[RISCV][GlobalISel] Add lowerReturn for calling conv

[nitinjohnraj]

Summary

Add minimal support to lower return, and introduce an OutgoingValueHandler and an OutgoingValueAssigner for returns.

Supports return values with integer, pointer and aggregate types.

(Update of D69808 - avoiding commandeering that revision)

[arsenm]

llvm/lib/CodeGen/TargetRegisterInfo.cpp

237–239 ↗

(On Diff #399993)

This needs its own patch and justification

llvm/lib/Target/RISCV/RISCVCallLowering.cpp

80–84 ↗

(On Diff #399993)

This looks like you're copying the broken form of splitToValueTypes from x86. I think trying to handle all of the argument marshalling in simple looking callbacks like this is unworkable. If you call the generic splitToValueTypes in CallLowering, it should try to handle everything for you

132–159 ↗

(On Diff #399993)

This is reinventing generic infrastructure

161 ↗

(On Diff #399993)

Ditto, this is a less complete copy of the generic infrastructure

203 ↗

(On Diff #399993)

Ditto

[lewis-revill]

Don't reinvent existing generic functions; instead introduce an OutgoingValueAssigner to allow us to call determineAndHandleAssignments.

[arsenm]

LGTM

llvm/test/CodeGen/RISCV/GlobalISel/irtranslator/ret.ll

145

Probably should have some vectors

[nitinjohnraj]

Rebased

[eopXD]

llvm/lib/Target/RISCV/GISel/RISCVCallLowering.cpp

83

This looks redundant.

llvm/lib/Target/RISCV/RISCVISelLowering.h

749

Can we avoid this change?

878

Why do we need to promote the static functions into the header here?

[nitinjohnraj]

Addressing comments

[nitinjohnraj]

This has been stagnant for a while, and there were a few open comments, so could I reaffirm the approval of this patch?

llvm/lib/Target/RISCV/RISCVISelLowering.h

749

This change is in order to use RISCVCCAssignFn in RISCVOutgoingValueAssigner above. We could maybe avoid that by inlining the type instead of changing the access specifier, but should we?

878

In order to use them on line 111 in llvm/lib/Target/RISCV/GISel/RISCVCallLowering.cpp. Should we maybe move this outside the RISCV namespace?

llvm/test/CodeGen/RISCV/GlobalISel/irtranslator/ret.ll

145

If I understand correctly, this patch doesn't handle vectors.

[arsenm]

llvm/test/CodeGen/RISCV/GlobalISel/irtranslator/ret.ll

103–104

New tests must use opaque pointers

[nitinjohnraj]

Used opaque pointer in test

[nitinjohnraj]

Fixed test

[craig.topper]

llvm/lib/Target/RISCV/GISel/RISCVCallLowering.cpp

123

Drop curly braces

llvm/lib/Target/RISCV/GISel/RISCVCallLowering.h

24

alphabetize the forward declarations

[nitinjohnraj]

Addressed comments

[nitinjohnraj]

Added lewis as co-author in commit message

llvm/lib/Target/RISCV/GISel/RISCVCallLowering.h

 //
 //===----------------------------------------------------------------------===//
 
 #ifndef LLVM_LIB_TARGET_RISCV_RISCVCALLLOWERING_H
 #define LLVM_LIB_TARGET_RISCV_RISCVCALLLOWERING_H
 
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/GlobalISel/CallLowering.h"
-#include "llvm/CodeGen/ValueTypes.h"
 
 namespace llvm {
 
+class MachineInstrBuilder;
+class MachineIRBuilder;
 class RISCVTargetLowering;
 
 class RISCVCallLowering : public CallLowering {
 
 public:
   RISCVCallLowering(const RISCVTargetLowering &TLI);
 
   bool lowerReturn(MachineIRBuilder &MIRBuiler, const Value *Val,
                    ArrayRef<Register> VRegs,
                    FunctionLoweringInfo &FLI) const override;
 
   bool lowerFormalArguments(MachineIRBuilder &MIRBuilder, const Function &F,
                             ArrayRef<ArrayRef<Register>> VRegs,
                             FunctionLoweringInfo &FLI) const override;
 
   bool lowerCall(MachineIRBuilder &MIRBuilder,
                  CallLoweringInfo &Info) const override;
+
+private:
+  bool lowerReturnVal(MachineIRBuilder &MIRBuilder, const Value *Val,
+                      ArrayRef<Register> VRegs, MachineInstrBuilder &Ret) const;
 };
 
