Files
revng-revng/lib/Lift/InstructionTranslator.cpp
T
2025-10-31 14:49:05 +01:00

1589 lines
52 KiB
C++

/// \file InstructionTranslator.cpp
/// \brief This file implements the logic to translate a libtcg instruction in
/// to LLVM IR.
//
// This file is distributed under the MIT License. See LICENSE.md for details.
//
#include <cstdint>
#include <fstream>
#include <queue>
#include <set>
#include <sstream>
#include "llvm/ADT/STLExtras.h"
#include "llvm/IR/Attributes.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/CFG.h"
#include "llvm/IR/GlobalVariable.h"
#include "llvm/IR/InstrTypes.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/Module.h"
#include "llvm/Support/Casting.h"
#include "revng/Lift/Lift.h"
#include "revng/Lift/VariableManager.h"
#include "revng/Model/FunctionTags.h"
#include "revng/Support/Assert.h"
#include "revng/Support/IRHelpers.h"
#include "revng/Support/RandomAccessIterator.h"
#include "revng/Support/Range.h"
#include "InstructionTranslator.h"
// This name is not present after `remove-newpc-calls`.
RegisterIRHelper NewPCHelper("newpc");
using namespace llvm;
static Logger Log("instruction-translator");
using IT = InstructionTranslator;
static uint64_t pc(LibTcgInstruction *Instr) {
revng_assert(Instr->opcode == LIBTCG_op_insn_start);
uint64_t PC = Instr->constant_args[0].constant;
if (Instr->nb_cargs > 1)
PC |= Instr->constant_args[1].constant << 32;
return PC;
}
/// Converts a libtcg condition into an LLVM predicate
///
/// \param Condition the input libtcg condition.
///
/// \return the corresponding LLVM predicate.
static CmpInst::Predicate conditionToPredicate(LibTcgCond Condition) {
switch (Condition) {
case LIBTCG_COND_EQ:
return CmpInst::ICMP_EQ;
case LIBTCG_COND_NE:
return CmpInst::ICMP_NE;
case LIBTCG_COND_LT:
return CmpInst::ICMP_SLT;
case LIBTCG_COND_GE:
return CmpInst::ICMP_SGE;
case LIBTCG_COND_LE:
return CmpInst::ICMP_SLE;
case LIBTCG_COND_GT:
return CmpInst::ICMP_SGT;
case LIBTCG_COND_LTU:
return CmpInst::ICMP_ULT;
case LIBTCG_COND_GEU:
return CmpInst::ICMP_UGE;
case LIBTCG_COND_LEU:
return CmpInst::ICMP_ULE;
case LIBTCG_COND_GTU:
return CmpInst::ICMP_UGT;
default:
revng_abort("Unknown libtcg condition");
}
}
/// Obtains the LLVM binary operation corresponding to the specified libtcg
/// opcode.
///
/// \param Opcode the libtcg opcode.
///
/// \return the LLVM binary operation matching opcode.
static Instruction::BinaryOps opcodeToBinaryOp(LibTcgOpcode Opcode) {
switch (Opcode) {
case LIBTCG_op_add_i32:
case LIBTCG_op_add_i64:
case LIBTCG_op_add2_i32:
case LIBTCG_op_add2_i64:
return Instruction::Add;
case LIBTCG_op_sub_i32:
case LIBTCG_op_sub_i64:
case LIBTCG_op_sub2_i32:
case LIBTCG_op_sub2_i64:
return Instruction::Sub;
case LIBTCG_op_mul_i32:
case LIBTCG_op_mul_i64:
return Instruction::Mul;
case LIBTCG_op_div_i32:
case LIBTCG_op_div_i64:
return Instruction::SDiv;
case LIBTCG_op_divu_i32:
case LIBTCG_op_divu_i64:
return Instruction::UDiv;
case LIBTCG_op_rem_i32:
case LIBTCG_op_rem_i64:
return Instruction::SRem;
case LIBTCG_op_remu_i32:
case LIBTCG_op_remu_i64:
return Instruction::URem;
case LIBTCG_op_and_i32:
case LIBTCG_op_and_i64:
return Instruction::And;
case LIBTCG_op_or_i32:
case LIBTCG_op_or_i64:
return Instruction::Or;
case LIBTCG_op_xor_i32:
case LIBTCG_op_xor_i64:
return Instruction::Xor;
case LIBTCG_op_shl_i32:
case LIBTCG_op_shl_i64:
return Instruction::Shl;
case LIBTCG_op_shr_i32:
case LIBTCG_op_shr_i64:
return Instruction::LShr;
case LIBTCG_op_sar_i32:
case LIBTCG_op_sar_i64:
return Instruction::AShr;
default:
revng_unreachable("libtcg opcode is not a binary operator");
}
}
/// Returns the maximum value which can be represented with the specified number
/// of bits.
static uint64_t getMaxValue(unsigned Bits) {
if (Bits == 32)
return 0xffffffff;
else if (Bits == 64)
return 0xffffffffffffffff;
else
revng_unreachable("Not the number of bits in an integer type");
}
/// Maps an opcode the corresponding input and output register size.
