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw a0, 8(sp)
+; RV32IFD-NEXT:    sw a1, 12(sp)
+; RV32IFD-NEXT:    fld ft0, 8(sp)
+; RV32IFD-NEXT:    fsqrt.d ft0, ft0
+; RV32IFD-NEXT:    fsd ft0, 8(sp)
+; RV32IFD-NEXT:    lw a0, 8(sp)
+; RV32IFD-NEXT:    lw a1, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.sqrt.f64(double %a)
+  ret double %1
+}
+
+declare double @llvm.powi.f64(double, i32)
+
+define double @powi_f64(double %a, i32 %b) {
+; RV32IFD-LABEL: powi_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp)
+; RV32IFD-NEXT:    call __powidf2
+; RV32IFD-NEXT:    lw ra, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.powi.f64(double %a, i32 %b)
+  ret double %1
+}
+
+declare double @llvm.sin.f64(double)
+
+define double @sin_f64(double %a) {
+; RV32IFD-LABEL: sin_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp)
+; RV32IFD-NEXT:    call sin
+; RV32IFD-NEXT:    lw ra, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.sin.f64(double %a)
+  ret double %1
+}
+
+declare double @llvm.cos.f64(double)
+
+define double @cos_f64(double %a) {
+; RV32IFD-LABEL: cos_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp)
+; RV32IFD-NEXT:    call cos
+; RV32IFD-NEXT:    lw ra, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.cos.f64(double %a)
+  ret double %1
+}
+
+; The sin+cos combination results in an FSINCOS SelectionDAG node.
+define double @sincos_f64(double %a) {
+; RV32IFD-LABEL: sincos_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -32
+; RV32IFD-NEXT:    sw ra, 28(sp)
+; RV32IFD-NEXT:    sw s1, 24(sp)
+; RV32IFD-NEXT:    sw s2, 20(sp)
+; RV32IFD-NEXT:    sw s3, 16(sp)
+; RV32IFD-NEXT:    sw s4, 12(sp)
+; RV32IFD-NEXT:    mv s2, a1
+; RV32IFD-NEXT:    mv s1, a0
+; RV32IFD-NEXT:    call sin
+; RV32IFD-NEXT:    mv s3, a0
+; RV32IFD-NEXT:    mv s4, a1
+; RV32IFD-NEXT:    mv a0, s1
+; RV32IFD-NEXT:    mv a1, s2
+; RV32IFD-NEXT:    call cos
+; RV32IFD-NEXT:    sw a0, 0(sp)
+; RV32IFD-NEXT:    sw a1, 4(sp)
+; RV32IFD-NEXT:    fld ft0, 0(sp)
+; RV32IFD-NEXT:    sw s3, 0(sp)
+; RV32IFD-NEXT:    sw s4, 4(sp)
+; RV32IFD-NEXT:    fld ft1, 0(sp)
+; RV32IFD-NEXT:    fadd.d ft0, ft1, ft0
+; RV32IFD-NEXT:    fsd ft0, 0(sp)
+; RV32IFD-NEXT:    lw a0, 0(sp)
+; RV32IFD-NEXT:    lw a1, 4(sp)
+; RV32IFD-NEXT:    lw s4, 12(sp)
+; RV32IFD-NEXT:    lw s3, 16(sp)
+; RV32IFD-NEXT:    lw s2, 20(sp)
+; RV32IFD-NEXT:    lw s1, 24(sp)
+; RV32IFD-NEXT:    lw ra, 28(sp)
+; RV32IFD-NEXT:    addi sp, sp, 32
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.sin.f64(double %a)
+  %2 = call double @llvm.cos.f64(double %a)
+  %3 = fadd double %1, %2
+  ret double %3
+}
+
+declare double @llvm.pow.f64(double, double)
+
+define double @pow_f64(double %a, double %b) {
+; RV32IFD-LABEL: pow_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp)
+; RV32IFD-NEXT:    call pow
+; RV32IFD-NEXT:    lw ra, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.pow.f64(double %a, double %b)
+  ret double %1
+}
+
+declare double @llvm.exp.f64(double)
+
+define double @exp_f64(double %a) {
+; RV32IFD-LABEL: exp_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp)
+; RV32IFD-NEXT:    call exp
+; RV32IFD-NEXT:    lw ra, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.exp.f64(double %a)
+  ret double %1
+}
+
+declare double @llvm.exp2.f64(double)
+
+define double @exp2_f64(double %a) {