 } // end namespace llvm
 
 #endif // LLVM_LIB_TARGET_RISCV_RISCVCALLLOWERING_H

llvm/lib/Target/RISCV/GISel/RISCVCallLowering.cpp

 //===-- RISCVCallLowering.cpp - Call lowering -------------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 /// \file
 /// This file implements the lowering of LLVM calls to machine code calls for
 /// GlobalISel.
 //
 //===----------------------------------------------------------------------===//
 
 #include "RISCVCallLowering.h"
 #include "RISCVISelLowering.h"
+#include "RISCVSubtarget.h"
+#include "llvm/CodeGen/Analysis.h"
 #include "llvm/CodeGen/GlobalISel/MachineIRBuilder.h"
 
 using namespace llvm;
 
+namespace {
+
+struct RISCVOutgoingValueAssigner : public CallLowering::OutgoingValueAssigner {
+private:
+  // The function used internally to assign args - we ignore the AssignFn stored
+  // by OutgoingValueAssigner since RISC-V implements its CC using a custom
+  // function with a different signature.
+  RISCVTargetLowering::RISCVCCAssignFn *RISCVAssignFn;
+
+  // Whether this is assigning args for a return.
+  bool IsRet;
+
+public:
+  RISCVOutgoingValueAssigner(
+      RISCVTargetLowering::RISCVCCAssignFn *RISCVAssignFn_, bool IsRet)
+      : CallLowering::OutgoingValueAssigner(nullptr),
+        RISCVAssignFn(RISCVAssignFn_), IsRet(IsRet) {}
+
+  bool assignArg(unsigned ValNo, EVT OrigVT, MVT ValVT, MVT LocVT,
+                 CCValAssign::LocInfo LocInfo,
+                 const CallLowering::ArgInfo &Info, ISD::ArgFlagsTy Flags,
+                 CCState &State) override {
+    MachineFunction &MF = State.getMachineFunction();
+    const DataLayout &DL = MF.getDataLayout();
+    const RISCVSubtarget &Subtarget = MF.getSubtarget<RISCVSubtarget>();
+
+    return RISCVAssignFn(DL, Subtarget.getTargetABI(), ValNo, ValVT, LocVT,
+                         LocInfo, Flags, State, /*IsFixed=*/true, IsRet,
+                         Info.Ty, *Subtarget.getTargetLowering(),
+                         /*FirstMaskArgument=*/std::nullopt);
+  }
+};
+
+struct RISCVOutgoingValueHandler : public CallLowering::OutgoingValueHandler {
+  RISCVOutgoingValueHandler(MachineIRBuilder &B, MachineRegisterInfo &MRI,
+                            MachineInstrBuilder MIB)
+      : OutgoingValueHandler(B, MRI), MIB(MIB) {}
+
+  MachineInstrBuilder MIB;
+
+  Register getStackAddress(uint64_t MemSize, int64_t Offset,
+                           MachinePointerInfo &MPO,
+                           ISD::ArgFlagsTy Flags) override {
+    llvm_unreachable("not implemented");
+  }
+
+  void assignValueToAddress(Register ValVReg, Register Addr, LLT MemTy,
+                            MachinePointerInfo &MPO, CCValAssign &VA) override {
+    llvm_unreachable("not implemented");
+  }
+
+  void assignValueToReg(Register ValVReg, Register PhysReg,
+                        CCValAssign VA) override {
+    Register ExtReg = extendRegister(ValVReg, VA);
+    MIRBuilder.buildCopy(PhysReg, ExtReg);
+    MIB.addUse(PhysReg, RegState::Implicit);
+  }
+};
+
+} // namespace
+
 RISCVCallLowering::RISCVCallLowering(const RISCVTargetLowering &TLI)
     : CallLowering(&TLI) {}
 