///
/// \return the size, in bits, of the registers used by the opcode.
static unsigned getRegisterSize(LibTcg &LibTcg, LibTcgOpcode Opcode) {
switch (Opcode) {
case LIBTCG_op_add2_i32:
case LIBTCG_op_add_i32:
case LIBTCG_op_andc_i32:
case LIBTCG_op_and_i32:
case LIBTCG_op_brcond2_i32:
case LIBTCG_op_brcond_i32:
case LIBTCG_op_bswap16_i32:
case LIBTCG_op_bswap32_i32:
case LIBTCG_op_deposit_i32:
case LIBTCG_op_div2_i32:
case LIBTCG_op_div_i32:
case LIBTCG_op_divu2_i32:
case LIBTCG_op_divu_i32:
case LIBTCG_op_eqv_i32:
case LIBTCG_op_ext16s_i32:
case LIBTCG_op_ext16u_i32:
case LIBTCG_op_ext8s_i32:
case LIBTCG_op_ext8u_i32:
case LIBTCG_op_extrl_i64_i32:
case LIBTCG_op_extrh_i64_i32:
case LIBTCG_op_ld16s_i32:
case LIBTCG_op_ld16u_i32:
case LIBTCG_op_ld8s_i32:
case LIBTCG_op_ld8u_i32:
case LIBTCG_op_ld_i32:
case LIBTCG_op_movcond_i32:
case LIBTCG_op_mov_i32:
case LIBTCG_op_mul_i32:
case LIBTCG_op_muls2_i32:
case LIBTCG_op_mulsh_i32:
case LIBTCG_op_mulu2_i32:
case LIBTCG_op_muluh_i32:
case LIBTCG_op_nand_i32:
case LIBTCG_op_neg_i32:
case LIBTCG_op_nor_i32:
case LIBTCG_op_not_i32:
case LIBTCG_op_orc_i32:
case LIBTCG_op_or_i32:
case LIBTCG_op_qemu_ld_a32_i32:
case LIBTCG_op_qemu_ld_a64_i32:
case LIBTCG_op_qemu_st_a32_i32:
case LIBTCG_op_qemu_st_a64_i32:
case LIBTCG_op_rem_i32:
case LIBTCG_op_remu_i32:
case LIBTCG_op_rotl_i32:
case LIBTCG_op_rotr_i32:
case LIBTCG_op_sar_i32:
case LIBTCG_op_setcond2_i32:
case LIBTCG_op_setcond_i32:
case LIBTCG_op_negsetcond_i32:
case LIBTCG_op_shl_i32:
case LIBTCG_op_shr_i32:
case LIBTCG_op_st16_i32:
case LIBTCG_op_st8_i32:
case LIBTCG_op_st_i32:
case LIBTCG_op_sub2_i32:
case LIBTCG_op_sub_i32:
case LIBTCG_op_xor_i32:
case LIBTCG_op_extract_i32:
case LIBTCG_op_sextract_i32:
case LIBTCG_op_extract2_i32:
case LIBTCG_op_clz_i32:
case LIBTCG_op_ctz_i32:
return 32;
case LIBTCG_op_add2_i64:
case LIBTCG_op_add_i64:
case LIBTCG_op_andc_i64:
case LIBTCG_op_and_i64:
case LIBTCG_op_brcond_i64:
case LIBTCG_op_bswap16_i64:
case LIBTCG_op_bswap32_i64:
case LIBTCG_op_bswap64_i64:
case LIBTCG_op_deposit_i64:
case LIBTCG_op_div2_i64:
case LIBTCG_op_div_i64:
case LIBTCG_op_divu2_i64:
case LIBTCG_op_divu_i64:
case LIBTCG_op_eqv_i64:
case LIBTCG_op_ext16s_i64:
case LIBTCG_op_ext16u_i64:
case LIBTCG_op_ext_i32_i64:
case LIBTCG_op_extu_i32_i64:
case LIBTCG_op_ext32s_i64:
case LIBTCG_op_ext32u_i64:
case LIBTCG_op_ext8s_i64:
case LIBTCG_op_ext8u_i64:
case LIBTCG_op_ld16s_i64:
case LIBTCG_op_ld16u_i64:
case LIBTCG_op_ld32s_i64:
case LIBTCG_op_ld32u_i64:
case LIBTCG_op_ld8s_i64:
case LIBTCG_op_ld8u_i64:
case LIBTCG_op_ld_i64:
case LIBTCG_op_movcond_i64:
case LIBTCG_op_mov_i64:
case LIBTCG_op_mul_i64:
case LIBTCG_op_muls2_i64:
case LIBTCG_op_mulsh_i64:
case LIBTCG_op_mulu2_i64:
case LIBTCG_op_muluh_i64:
case LIBTCG_op_nand_i64:
case LIBTCG_op_neg_i64:
case LIBTCG_op_nor_i64:
case LIBTCG_op_not_i64:
case LIBTCG_op_orc_i64:
case LIBTCG_op_or_i64:
case LIBTCG_op_qemu_ld_a32_i64:
case LIBTCG_op_qemu_ld_a64_i64:
case LIBTCG_op_qemu_st_a32_i64:
case LIBTCG_op_qemu_st_a64_i64:
case LIBTCG_op_rem_i64:
case LIBTCG_op_remu_i64:
case LIBTCG_op_rotl_i64:
case LIBTCG_op_rotr_i64:
case LIBTCG_op_sar_i64:
case LIBTCG_op_setcond_i64:
case LIBTCG_op_negsetcond_i64:
case LIBTCG_op_shl_i64:
case LIBTCG_op_shr_i64:
case LIBTCG_op_st16_i64:
case LIBTCG_op_st32_i64:
case LIBTCG_op_st8_i64:
case LIBTCG_op_st_i64:
case LIBTCG_op_sub2_i64:
case LIBTCG_op_sub_i64:
case LIBTCG_op_xor_i64:
case LIBTCG_op_clz_i64:
case LIBTCG_op_ctz_i64:
case LIBTCG_op_extract2_i64:
case LIBTCG_op_extract_i64:
case LIBTCG_op_sextract_i64:
return 64;
case LIBTCG_op_br:
case LIBTCG_op_call:
case LIBTCG_op_insn_start:
case LIBTCG_op_discard:
case LIBTCG_op_exit_tb:
case LIBTCG_op_goto_tb:
case LIBTCG_op_goto_ptr:
case LIBTCG_op_set_label:
return 0;
default: {
// For debugging purposes printing the actual opcode
// really helps.