+; RV32IFD-LABEL: exp2_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp)
+; RV32IFD-NEXT:    call exp2
+; RV32IFD-NEXT:    lw ra, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.exp2.f64(double %a)
+  ret double %1
+}
+
+declare double @llvm.log.f64(double)
+
+define double @log_f64(double %a) {
+; RV32IFD-LABEL: log_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp)
+; RV32IFD-NEXT:    call log
+; RV32IFD-NEXT:    lw ra, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.log.f64(double %a)
+  ret double %1
+}
+
+declare double @llvm.log10.f64(double)
+
+define double @log10_f64(double %a) {
+; RV32IFD-LABEL: log10_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp)
+; RV32IFD-NEXT:    call log10
+; RV32IFD-NEXT:    lw ra, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.log10.f64(double %a)
+  ret double %1
+}
+
+declare double @llvm.log2.f64(double)
+
+define double @log2_f64(double %a) {
+; RV32IFD-LABEL: log2_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp)
+; RV32IFD-NEXT:    call log2
+; RV32IFD-NEXT:    lw ra, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.log2.f64(double %a)
+  ret double %1
+}
+
+declare double @llvm.fma.f64(double, double, double)
+
+; TODO: Select RISC-V FMA instruction.
+define double @fma_f64(double %a, double %b, double %c) {
+; RV32IFD-LABEL: fma_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp)
+; RV32IFD-NEXT:    call fma
+; RV32IFD-NEXT:    lw ra, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.fma.f64(double %a, double %b, double %c)
+  ret double %1
+}
+
+declare double @llvm.fabs.f64(double)
+
+define double @fabs_f64(double %a) {
+; RV32IFD-LABEL: fabs_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw a0, 8(sp)
+; RV32IFD-NEXT:    sw a1, 12(sp)
+; RV32IFD-NEXT:    fld ft0, 8(sp)
+; RV32IFD-NEXT:    fabs.d ft0, ft0
+; RV32IFD-NEXT:    fsd ft0, 8(sp)
+; RV32IFD-NEXT:    lw a0, 8(sp)
+; RV32IFD-NEXT:    lw a1, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.fabs.f64(double %a)
+  ret double %1
+}
+
+declare double @llvm.minnum.f64(double, double)
+
+define double @minnum_f64(double %a, double %b) nounwind {
+; RV32IFD-LABEL: minnum_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw a2, 8(sp)
+; RV32IFD-NEXT:    sw a3, 12(sp)
+; RV32IFD-NEXT:    fld ft0, 8(sp)
+; RV32IFD-NEXT:    sw a0, 8(sp)
+; RV32IFD-NEXT:    sw a1, 12(sp)
+; RV32IFD-NEXT:    fld ft1, 8(sp)
+; RV32IFD-NEXT:    fmin.d ft0, ft1, ft0
+; RV32IFD-NEXT:    fsd ft0, 8(sp)
+; RV32IFD-NEXT:    lw a0, 8(sp)
+; RV32IFD-NEXT:    lw a1, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.minnum.f64(double %a, double %b)
+  ret double %1
+}
+
+declare double @llvm.maxnum.f64(double, double)
+
+define double @maxnum_f64(double %a, double %b) nounwind {
+; RV32IFD-LABEL: maxnum_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw a2, 8(sp)
+; RV32IFD-NEXT:    sw a3, 12(sp)
+; RV32IFD-NEXT:    fld ft0, 8(sp)
+; RV32IFD-NEXT:    sw a0, 8(sp)
+; RV32IFD-NEXT:    sw a1, 12(sp)
+; RV32IFD-NEXT:    fld ft1, 8(sp)
+; RV32IFD-NEXT:    fmax.d ft0, ft1, ft0
+; RV32IFD-NEXT:    fsd ft0, 8(sp)
+; RV32IFD-NEXT:    lw a0, 8(sp)
+; RV32IFD-NEXT:    lw a1, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.maxnum.f64(double %a, double %b)
+  ret double %1
+}
+
+; TODO: FMINNAN and FMAXNAN aren't handled in
+; SelectionDAGLegalize::ExpandNode.