+bool RISCVCallLowering::lowerReturnVal(MachineIRBuilder &MIRBuilder,
+                                       const Value *Val,
+                                       ArrayRef<Register> VRegs,
+                                       MachineInstrBuilder &Ret) const {
+  if (!Val)
+    return true;
+
+  // TODO: Only integer, pointer and aggregate types are supported now.
+  if (!Val->getType()->isIntOrPtrTy() && !Val->getType()->isAggregateType())
+    return false;
+
+  MachineFunction &MF = MIRBuilder.getMF();
+  const DataLayout &DL = MF.getDataLayout();
+  const Function &F = MF.getFunction();
+  CallingConv::ID CC = F.getCallingConv();
+
+  ArgInfo OrigRetInfo(VRegs, Val->getType(), 0);
+  setArgFlags(OrigRetInfo, AttributeList::ReturnIndex, DL, F);
+
+  SmallVector<ArgInfo, 4> SplitRetInfos;
+  splitToValueTypes(OrigRetInfo, SplitRetInfos, DL, CC);
+
+  RISCVOutgoingValueAssigner Assigner(
+      CC == CallingConv::Fast ? RISCV::CC_RISCV_FastCC : RISCV::CC_RISCV,
+      /*IsRet=*/true);
+  RISCVOutgoingValueHandler Handler(MIRBuilder, MF.getRegInfo(), Ret);
+  return determineAndHandleAssignments(Handler, Assigner, SplitRetInfos,
+                                       MIRBuilder, CC, F.isVarArg());
+}
+
 bool RISCVCallLowering::lowerReturn(MachineIRBuilder &MIRBuilder,
                                     const Value *Val, ArrayRef<Register> VRegs,
                                     FunctionLoweringInfo &FLI) const {
-
+  assert(!Val == VRegs.empty() && "Return value without a vreg");
   MachineInstrBuilder Ret = MIRBuilder.buildInstrNoInsert(RISCV::PseudoRET);
 
-  if (Val != nullptr) {
+  if (!lowerReturnVal(MIRBuilder, Val, VRegs, Ret))
     return false;
-  }
+
   MIRBuilder.insertInstr(Ret);
   return true;
 }
 
 bool RISCVCallLowering::lowerFormalArguments(MachineIRBuilder &MIRBuilder,
                                              const Function &F,
                                              ArrayRef<ArrayRef<Register>> VRegs,
                                              FunctionLoweringInfo &FLI) const {

llvm/lib/Target/RISCV/RISCVISelLowering.h

public:
   bool lowerInterleavedLoad(LoadInst *LI,
                             ArrayRef<ShuffleVectorInst *> Shuffles,
                             ArrayRef<unsigned> Indices,
                             unsigned Factor) const override;
 
   bool lowerInterleavedStore(StoreInst *SI, ShuffleVectorInst *SVI,
                              unsigned Factor) const override;
 
-private:
   /// RISCVCCAssignFn - This target-specific function extends the default
   /// CCValAssign with additional information used to lower RISC-V calling
   /// conventions.
   typedef bool RISCVCCAssignFn(const DataLayout &DL, RISCVABI::ABI,
                                unsigned ValNo, MVT ValVT, MVT LocVT,
                                CCValAssign::LocInfo LocInfo,
                                ISD::ArgFlagsTy ArgFlags, CCState &State,
                                bool IsFixed, bool IsRet, Type *OrigTy,
                                const RISCVTargetLowering &TLI,
                                std::optional<unsigned> FirstMaskArgument);
 
+private:
   void analyzeInputArgs(MachineFunction &MF, CCState &CCInfo,
                         const SmallVectorImpl<ISD::InputArg> &Ins, bool IsRet,
                         RISCVCCAssignFn Fn) const;
   void analyzeOutputArgs(MachineFunction &MF, CCState &CCInfo,
                          const SmallVectorImpl<ISD::OutputArg> &Outs,
                          bool IsRet, CallLoweringInfo *CLI,
                          RISCVCCAssignFn Fn) const;
 
private:
   bool shouldNormalizeToSelectSequence(LLVMContext &, EVT) const override {
     return false;
   };
 