std::stringstream ErrSS;
ErrSS << "Unexpected libtcg opcode [" << Opcode
<< "]: " << LibTcg.instructionName(Opcode);
revng_unreachable(ErrSS.str().c_str());
}
}
}
static Value *genDeposit(revng::IRBuilder &Builder,
unsigned RegisterSize,
Value *Into,
Value *From,
Value *Offset,
Value *Length) {
revng_assert(isa<ConstantInt>(Offset));
uint64_t ConstOffset = cast<ConstantInt>(Offset)->getLimitedValue();
if (ConstOffset == RegisterSize)
return Into;
revng_assert(isa<ConstantInt>(Length));
uint64_t ConstLength = cast<ConstantInt>(Length)->getLimitedValue();
uint64_t Bits = 0;
// Thou shall not << 32
if (ConstLength == RegisterSize)
Bits = getMaxValue(RegisterSize);
else
Bits = (1UL << ConstLength) - 1;
// result = (t1 & ~(bits << position)) | ((t2 & bits) << position)
uint64_t BaseMask = ~(Bits << ConstOffset);
Value *MaskedBase = Builder.CreateAnd(Into, BaseMask);
Value *Deposit = Builder.CreateAnd(From, Bits);
Value *ShiftedDeposit = Builder.CreateShl(Deposit, ConstOffset);
Value *Result = Builder.CreateOr(MaskedBase, ShiftedDeposit);
return Result;
}
/// Create a compare instruction given a comparison operator and the operands
///
/// \param Builder the builder to use to create the instruction.
/// \param Condition the libtcg condition.
/// \param FirstOperand the first operand of the comparison.
/// \param SecondOperand the second operand of the comparison.
///
/// \return a compare instruction.
template<typename T>
static Value *createICmp(T &Builder,
LibTcgCond Condition,
Value *FirstOperand,
Value *SecondOperand) {
return Builder.CreateICmp(conditionToPredicate(Condition),
FirstOperand,
SecondOperand);
}
using LBM = IT::LabeledBlocksMap;
IT::InstructionTranslator(class LibTcg &LibTcg,
revng::IRBuilder &Builder,
VariableManager &Variables,
JumpTargetManager &JumpTargets,
std::vector<BasicBlock *> Blocks,
bool EndianessMismatch,
ProgramCounterHandler *PCH) :
LibTcg(LibTcg),
Builder(Builder),
Variables(Variables),
JumpTargets(JumpTargets),
Blocks(Blocks),
TheModule(*Builder.GetInsertBlock()->getParent()->getParent()),
TheFunction(Builder.GetInsertBlock()->getParent()),
EndianessMismatch(EndianessMismatch),
NewPCMarker(nullptr),
LastPC(MetaAddress::invalid()),
PCH(PCH) {
auto &Context = TheModule.getContext();
using FT = FunctionType;
// The newpc function call takes the following parameters:
//
// * BasicBlockID of the instruction in string form
// * instruction size
// * isJT (-1: unknown, 0: no, 1: yes)
// * inlining index
// * pointer to the disassembled instruction
// * all the local variables used by this instruction
auto *NewPCMarkerTy = FT::get(Type::getVoidTy(Context),
{ Type::getInt8PtrTy(Context),
Type::getInt64Ty(Context),
Type::getInt32Ty(Context),
Type::getInt32Ty(Context),
Type::getInt8PtrTy(Context) },
true);
NewPCMarker = createIRHelper("newpc",
TheModule,
NewPCMarkerTy,
GlobalValue::ExternalLinkage);
FunctionTags::Marker.addTo(NewPCMarker);
NewPCMarker->addFnAttr(Attribute::WillReturn);
NewPCMarker->addFnAttr(Attribute::NoUnwind);
NewPCMarker->addFnAttr(Attribute::NoMerge);
}
void IT::finalizeNewPCMarkers() {
size_t FixedArgCount = NewPCMarker->arg_size();
llvm::SmallVector<CallInst *, 4> CallsToRemove;
for (User *U : NewPCMarker->users()) {
auto *Call = cast<CallInst>(U);
// Report the instruction on the coverage CSV
using namespace NewPCArguments;
MetaAddress PC = addressFromNewPC(Call);
uint64_t Size = getLimitedValue(Call->getArgOperand(InstructionSize));
bool IsJT = JumpTargets.isJumpTarget(PC);
// We already finished discovering new code to translate, so we can remove
// the references to local variables as argument of the calls to newpc and
// create room for more optimizations.
if (Call->arg_size() != FixedArgCount) {
SmallVector<Value *, 8> Args;
auto *AI = Call->arg_begin();
for (size_t Idx = 0; Idx < FixedArgCount; ++Idx, ++AI)
Args.emplace_back(*AI);
auto *NewCall = CallInst::Create(NewPCMarker, Args, "", Call);
NewCall->setCallingConv(Call->getCallingConv());
NewCall->setDebugLoc(Call->getDebugLoc());
NewCall->copyMetadata(*Call);
// Note: we intentionally do not copy attributes. We do not expect to have
// any and removing those on extra arguments leads to a mysterious
// failure in verify "Attribute after last parameter".