+
+; declare double @llvm.minimum.f64(double, double)
 
-; The call to ffloor is introduced very late, meaning this test case covers
-; aspects of passing f64 on RV32D soft-float that double-calling-conv.ll
-; doesn't.
+; define double @fminimum_f64(double %a, double %b) nounwind {
+;   %1 = call double @llvm.minimum.f64(double %a, double %b)
+;   ret double %1
+; }
+
+; declare double @llvm.maximum.f64(double, double)
+
+; define double @fmaximum_f64(double %a, double %b) nounwind {
+;   %1 = call double @llvm.maximum.f64(double %a, double %b)
+;   ret double %1
+; }
+
+declare double @llvm.copysign.f64(double, double)
+
+define double @copysign_f64(double %a, double %b) nounwind {
+; RV32IFD-LABEL: copysign_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw a2, 8(sp)
+; RV32IFD-NEXT:    sw a3, 12(sp)
+; RV32IFD-NEXT:    fld ft0, 8(sp)
+; RV32IFD-NEXT:    sw a0, 8(sp)
+; RV32IFD-NEXT:    sw a1, 12(sp)
+; RV32IFD-NEXT:    fld ft1, 8(sp)
+; RV32IFD-NEXT:    fsgnj.d ft0, ft1, ft0
+; RV32IFD-NEXT:    fsd ft0, 8(sp)
+; RV32IFD-NEXT:    lw a0, 8(sp)
+; RV32IFD-NEXT:    lw a1, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.copysign.f64(double %a, double %b)
+  ret double %1
+}
+
+declare double @llvm.floor.f64(double)
 
-define double @foo(double %a) nounwind {
-; RV32IFD-LABEL: foo:
+define double @floor_f64(double %a) {
+; RV32IFD-LABEL: floor_f64:
 ; RV32IFD:       # %bb.0:
 ; RV32IFD-NEXT:    addi sp, sp, -16
 ; RV32IFD-NEXT:    sw ra, 12(sp)
 ; RV32IFD-NEXT:    call floor
 ; RV32IFD-NEXT:    lw ra, 12(sp)
 ; RV32IFD-NEXT:    addi sp, sp, 16
 ; RV32IFD-NEXT:    ret
   %1 = call double @llvm.floor.f64(double %a)
-  ret double %1
+  ret double %1
 }
+
+declare double @llvm.ceil.f64(double)
+
+define double @ceil_f64(double %a) {
+; RV32IFD-LABEL: ceil_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp)
+; RV32IFD-NEXT:    call ceil
+; RV32IFD-NEXT:    lw ra, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.ceil.f64(double %a)
+  ret double %1
+}
+
+declare double @llvm.trunc.f64(double)
+
+define double @trunc_f64(double %a) {
+; RV32IFD-LABEL: trunc_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp)
+; RV32IFD-NEXT:    call trunc
+; RV32IFD-NEXT:    lw ra, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.trunc.f64(double %a)
+  ret double %1
+}
+
+declare double @llvm.rint.f64(double)
+
+define double @rint_f64(double %a) {
+; RV32IFD-LABEL: rint_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp)
+; RV32IFD-NEXT:    call rint
+; RV32IFD-NEXT:    lw ra, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.rint.f64(double %a)
+  ret double %1
+}
+
+declare double @llvm.nearbyint.f64(double)
+
+define double @nearbyint_f64(double %a) {
+; RV32IFD-LABEL: nearbyint_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp)
+; RV32IFD-NEXT:    call nearbyint
+; RV32IFD-NEXT:    lw ra, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.nearbyint.f64(double %a)
+  ret double %1
+}
+
+declare double @llvm.round.f64(double)
+
+define double @round_f64(double %a) {
+; RV32IFD-LABEL: round_f64:
+; RV32IFD:       # %bb.0:
+; RV32IFD-NEXT:    addi sp, sp, -16
+; RV32IFD-NEXT:    sw ra, 12(sp)
+; RV32IFD-NEXT:    call round
+; RV32IFD-NEXT:    lw ra, 12(sp)
+; RV32IFD-NEXT:    addi sp, sp, 16
+; RV32IFD-NEXT:    ret
+  %1 = call double @llvm.round.f64(double %a)
+  ret double %1
+}

test/CodeGen/RISCV/float-intrinsics.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+f -verify-machineinstrs < %s \
+; RUN:   | FileCheck -check-prefix=RV32IF %s
+; RUN: llc -mtriple=riscv32 -mattr=+d -verify-machineinstrs < %s \