   /// For available scheduling models FDIV + two independent FMULs are much
   /// faster than two FDIVs.
   unsigned combineRepeatedFPDivisors() const override;
 };
+
+namespace RISCV {
+
+bool CC_RISCV(const DataLayout &DL, RISCVABI::ABI ABI, unsigned ValNo,
+              MVT ValVT, MVT LocVT, CCValAssign::LocInfo LocInfo,
+              ISD::ArgFlagsTy ArgFlags, CCState &State, bool IsFixed,
+              bool IsRet, Type *OrigTy, const RISCVTargetLowering &TLI,
+              std::optional<unsigned> FirstMaskArgument);
+
+bool CC_RISCV_FastCC(const DataLayout &DL, RISCVABI::ABI ABI, unsigned ValNo,
+                     MVT ValVT, MVT LocVT, CCValAssign::LocInfo LocInfo,
+                     ISD::ArgFlagsTy ArgFlags, CCState &State, bool IsFixed,
+                     bool IsRet, Type *OrigTy, const RISCVTargetLowering &TLI,
+                     std::optional<unsigned> FirstMaskArgument);
+
+bool CC_RISCV_GHC(unsigned ValNo, MVT ValVT, MVT LocVT,
+                  CCValAssign::LocInfo LocInfo, ISD::ArgFlagsTy ArgFlags,
+                  CCState &State);
+} // end namespace RISCV
+
 namespace RISCVVIntrinsicsTable {
 
 struct RISCVVIntrinsicInfo {
   unsigned IntrinsicID;
   uint8_t ScalarOperand;
   uint8_t VLOperand;
   bool hasScalarOperand() const {
     // 0xF is not valid. See NoScalarOperand in IntrinsicsRISCV.td.

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

static unsigned allocateRVVReg(MVT ValVT, unsigned ValNo,
   if (RC == &RISCV::VRM4RegClass)
     return State.AllocateReg(ArgVRM4s);
   if (RC == &RISCV::VRM8RegClass)
     return State.AllocateReg(ArgVRM8s);
   llvm_unreachable("Unhandled register class for ValueType");
 }
 
 // Implements the RISC-V calling convention. Returns true upon failure.
-static bool CC_RISCV(const DataLayout &DL, RISCVABI::ABI ABI, unsigned ValNo,
+bool RISCV::CC_RISCV(const DataLayout &DL, RISCVABI::ABI ABI, unsigned ValNo,
                      MVT ValVT, MVT LocVT, CCValAssign::LocInfo LocInfo,
                      ISD::ArgFlagsTy ArgFlags, CCState &State, bool IsFixed,
                      bool IsRet, Type *OrigTy, const RISCVTargetLowering &TLI,
                      std::optional<unsigned> FirstMaskArgument) {
   unsigned XLen = DL.getLargestLegalIntTypeSizeInBits();
   assert(XLen == 32 || XLen == 64);
   MVT XLenVT = XLen == 32 ? MVT::i32 : MVT::i64;
 
if (VA.getLocReg() == RISCV::X17) {
     RegInfo.addLiveIn(VA.getLocReg() + 1, HiVReg);
     Hi = DAG.getCopyFromReg(Chain, DL, HiVReg, MVT::i32);
   }
   return DAG.getNode(RISCVISD::BuildPairF64, DL, MVT::f64, Lo, Hi);
 }
 
 // FastCC has less than 1% performance improvement for some particular
 // benchmark. But theoretically, it may has benenfit for some cases.
-static bool CC_RISCV_FastCC(const DataLayout &DL, RISCVABI::ABI ABI,
+bool RISCV::CC_RISCV_FastCC(const DataLayout &DL, RISCVABI::ABI ABI,
                             unsigned ValNo, MVT ValVT, MVT LocVT,
                             CCValAssign::LocInfo LocInfo,
                             ISD::ArgFlagsTy ArgFlags, CCState &State,
                             bool IsFixed, bool IsRet, Type *OrigTy,
                             const RISCVTargetLowering &TLI,
                             std::optional<unsigned> FirstMaskArgument) {
 
   // X5 and X6 might be used for save-restore libcall.
if (LocVT.isVector()) {
     }
 
     return false;
   }
 
   return true; // CC didn't match.
 }
 
-static bool CC_RISCV_GHC(unsigned ValNo, MVT ValVT, MVT LocVT,
+bool RISCV::CC_RISCV_GHC(unsigned ValNo, MVT ValVT, MVT LocVT,
                          CCValAssign::LocInfo LocInfo,
                          ISD::ArgFlagsTy ArgFlags, CCState &State) {
 
   if (ArgFlags.isNest()) {
     report_fatal_error(
         "Attribute 'nest' is not supported in GHC calling convention");
   }
 
SDValue RISCVTargetLowering::LowerFormalArguments(
   // Used with vargs to acumulate store chains.
   std::vector<SDValue> OutChains;
 