revng_assert(Call->use_empty());
CallsToRemove.push_back(Call);
}
}
for (auto *Call : CallsToRemove)
eraseFromParent(Call);
}
SmallSet<unsigned, 1> IT::preprocess(const LibTcgTranslationBlock &TB) {
SmallSet<unsigned, 1> Result;
for (unsigned I = 0; I < TB.instruction_count; ++I) {
LibTcgInstruction &Instruction = TB.list[I];
switch (Instruction.opcode) {
case LIBTCG_op_mov_i32:
case LIBTCG_op_mov_i64:
break;
default:
continue;
}
LibTcgArgument Argument = Instruction.output_args[0];
revng_assert(Argument.kind == LIBTCG_ARG_TEMP);
LibTcgTemp *Temp = Argument.temp;
if (Temp->kind != LIBTCG_TEMP_GLOBAL)
continue;
if (strcmp("btarget", Temp->name) != 0)
continue;
for (unsigned J = I + 1; J < TB.instruction_count; ++J) {
LibTcgOpcode Opcode = TB.list[J].opcode;
if (Opcode == LIBTCG_op_insn_start)
Result.insert(J);
}
break;
}
return Result;
}
CallInst *IT::emitNewPCCall(revng::IRBuilder &Builder,
MetaAddress PC,
uint64_t Size) const {
PointerType *Int8PtrTy = getStringPtrType(TheModule.getContext());
auto *Int8NullPtr = ConstantPointerNull::get(Int8PtrTy);
std::vector<Value *> Args = { BasicBlockID(PC).toValue(&TheModule),
Builder.getInt64(Size),
Builder.getInt32(-1),
Builder.getInt32(0),
Int8NullPtr };
// Insert a call to NewPCMarker capturing all the currently live temporaries
// which might be alive across an instruction boundary. This prevents SROA
// from transforming them in SSA values, which is bad in case we have to
// split a basic block
for (AllocaInst *V : Variables.getLiveVariables())
Args.push_back(V);
return Builder.CreateCall(NewPCMarker, Args);
}
std::tuple<IT::TranslationResult, MetaAddress, MetaAddress>
IT::newInstruction(LibTcgInstruction *Instr,
LibTcgInstruction *Next,
MetaAddress StartPC,
MetaAddress EndPC,
bool IsFirst) {
using R = std::tuple<TranslationResult, MetaAddress, MetaAddress>;
revng_assert(Instr != nullptr);
LLVMContext &Context = TheModule.getContext();
// A new original instruction, let's create a new metadata node
// referencing it for all the next instructions to come
MetaAddress PC = StartPC.replaceAddress(pc(Instr));
// Prevent translation of non-executable code
if (not JumpTargets.isExecutableAddress(PC))
return R{ Abort, MetaAddress::invalid(), MetaAddress::invalid() };
// Compute NextPC
MetaAddress NextPC = MetaAddress::invalid();
if (Next != nullptr)
NextPC = StartPC.replaceAddress(pc(Next));
else
NextPC = EndPC;
if (!IsFirst) {
// Check if this PC already has a block and use it
bool ShouldContinue;
BasicBlock *DivergeTo = JumpTargets.newPC(PC, ShouldContinue);
if (DivergeTo != nullptr) {
Builder.CreateBr(DivergeTo);
if (ShouldContinue) {
// The block is empty, let's fill it
Blocks.push_back(DivergeTo);
Builder.SetInsertPoint(DivergeTo);
} else {
// The block contains already translated code, early exit
return R{ Stop, PC, NextPC };
}
}
}
// Variables.newBasicBlock();
revng_assert(NextPC - PC);
auto *Call = emitNewPCCall(Builder, PC, *(NextPC - PC));
if (!IsFirst) {
// Inform the JumpTargetManager about the new PC we met
BasicBlock::iterator CurrentIt = Builder.GetInsertPoint();
if (CurrentIt == Builder.GetInsertBlock()->begin())
revng_assert(JumpTargets.getBlockAt(PC) == Builder.GetInsertBlock());
else
JumpTargets.registerInstruction(PC, Call);
}
return R{ Success, PC, NextPC };
}
IT::TranslationResult IT::translateCall(LibTcgInstruction *Instruction,
MetaAddress PC,
unsigned SinceInstructionStart) {
std::vector<Value *> InArgs;
for (uint8_t I = 0; I < Instruction->nb_iargs; ++I) {
auto *Load = Variables.load(Builder, &Instruction->input_args[I]);
if (Load == nullptr)
return Abort;
InArgs.push_back(Load);
}
const auto GetValueType = [](Value *Argument) { return Argument->getType(); };
auto ValueTypes = llvm::map_range(InArgs, GetValueType);
std::vector<Type *> InArgsType(ValueTypes.begin(), ValueTypes.end());
LibTcgHelperInfo Info = LibTcg.helperInfo(Instruction);
std::string HelperName = "helper_" + std::string(Info.func_name);
Function *Helper = TheModule.getFunction(HelperName);
revng_assert(Helper != nullptr);
FunctionTags::Helper.addTo(Helper);
// Emit a call to the helper
FunctionType *HelperType = Helper->getFunctionType();
for (unsigned I = 0; I < InArgs.size(); ++I) {
Type *FormalArgumentType = HelperType->getFunctionParamType(I);
InArgs[I] = Builder.CreateBitOrPointerCast(InArgs[I], FormalArgumentType);
}
CallInst *Result = Builder.CreateCall(Helper, InArgs);
// Handle return values and perform sanity checks
Type *ReturnType = HelperType->getReturnType();
switch (Instruction->nb_oargs) {
case 0:
revng_assert(ReturnType->isVoidTy());
break;
case 1: {
revng_assert(ReturnType->isIntegerTy() or ReturnType->isPointerTy());
Value *ResultDestination = Variables
.getOrCreate(&Instruction->output_args[0]);
revng_assert(ResultDestination != nullptr);
Builder.CreateStore(Result, ResultDestination);
} break;
default: {
auto *ReturnStruct = cast<StructType>(ReturnType);
revng_assert(ReturnStruct->getNumElements() == Instruction->nb_oargs);
for (unsigned I = 0; I < Instruction->nb_oargs; ++I) {
Value *ResultDestination = Variables
.getOrCreate(&Instruction->output_args[I]);
revng_assert(ResultDestination != nullptr);
Builder.CreateStore(Builder.CreateExtractValue(Result, I),
ResultDestination);
}
} break;
}
if (Info.func_flags & LIBTCG_CALL_NO_RETURN) {
handleExitTB();
}
return Success;
}
IT::TranslationResult IT::translate(LibTcgInstruction *Instr,
MetaAddress PC,
unsigned SinceInstructionStart,
MetaAddress NextPC) {
std::vector<Value *> InArgs;
for (unsigned I = 0; I < Instr->nb_iargs; ++I) {
auto *Load = Variables.load(Builder, &Instr->input_args[I]);
if (Load == nullptr) {
revng_log(Log,
"Aborting translation of instruction #"
<< SinceInstructionStart << " in " << PC
<< " due to input argument " << I << ".");
return Abort;
}
InArgs.push_back(Load);
}
// TODO: constant args are not widely used. Consider accessing constant_args
// directly in translateOpcode where needed.