+; RUN:   | FileCheck -check-prefix=RV32IF %s
+
+declare float @llvm.sqrt.f32(float)
+
+define float @sqrt_f32(float %a) {
+; RV32IF-LABEL: sqrt_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    fmv.w.x ft0, a0
+; RV32IF-NEXT:    fsqrt.s ft0, ft0
+; RV32IF-NEXT:    fmv.x.w a0, ft0
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.sqrt.f32(float %a)
+  ret float %1
+}
+
+declare float @llvm.powi.f32(float, i32)
+
+define float @powi_f32(float %a, i32 %b) {
+; RV32IF-LABEL: powi_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call __powisf2
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.powi.f32(float %a, i32 %b)
+  ret float %1
+}
+
+declare float @llvm.sin.f32(float)
+
+define float @sin_f32(float %a) {
+; RV32IF-LABEL: sin_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call sinf
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.sin.f32(float %a)
+  ret float %1
+}
+
+declare float @llvm.cos.f32(float)
+
+define float @cos_f32(float %a) {
+; RV32IF-LABEL: cos_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call cosf
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.cos.f32(float %a)
+  ret float %1
+}
+
+; The sin+cos combination results in an FSINCOS SelectionDAG node.
+define float @sincos_f32(float %a) {
+; RV32IF-LABEL: sincos_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    sw s1, 8(sp)
+; RV32IF-NEXT:    sw s2, 4(sp)
+; RV32IF-NEXT:    mv s1, a0
+; RV32IF-NEXT:    call sinf
+; RV32IF-NEXT:    mv s2, a0
+; RV32IF-NEXT:    mv a0, s1
+; RV32IF-NEXT:    call cosf
+; RV32IF-NEXT:    fmv.w.x ft0, a0
+; RV32IF-NEXT:    fmv.w.x ft1, s2
+; RV32IF-NEXT:    fadd.s ft0, ft1, ft0
+; RV32IF-NEXT:    fmv.x.w a0, ft0
+; RV32IF-NEXT:    lw s2, 4(sp)
+; RV32IF-NEXT:    lw s1, 8(sp)
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.sin.f32(float %a)
+  %2 = call float @llvm.cos.f32(float %a)
+  %3 = fadd float %1, %2
+  ret float %3
+}
+
+declare float @llvm.pow.f32(float, float)
+
+define float @pow_f32(float %a, float %b) {
+; RV32IF-LABEL: pow_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call powf
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.pow.f32(float %a, float %b)
+  ret float %1
+}
+
+declare float @llvm.exp.f32(float)
+
+define float @exp_f32(float %a) {
+; RV32IF-LABEL: exp_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call expf
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.exp.f32(float %a)
+  ret float %1
+}
+
+declare float @llvm.exp2.f32(float)
+
+define float @exp2_f32(float %a) {
+; RV32IF-LABEL: exp2_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call exp2f
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.exp2.f32(float %a)
+  ret float %1
+}
+
+declare float @llvm.log.f32(float)
+
+define float @log_f32(float %a) {
+; RV32IF-LABEL: log_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call logf
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.log.f32(float %a)
+  ret float %1
+}
+
+declare float @llvm.log10.f32(float)
+
+define float @log10_f32(float %a) {
+; RV32IF-LABEL: log10_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call log10f
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.log10.f32(float %a)
+  ret float %1
+}
+
+declare float @llvm.log2.f32(float)
+
+define float @log2_f32(float %a) {
+; RV32IF-LABEL: log2_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call log2f
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.log2.f32(float %a)
+  ret float %1
+}
+
+declare float @llvm.fma.f32(float, float, float)
+
+; TODO: Select RISC-V FMA instruction.