   // Assign locations to all of the incoming arguments.
   SmallVector<CCValAssign, 16> ArgLocs;
   CCState CCInfo(CallConv, IsVarArg, MF, ArgLocs, *DAG.getContext());
 
   if (CallConv == CallingConv::GHC)
-    CCInfo.AnalyzeFormalArguments(Ins, CC_RISCV_GHC);
+    CCInfo.AnalyzeFormalArguments(Ins, RISCV::CC_RISCV_GHC);
   else
     analyzeInputArgs(MF, CCInfo, Ins, /*IsRet=*/false,
-                     CallConv == CallingConv::Fast ? CC_RISCV_FastCC
-                                                   : CC_RISCV);
+                     CallConv == CallingConv::Fast ? RISCV::CC_RISCV_FastCC
+                                                   : RISCV::CC_RISCV);
 
   for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
     CCValAssign &VA = ArgLocs[i];
     SDValue ArgValue;
     // Passing f64 on RV32D with a soft float ABI must be handled as a special
     // case.
     if (VA.getLocVT() == MVT::i32 && VA.getValVT() == MVT::f64)
       ArgValue = unpackF64OnRV32DSoftABI(DAG, Chain, VA, DL);
SDValue RISCVTargetLowering::LowerCall(CallLoweringInfo &CLI,
 
   MachineFunction &MF = DAG.getMachineFunction();
 
   // Analyze the operands of the call, assigning locations to each operand.
   SmallVector<CCValAssign, 16> ArgLocs;
   CCState ArgCCInfo(CallConv, IsVarArg, MF, ArgLocs, *DAG.getContext());
 
   if (CallConv == CallingConv::GHC)
-    ArgCCInfo.AnalyzeCallOperands(Outs, CC_RISCV_GHC);
+    ArgCCInfo.AnalyzeCallOperands(Outs, RISCV::CC_RISCV_GHC);
   else
     analyzeOutputArgs(MF, ArgCCInfo, Outs, /*IsRet=*/false, &CLI,
-                      CallConv == CallingConv::Fast ? CC_RISCV_FastCC
-                                                    : CC_RISCV);
+                      CallConv == CallingConv::Fast ? RISCV::CC_RISCV_FastCC
+                                                    : RISCV::CC_RISCV);
 
   // Check if it's really possible to do a tail call.
   if (IsTailCall)
     IsTailCall = isEligibleForTailCallOptimization(ArgCCInfo, CLI, MF, ArgLocs);
 
   if (IsTailCall)
     ++NumTailCalls;
   else if (CLI.CB && CLI.CB->isMustTailCall())
SDValue RISCVTargetLowering::LowerCall(CallLoweringInfo &CLI,
 
   // Mark the end of the call, which is glued to the call itself.
   Chain = DAG.getCALLSEQ_END(Chain, NumBytes, 0, Glue, DL);
   Glue = Chain.getValue(1);
 
   // Assign locations to each value returned by this call.
   SmallVector<CCValAssign, 16> RVLocs;
   CCState RetCCInfo(CallConv, IsVarArg, MF, RVLocs, *DAG.getContext());
-  analyzeInputArgs(MF, RetCCInfo, Ins, /*IsRet=*/true, CC_RISCV);
+  analyzeInputArgs(MF, RetCCInfo, Ins, /*IsRet=*/true, RISCV::CC_RISCV);
 
   // Copy all of the result registers out of their specified physreg.
   for (auto &VA : RVLocs) {
     // Copy the value out
     SDValue RetValue =
         DAG.getCopyFromReg(Chain, DL, VA.getLocReg(), VA.getLocVT(), Glue);
     // Glue the RetValue to the end of the call sequence
     Chain = RetValue.getValue(1);
bool RISCVTargetLowering::CanLowerReturn(
   std::optional<unsigned> FirstMaskArgument;
   if (Subtarget.hasVInstructions())
     FirstMaskArgument = preAssignMask(Outs);
 
   for (unsigned i = 0, e = Outs.size(); i != e; ++i) {
     MVT VT = Outs[i].VT;
     ISD::ArgFlagsTy ArgFlags = Outs[i].Flags;
     RISCVABI::ABI ABI = MF.getSubtarget<RISCVSubtarget>().getTargetABI();
-    if (CC_RISCV(MF.getDataLayout(), ABI, i, VT, VT, CCValAssign::Full,
+    if (RISCV::CC_RISCV(MF.getDataLayout(), ABI, i, VT, VT, CCValAssign::Full,
                  ArgFlags, CCInfo, /*IsFixed=*/true, /*IsRet=*/true, nullptr,
                  *this, FirstMaskArgument))
       return false;
   }
   return true;
 }
 