std::vector<LibTcgArgument> ConstArgs;
{
unsigned RegisterSize = getRegisterSize(LibTcg, Instr->opcode);
Type *RegisterType = nullptr;
if (RegisterSize == 32)
RegisterType = Builder.getInt32Ty();
else if (RegisterSize == 64 or RegisterSize == 0)
RegisterType = Builder.getInt64Ty();
else if (RegisterSize != 0)
revng_unreachable("Unexpected register size");
for (unsigned I = 0; I < Instr->nb_cargs; ++I) {
if (Instr->constant_args[I].kind == LIBTCG_ARG_CONSTANT) {
InArgs.push_back(ConstantInt::get(RegisterType,
Instr->constant_args[I].constant));
} else {
ConstArgs.push_back(Instr->constant_args[I]);
}
}
}
LastPC = PC;
auto Result = translateOpcode(Instr->opcode, ConstArgs, InArgs);
revng_assert(Result.size() == Instr->nb_oargs);
// TODO: use ZipIterator here
for (unsigned I = 0; I < Result.size(); I++) {
auto *Destination = Variables.getOrCreate(&Instr->output_args[I]);
if (Destination == nullptr) {
revng_log(Log,
"Aborting translation of instruction #"
<< SinceInstructionStart << " in " << PC
<< " due to output argument " << I << ".");
return Abort;
}
auto *Store = Builder.CreateStore(Result[I], Destination);
if (PCH->affectsPC(Store)) {
// This is a PC-related store
PCH->handleStore(Builder, Store);
}
Value *StoredValue = Store->getValueOperand();
SmallVector<ConstantInt *> Constants;
if (auto *Constant = dyn_cast<ConstantInt>(StoredValue)) {
Constants.push_back(Constant);
} else if (auto *Select = dyn_cast<SelectInst>(StoredValue)) {
if (auto *Constant = dyn_cast<ConstantInt>(Select->getTrueValue()))
Constants.push_back(Constant);
if (auto *Constant = dyn_cast<ConstantInt>(Select->getFalseValue()))
Constants.push_back(Constant);
}
for (auto *Constant : Constants) {
MetaAddress Address = JumpTargets.fromPC(Constant->getLimitedValue());
if (Address.isValid() and PC != Address and JumpTargets.isPC(Address)
and not JumpTargets.hasJT(Address)) {
JumpTargets.registerSimpleLiteral(Address);
}
}
}
return Success;
}
void IT::registerDirectJumps() {
for (BasicBlock *ExitBB : ExitBlocks) {
auto &&[Result, NextPC] = PCH->getUniqueJumpTarget(ExitBB);
if (Result == NextJumpTarget::Unique and JumpTargets.isPC(NextPC)
and not JumpTargets.hasJT(NextPC)) {
JumpTargets.registerJT(NextPC, JTReason::DirectJump);
}
}
ExitBlocks.clear();
}
int64_t IT::getEnvOffset(Instruction &I, int64_t Offset) const {
Value *Pointer = nullptr;
if (auto *Load = dyn_cast<LoadInst>(&I))
Pointer = Load->getPointerOperand();
else if (auto *Store = dyn_cast<StoreInst>(&I))
Pointer = Store->getPointerOperand();
// Check if we're loading from env directly
if (Variables.isEnv(Pointer))
return Offset;
// Handle simple alloc
auto *Alloca = dyn_cast<AllocaInst>(Pointer);
revng_assert(Alloca != nullptr);
// Look for the last store there
bool AddendFound = false;
BasicBlock *Current = Builder.GetInsertBlock();
for (Instruction &I : llvm::make_range(Current->rbegin(), Current->rend())) {
if (auto *Store = dyn_cast<StoreInst>(&I)) {
Value *StorePointer = Store->getPointerOperand();
// Only accept store to allocas
revng_check(isa<AllocaInst>(StorePointer)
or isa<GlobalVariable>(StorePointer));
// Check if we found a store targeting our alloca
if (StorePointer == Alloca) {
// Extract base and addend
auto *Add = cast<BinaryOperator>(Store->getValueOperand());
revng_check(Add->getOpcode() == llvm::Instruction::Add);
revng_check(isa<ConstantInt>(Add->getOperand(1)));
Pointer = Add->getOperand(0);
revng_assert(Variables.isEnv(Pointer));
Offset += cast<ConstantInt>(Add->getOperand(1))->getLimitedValue();
return Offset;
}
} else if (isa<CallBase>(&I)) {
// Abort in case we find a call
revng_abort();
} else {
// Skip over instructions without side effects
}
}
revng_abort();
}
std::vector<Value *>
IT::translateOpcode(LibTcgOpcode Opcode,
std::vector<LibTcgArgument> ConstArguments,
std::vector<Value *> InArguments) {
LLVMContext &Context = TheModule.getContext();
unsigned RegisterSize = getRegisterSize(LibTcg, Opcode);
Type *RegisterType = nullptr;
if (RegisterSize == 32)
RegisterType = Builder.getInt32Ty();
else if (RegisterSize == 64)
RegisterType = Builder.getInt64Ty();
else if (RegisterSize != 0)
revng_unreachable("Unexpected register size");