+define float @fma_f32(float %a, float %b, float %c) {
+; RV32IF-LABEL: fma_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call fmaf
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.fma.f32(float %a, float %b, float %c)
+  ret float %1
+}
+
+declare float @llvm.fabs.f32(float)
+
+define float @fabs_f32(float %a) {
+; RV32IF-LABEL: fabs_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    lui a1, 524288
+; RV32IF-NEXT:    addi a1, a1, -1
+; RV32IF-NEXT:    and a0, a0, a1
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.fabs.f32(float %a)
+  ret float %1
+}
+
+declare float @llvm.minnum.f32(float, float)
+
+define float @minnum_f32(float %a, float %b) nounwind {
+; RV32IF-LABEL: minnum_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    fmv.w.x ft0, a1
+; RV32IF-NEXT:    fmv.w.x ft1, a0
+; RV32IF-NEXT:    fmin.s ft0, ft1, ft0
+; RV32IF-NEXT:    fmv.x.w a0, ft0
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.minnum.f32(float %a, float %b)
+  ret float %1
+}
+
+declare float @llvm.maxnum.f32(float, float)
+
+define float @maxnum_f32(float %a, float %b) nounwind {
+; RV32IF-LABEL: maxnum_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    fmv.w.x ft0, a1
+; RV32IF-NEXT:    fmv.w.x ft1, a0
+; RV32IF-NEXT:    fmax.s ft0, ft1, ft0
+; RV32IF-NEXT:    fmv.x.w a0, ft0
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.maxnum.f32(float %a, float %b)
+  ret float %1
+}
+
+; TODO: FMINNAN and FMAXNAN aren't handled in
+; SelectionDAGLegalize::ExpandNode.
+
+; declare float @llvm.minimum.f32(float, float)
+
+; define float @fminimum_f32(float %a, float %b) nounwind {
+;   %1 = call float @llvm.minimum.f32(float %a, float %b)
+;   ret float %1
+; }
+
+; declare float @llvm.maximum.f32(float, float)
+
+; define float @fmaximum_f32(float %a, float %b) nounwind {
+;   %1 = call float @llvm.maximum.f32(float %a, float %b)
+;   ret float %1
+; }
+
+declare float @llvm.copysign.f32(float, float)
+
+define float @copysign_f32(float %a, float %b) nounwind {
+; RV32IF-LABEL: copysign_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    fmv.w.x ft0, a1
+; RV32IF-NEXT:    fmv.w.x ft1, a0
+; RV32IF-NEXT:    fsgnj.s ft0, ft1, ft0
+; RV32IF-NEXT:    fmv.x.w a0, ft0
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.copysign.f32(float %a, float %b)
+  ret float %1
+}
+
+declare float @llvm.floor.f32(float)
+
+define float @floor_f32(float %a) {
+; RV32IF-LABEL: floor_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call floorf
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.floor.f32(float %a)
+  ret float %1
+}
+
+declare float @llvm.ceil.f32(float)
+
+define float @ceil_f32(float %a) {
+; RV32IF-LABEL: ceil_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call ceilf
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.ceil.f32(float %a)
+  ret float %1
+}
+
+declare float @llvm.trunc.f32(float)
+
+define float @trunc_f32(float %a) {
+; RV32IF-LABEL: trunc_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call truncf
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.trunc.f32(float %a)
+  ret float %1
+}
+
+declare float @llvm.rint.f32(float)
+
+define float @rint_f32(float %a) {
+; RV32IF-LABEL: rint_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call rintf
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.rint.f32(float %a)
+  ret float %1
+}
+
+declare float @llvm.nearbyint.f32(float)
+
+define float @nearbyint_f32(float %a) {
+; RV32IF-LABEL: nearbyint_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call nearbyintf
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.nearbyint.f32(float %a)
+  ret float %1
+}
+
+declare float @llvm.round.f32(float)
+
+define float @round_f32(float %a) {
+; RV32IF-LABEL: round_f32:
+; RV32IF:       # %bb.0:
+; RV32IF-NEXT:    addi sp, sp, -16
+; RV32IF-NEXT:    sw ra, 12(sp)
+; RV32IF-NEXT:    call roundf
+; RV32IF-NEXT:    lw ra, 12(sp)
+; RV32IF-NEXT:    addi sp, sp, 16
+; RV32IF-NEXT:    ret
+  %1 = call float @llvm.round.f32(float %a)
+  ret float %1
+}