 SDValue
 RISCVTargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
                                  bool IsVarArg,
                                  const SmallVectorImpl<ISD::OutputArg> &Outs,
                                  const SmallVectorImpl<SDValue> &OutVals,
                                  const SDLoc &DL, SelectionDAG &DAG) const {
   MachineFunction &MF = DAG.getMachineFunction();
   const RISCVSubtarget &STI = MF.getSubtarget<RISCVSubtarget>();
 
   // Stores the assignment of the return value to a location.
   SmallVector<CCValAssign, 16> RVLocs;
 
   // Info about the registers and stack slot.
   CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(), RVLocs,
                  *DAG.getContext());
 
   analyzeOutputArgs(DAG.getMachineFunction(), CCInfo, Outs, /*IsRet=*/true,
-                    nullptr, CC_RISCV);
+                    nullptr, RISCV::CC_RISCV);
 
   if (CallConv == CallingConv::GHC && !RVLocs.empty())
     report_fatal_error("GHC functions return void only");
 
   SDValue Glue;
   SmallVector<SDValue, 4> RetOps(1, Chain);
 
   // Copy the result values into the output registers.

llvm/test/CodeGen/RISCV/GlobalISel/irtranslator/ret.ll

+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+; RUN: llc -mtriple=riscv32 -global-isel -stop-after=irtranslator -verify-machineinstrs < %s \
+; RUN:   | FileCheck -check-prefix=RV32I %s
+; RUN: llc -mtriple=riscv64 -global-isel -stop-after=irtranslator -verify-machineinstrs < %s \
+; RUN:   | FileCheck -check-prefix=RV64I %s
+
+define void @test_ret_void() {
+  ; RV32I-LABEL: name: test_ret_void
+  ; RV32I: bb.1.entry:
+  ; RV32I-NEXT:   PseudoRET
+  ; RV64I-LABEL: name: test_ret_void
+  ; RV64I: bb.1.entry:
+  ; RV64I-NEXT:   PseudoRET
+entry:
+  ret void
+}
+
+define i8 @test_ret_i8() {
+  ; RV32I-LABEL: name: test_ret_i8
+  ; RV32I: bb.1.entry:
+  ; RV32I-NEXT:   [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 1
+  ; RV32I-NEXT:   [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[C]](s8)
+  ; RV32I-NEXT:   $x10 = COPY [[ANYEXT]](s32)
+  ; RV32I-NEXT:   PseudoRET implicit $x10
+  ; RV64I-LABEL: name: test_ret_i8
+  ; RV64I: bb.1.entry:
+  ; RV64I-NEXT:   [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 1
+  ; RV64I-NEXT:   [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[C]](s8)
+  ; RV64I-NEXT:   $x10 = COPY [[ANYEXT]](s64)
+  ; RV64I-NEXT:   PseudoRET implicit $x10
+entry:
+  ret i8 1
+}
+
+define zeroext i8 @test_ret_i8_zext() {
+  ; RV32I-LABEL: name: test_ret_i8_zext
+  ; RV32I: bb.1.entry:
+  ; RV32I-NEXT:   [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 1
+  ; RV32I-NEXT:   [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[C]](s8)
+  ; RV32I-NEXT:   $x10 = COPY [[ZEXT]](s32)
+  ; RV32I-NEXT:   PseudoRET implicit $x10
+  ; RV64I-LABEL: name: test_ret_i8_zext
+  ; RV64I: bb.1.entry:
+  ; RV64I-NEXT:   [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 1
+  ; RV64I-NEXT:   [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[C]](s8)
+  ; RV64I-NEXT:   $x10 = COPY [[ZEXT]](s64)
+  ; RV64I-NEXT:   PseudoRET implicit $x10
+entry:
+  ret i8 1
+}
+
+define signext i16 @test_ret_i16_sext() {
+  ; RV32I-LABEL: name: test_ret_i16_sext
+  ; RV32I: bb.1.entry:
+  ; RV32I-NEXT:   [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 1
+  ; RV32I-NEXT:   [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[C]](s16)
+  ; RV32I-NEXT:   $x10 = COPY [[SEXT]](s32)
+  ; RV32I-NEXT:   PseudoRET implicit $x10
+  ; RV64I-LABEL: name: test_ret_i16_sext
+  ; RV64I: bb.1.entry:
+  ; RV64I-NEXT:   [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 1
+  ; RV64I-NEXT:   [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[C]](s16)
+  ; RV64I-NEXT:   $x10 = COPY [[SEXT]](s64)
+  ; RV64I-NEXT:   PseudoRET implicit $x10
+entry:
+  ret i16 1
+}
+
+define i32 @test_ret_i32() {
+  ; RV32I-LABEL: name: test_ret_i32
+  ; RV32I: bb.1.entry:
+  ; RV32I-NEXT:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+  ; RV32I-NEXT:   $x10 = COPY [[C]](s32)
+  ; RV32I-NEXT:   PseudoRET implicit $x10
+  ; RV64I-LABEL: name: test_ret_i32
+  ; RV64I: bb.1.entry:
+  ; RV64I-NEXT:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+  ; RV64I-NEXT:   [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[C]](s32)
+  ; RV64I-NEXT:   $x10 = COPY [[ANYEXT]](s64)
+  ; RV64I-NEXT:   PseudoRET implicit $x10
+entry:
+  ret i32 1
+}
+
+define i64 @test_ret_i64() {
+  ; RV32I-LABEL: name: test_ret_i64
+  ; RV32I: bb.1.entry:
+  ; RV32I-NEXT:   [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 4294967296
+  ; RV32I-NEXT:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[C]](s64)
+  ; RV32I-NEXT:   $x10 = COPY [[UV]](s32)
+  ; RV32I-NEXT:   $x11 = COPY [[UV1]](s32)
+  ; RV32I-NEXT:   PseudoRET implicit $x10, implicit $x11
+  ; RV64I-LABEL: name: test_ret_i64
+  ; RV64I: bb.1.entry:
+  ; RV64I-NEXT:   [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 4294967296
+  ; RV64I-NEXT:   $x10 = COPY [[C]](s64)
+  ; RV64I-NEXT:   PseudoRET implicit $x10
+entry:
+  ret i64 4294967296
+}
+
+define ptr @test_ret_ptr() {
+  ; RV32I-LABEL: name: test_ret_ptr
+  ; RV32I: bb.1.entry:
+  ; RV32I-NEXT:   [[DEF:%[0-9]+]]:_(p0) = G_IMPLICIT_DEF
+  ; RV32I-NEXT:   $x10 = COPY [[DEF]](p0)
+  ; RV32I-NEXT:   PseudoRET implicit $x10
+  ; RV64I-LABEL: name: test_ret_ptr
+  ; RV64I: bb.1.entry:
+  ; RV64I-NEXT:   [[DEF:%[0-9]+]]:_(p0) = G_IMPLICIT_DEF
+  ; RV64I-NEXT:   $x10 = COPY [[DEF]](p0)
+  ; RV64I-NEXT:   PseudoRET implicit $x10
+entry:
+  ret ptr undef
+}
+
+define [2 x i32] @test_ret_2xi32() {
+  ; RV32I-LABEL: name: test_ret_2xi32
+  ; RV32I: bb.1.entry:
+  ; RV32I-NEXT:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+  ; RV32I-NEXT:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
+  ; RV32I-NEXT:   $x10 = COPY [[C]](s32)
+  ; RV32I-NEXT:   $x11 = COPY [[C1]](s32)
+  ; RV32I-NEXT:   PseudoRET implicit $x10, implicit $x11
+  ; RV64I-LABEL: name: test_ret_2xi32
+  ; RV64I: bb.1.entry:
+  ; RV64I-NEXT:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+  ; RV64I-NEXT:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
+  ; RV64I-NEXT:   [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[C]](s32)
+  ; RV64I-NEXT:   [[ANYEXT1:%[0-9]+]]:_(s64) = G_ANYEXT [[C1]](s32)
+  ; RV64I-NEXT:   $x10 = COPY [[ANYEXT]](s64)
+  ; RV64I-NEXT:   $x11 = COPY [[ANYEXT1]](s64)
+  ; RV64I-NEXT:   PseudoRET implicit $x10, implicit $x11
+entry:
+  ret [2 x i32] [i32 1, i32 2]
+}