switch (Opcode) {
case LIBTCG_op_discard:
// Let's overwrite the discarded temporary with a 0
return { ConstantInt::get(RegisterType, 0) };
case LIBTCG_op_mov_i32:
case LIBTCG_op_mov_i64:
if (auto *Constant = dyn_cast<ConstantInt>(InArguments[0])) {
return { Constant };
} else {
return { Builder.CreateTrunc(InArguments[0], RegisterType) };
}
case LIBTCG_op_setcond_i32:
case LIBTCG_op_setcond_i64: {
revng_assert(ConstArguments.size() > 0
and ConstArguments[0].kind == LIBTCG_ARG_COND);
Value *Compare = createICmp(Builder,
ConstArguments[0].cond,
InArguments[0],
InArguments[1]);
return { Builder.CreateZExt(Compare, RegisterType) };
}
case LIBTCG_op_negsetcond_i32:
case LIBTCG_op_negsetcond_i64: {
revng_assert(ConstArguments.size() > 0
and ConstArguments[0].kind == LIBTCG_ARG_COND);
Value *Compare = createICmp(Builder,
ConstArguments[0].cond,
InArguments[0],
InArguments[1]);
auto *Zero = ConstantInt::get(RegisterType, 0);
Value *Result = Builder.CreateZExt(Compare, RegisterType);
return { Builder.CreateSub(Zero, Result) };
}
case LIBTCG_op_movcond_i32: // Resist the fallthrough temptation
case LIBTCG_op_movcond_i64: {
revng_assert(ConstArguments[0].kind == LIBTCG_ARG_COND);
Value *Compare = createICmp(Builder,
ConstArguments[0].cond,
InArguments[0],
InArguments[1]);
Value *Select = Builder.CreateSelect(Compare,
InArguments[2],
InArguments[3]);
return { Select };
}
case LIBTCG_op_qemu_ld_a32_i32:
case LIBTCG_op_qemu_ld_a64_i32:
case LIBTCG_op_qemu_ld_a32_i64:
case LIBTCG_op_qemu_ld_a64_i64:
case LIBTCG_op_qemu_st_a32_i32:
case LIBTCG_op_qemu_st_a64_i32:
case LIBTCG_op_qemu_st_a32_i64:
case LIBTCG_op_qemu_st_a64_i64: {
revng_assert(ConstArguments[0].kind == LIBTCG_ARG_MEM_OP_INDEX);
LibTcgMemOp MemoryOp = ConstArguments[0].mem_op_index.op;
unsigned Alignment = 1;
// Load size
IntegerType *MemoryType = nullptr;
auto MemoryOpSize = static_cast<LibTcgMemOp>(MemoryOp & LIBTCG_MO_SIZE);
switch (MemoryOpSize) {
case LIBTCG_MO_8:
MemoryType = Builder.getInt8Ty();
break;
case LIBTCG_MO_16:
MemoryType = Builder.getInt16Ty();
break;
case LIBTCG_MO_32:
MemoryType = Builder.getInt32Ty();
break;
case LIBTCG_MO_64:
MemoryType = Builder.getInt64Ty();
break;
default:
revng_unreachable("Unexpected load size");
}
// If necessary, handle endianness mismatch
// TODO: it might be a bit overkill, but it be nice to make this function
// template-parametric w.r.t. endianness mismatch
Function *BSwapFunction = nullptr;
if (MemoryType != Builder.getInt8Ty() and EndianessMismatch)
BSwapFunction = Intrinsic::getDeclaration(&TheModule,
Intrinsic::bswap,
{ MemoryType });
// Is the memory op a sign extended load?
bool SignExtend = (MemoryOp & LIBTCG_MO_SIGN) != 0;
Value *Pointer = nullptr;
if (Opcode == LIBTCG_op_qemu_ld_a32_i32
or Opcode == LIBTCG_op_qemu_ld_a64_i32
or Opcode == LIBTCG_op_qemu_ld_a32_i64
or Opcode == LIBTCG_op_qemu_ld_a64_i64) {
Pointer = Builder.CreateIntToPtr(InArguments[0],
MemoryType->getPointerTo());
auto *Load = Builder.CreateAlignedLoad(MemoryType,
Pointer,
MaybeAlign(Alignment));
Value *Loaded = Load;
if (BSwapFunction != nullptr)
Loaded = Builder.CreateCall(BSwapFunction, Load);
if (SignExtend)
return { Builder.CreateSExt(Loaded, RegisterType) };
else
return { Builder.CreateZExt(Loaded, RegisterType) };
} else if (Opcode == LIBTCG_op_qemu_st_a32_i32
or Opcode == LIBTCG_op_qemu_st_a64_i32
or Opcode == LIBTCG_op_qemu_st_a32_i64
or Opcode == LIBTCG_op_qemu_st_a64_i64) {
Pointer = Builder.CreateIntToPtr(InArguments[1],
MemoryType->getPointerTo());
Value *Value = Builder.CreateTrunc(InArguments[0], MemoryType);
if (BSwapFunction != nullptr)
Value = Builder.CreateCall(BSwapFunction, Value);
auto *Store = Builder.CreateAlignedStore(Value,
Pointer,
MaybeAlign(Alignment));
// If we're writing somewhere an immediate, register it for exploration
if (auto *Constant = dyn_cast<ConstantInt>(Store->getValueOperand())) {
MetaAddress Address = JumpTargets.fromPC(Constant->getLimitedValue());
if (Address.isValid() and JumpTargets.isPC(Address)
and not JumpTargets.hasJT(Address)) {
JumpTargets.registerSimpleLiteral(Address);
}
}
return {};
} else {
revng_unreachable("Unknown load type");
}
}
case LIBTCG_op_ld8u_i32:
case LIBTCG_op_ld8s_i32:
case LIBTCG_op_ld16u_i32:
case LIBTCG_op_ld16s_i32:
case LIBTCG_op_ld_i32:
case LIBTCG_op_ld8u_i64:
case LIBTCG_op_ld8s_i64:
case LIBTCG_op_ld16u_i64:
case LIBTCG_op_ld16s_i64:
case LIBTCG_op_ld32u_i64:
case LIBTCG_op_ld32s_i64:
case LIBTCG_op_ld_i64: {
bool Signed = false;
switch (Opcode) {
case LIBTCG_op_ld_i32:
case LIBTCG_op_ld_i64:
case LIBTCG_op_ld8u_i32:
case LIBTCG_op_ld16u_i32:
case LIBTCG_op_ld8u_i64:
case LIBTCG_op_ld16u_i64:
case LIBTCG_op_ld32u_i64:
Signed = false;
break;
case LIBTCG_op_ld8s_i32:
case LIBTCG_op_ld16s_i32:
case LIBTCG_op_ld8s_i64:
case LIBTCG_op_ld16s_i64:
case LIBTCG_op_ld32s_i64:
Signed = true;
break;
default:
revng_unreachable("Unexpected opcode");
}
unsigned LoadSize;
switch (Opcode) {
case LIBTCG_op_ld8u_i32:
case LIBTCG_op_ld8s_i32:
case LIBTCG_op_ld8u_i64:
case LIBTCG_op_ld8s_i64:
LoadSize = 1;
break;
case LIBTCG_op_ld16u_i32:
case LIBTCG_op_ld16s_i32:
case LIBTCG_op_ld16u_i64:
case LIBTCG_op_ld16s_i64:
LoadSize = 2;
break;
case LIBTCG_op_ld_i32:
case LIBTCG_op_ld32u_i64:
case LIBTCG_op_ld32s_i64:
LoadSize = 4;
break;
case LIBTCG_op_ld_i64:
LoadSize = 8;
break;
default:
revng_unreachable("Unexpected opcode");
}
auto *Base = dyn_cast<LoadInst>(InArguments[0]);
int64_t Offset = cast<ConstantInt>(InArguments[1])->getLimitedValue();
Offset = getEnvOffset(*Base, Offset);
Value *Result = Variables.loadFromEnvOffset(Builder, LoadSize, Offset);
revng_assert(Result != nullptr);
// Zero/sign extend in the target dimension
if (Signed)
return { Builder.CreateSExt(Result, RegisterType) };
else
return { Builder.CreateZExt(Result, RegisterType) };
}
case LIBTCG_op_st8_i32:
case LIBTCG_op_st16_i32:
case LIBTCG_op_st_i32:
case LIBTCG_op_st8_i64:
case LIBTCG_op_st16_i64:
case LIBTCG_op_st32_i64:
case LIBTCG_op_st_i64: {
unsigned StoreSize;
switch (Opcode) {
case LIBTCG_op_st8_i32:
case LIBTCG_op_st8_i64:
StoreSize = 1;
break;
case LIBTCG_op_st16_i32:
case LIBTCG_op_st16_i64:
StoreSize = 2;
break;
case LIBTCG_op_st_i32:
case LIBTCG_op_st32_i64:
StoreSize = 4;
break;
case LIBTCG_op_st_i64:
StoreSize = 8;
break;
default:
revng_unreachable("Unexpected opcode");
}
// For host stores, right now we handle a couple of simple situations.
// TODO: the more appropriate thing to do would be to leave these as memory
// accesses relative to env, eventually run SROA and *then* promote
// them to CSV accesses.
auto *Load = cast<LoadInst>(InArguments[1]);
int64_t Offset = cast<ConstantInt>(InArguments[2])->getLimitedValue();
Offset = getEnvOffset(*Load, Offset);
revng_assert(isa<ConstantInt>(InArguments[2]));
auto Result = Variables.storeToEnvOffset(Builder,
StoreSize,
Offset,
InArguments[0]);
PCH->handleStore(Builder, *Result);
return {};
}
case LIBTCG_op_add_i32:
case LIBTCG_op_sub_i32:
case LIBTCG_op_mul_i32:
case LIBTCG_op_div_i32:
case LIBTCG_op_divu_i32:
case LIBTCG_op_rem_i32:
case LIBTCG_op_remu_i32:
case LIBTCG_op_and_i32:
case LIBTCG_op_or_i32:
case LIBTCG_op_xor_i32:
case LIBTCG_op_shl_i32:
case LIBTCG_op_shr_i32:
case LIBTCG_op_sar_i32:
case LIBTCG_op_add_i64:
case LIBTCG_op_sub_i64:
case LIBTCG_op_mul_i64:
case LIBTCG_op_div_i64:
case LIBTCG_op_divu_i64:
case LIBTCG_op_rem_i64:
case LIBTCG_op_remu_i64:
case LIBTCG_op_and_i64:
case LIBTCG_op_or_i64:
case LIBTCG_op_xor_i64:
case LIBTCG_op_shl_i64:
case LIBTCG_op_shr_i64:
case LIBTCG_op_sar_i64: {
// TODO: assert on sizes?
Instruction::BinaryOps BinaryOp = opcodeToBinaryOp(Opcode);
Value *Operation = Builder.CreateBinOp(BinaryOp,
InArguments[0],
InArguments[1]);
return { Operation };
}
case LIBTCG_op_div2_i32:
case LIBTCG_op_divu2_i32:
case LIBTCG_op_div2_i64:
case LIBTCG_op_divu2_i64: {
Instruction::BinaryOps DivisionOp, RemainderOp;
if (Opcode == LIBTCG_op_div2_i32 or Opcode == LIBTCG_op_div2_i64) {
DivisionOp = Instruction::SDiv;
RemainderOp = Instruction::SRem;
} else if (Opcode == LIBTCG_op_divu2_i32 or Opcode == LIBTCG_op_divu2_i64) {
DivisionOp = Instruction::UDiv;
RemainderOp = Instruction::URem;
} else {
revng_unreachable("Unknown operation type");
}
// TODO: we're ignoring InArguments[1], which is the MSB
// TODO: assert on sizes?
Value *Division = Builder.CreateBinOp(DivisionOp,
InArguments[0],
InArguments[2]);
Value *Remainder = Builder.CreateBinOp(RemainderOp,
InArguments[0],
InArguments[2]);
return { Division, Remainder };
}
case LIBTCG_op_rotr_i32:
case LIBTCG_op_rotr_i64:
case LIBTCG_op_rotl_i32:
case LIBTCG_op_rotl_i64: {
Value *Bits = ConstantInt::get(RegisterType, RegisterSize);
Instruction::BinaryOps FirstShiftOp, SecondShiftOp;
if (Opcode == LIBTCG_op_rotl_i32 or Opcode == LIBTCG_op_rotl_i64) {
FirstShiftOp = Instruction::Shl;
SecondShiftOp = Instruction::LShr;
} else if (Opcode == LIBTCG_op_rotr_i32 or Opcode == LIBTCG_op_rotr_i64) {
FirstShiftOp = Instruction::LShr;
SecondShiftOp = Instruction::Shl;
} else {
revng_unreachable("Unexpected opcode");
}
Value *FirstShift = Builder.CreateBinOp(FirstShiftOp,
InArguments[0],
InArguments[1]);
Value *SecondShiftAmount = Builder.CreateSub(Bits, InArguments[1]);
Value *SecondShift = Builder.CreateBinOp(SecondShiftOp,
InArguments[0],
SecondShiftAmount);
return { Builder.CreateOr(FirstShift, SecondShift) };
}
case LIBTCG_op_deposit_i32:
case LIBTCG_op_deposit_i64: {
Value *Result = genDeposit(Builder,
RegisterSize,
InArguments[0],
InArguments[1],
InArguments[2],
InArguments[3]);
return { Result };
}
case LIBTCG_op_ext8s_i32:
case LIBTCG_op_ext16s_i32:
case LIBTCG_op_ext8u_i32:
case LIBTCG_op_ext16u_i32:
case LIBTCG_op_ext8s_i64:
case LIBTCG_op_ext16s_i64:
case LIBTCG_op_ext32s_i64:
case LIBTCG_op_ext8u_i64:
case LIBTCG_op_ext16u_i64:
case LIBTCG_op_ext32u_i64:
case LIBTCG_op_ext_i32_i64:
case LIBTCG_op_extu_i32_i64: {
Type *SourceType = nullptr;
switch (Opcode) {
case LIBTCG_op_ext8s_i32:
case LIBTCG_op_ext8u_i32:
case LIBTCG_op_ext8s_i64:
case LIBTCG_op_ext8u_i64:
SourceType = Builder.getInt8Ty();
break;
case LIBTCG_op_ext16s_i32:
case LIBTCG_op_ext16u_i32:
case LIBTCG_op_ext16s_i64:
case LIBTCG_op_ext16u_i64:
SourceType = Builder.getInt16Ty();
break;
case LIBTCG_op_ext32s_i64:
case LIBTCG_op_ext32u_i64:
case LIBTCG_op_ext_i32_i64:
case LIBTCG_op_extu_i32_i64:
SourceType = Builder.getInt32Ty();
break;
default:
revng_unreachable("Unexpected opcode");
}
Value *Truncated = Builder.CreateTrunc(InArguments[0], SourceType);
switch (Opcode) {
case LIBTCG_op_ext8s_i32:
case LIBTCG_op_ext8s_i64:
case LIBTCG_op_ext16s_i32:
case LIBTCG_op_ext16s_i64:
case LIBTCG_op_ext32s_i64:
case LIBTCG_op_ext_i32_i64:
return { Builder.CreateSExt(Truncated, RegisterType) };
case LIBTCG_op_ext8u_i32:
case LIBTCG_op_ext8u_i64:
case LIBTCG_op_ext16u_i32:
case LIBTCG_op_ext16u_i64:
case LIBTCG_op_ext32u_i64:
case LIBTCG_op_extu_i32_i64:
return { Builder.CreateZExt(Truncated, RegisterType) };
default:
revng_unreachable("Unexpected opcode");
}
}
case LIBTCG_op_extrl_i64_i32: {
return { Builder.CreateTrunc(InArguments[0], Builder.getInt32Ty()) };
}
case LIBTCG_op_extrh_i64_i32: {
Value *Shifted = Builder.CreateAShr(InArguments[0],
ConstantInt::get(Builder.getInt64Ty(),
32));
return { Builder.CreateTrunc(Shifted, Builder.getInt32Ty()) };
}
case LIBTCG_op_not_i32:
case LIBTCG_op_not_i64:
return { Builder.CreateXor(InArguments[0], getMaxValue(RegisterSize)) };
case LIBTCG_op_neg_i32:
case LIBTCG_op_neg_i64: {
auto *InitialValue = ConstantInt::get(RegisterType, 0);
return { Builder.CreateSub(InitialValue, InArguments[0]) };
}
case LIBTCG_op_andc_i32:
case LIBTCG_op_andc_i64:
case LIBTCG_op_orc_i32:
case LIBTCG_op_orc_i64:
case LIBTCG_op_eqv_i32:
case LIBTCG_op_eqv_i64: {
Instruction::BinaryOps ExternalOp;
switch (Opcode) {
case LIBTCG_op_andc_i32:
case LIBTCG_op_andc_i64:
ExternalOp = Instruction::And;
break;
case LIBTCG_op_orc_i32:
case LIBTCG_op_orc_i64:
ExternalOp = Instruction::Or;
break;
case LIBTCG_op_eqv_i32:
case LIBTCG_op_eqv_i64:
ExternalOp = Instruction::Xor;
break;
default:
revng_unreachable("Unexpected opcode");
}
Value *Negate = Builder.CreateXor(InArguments[1],
getMaxValue(RegisterSize));
Value *Result = Builder.CreateBinOp(ExternalOp, InArguments[0], Negate);
return { Result };
}
case LIBTCG_op_nand_i32:
case LIBTCG_op_nand_i64: {
Value *AndValue = Builder.CreateAnd(InArguments[0], InArguments[1]);
Value *Result = Builder.CreateXor(AndValue, getMaxValue(RegisterSize));
return { Result };
}
case LIBTCG_op_nor_i32:
case LIBTCG_op_nor_i64: {
Value *OrValue = Builder.CreateOr(InArguments[0], InArguments[1]);
Value *Result = Builder.CreateXor(OrValue, getMaxValue(RegisterSize));
return { Result };
}
case LIBTCG_op_bswap16_i32:
case LIBTCG_op_bswap32_i32:
case LIBTCG_op_bswap16_i64:
case LIBTCG_op_bswap32_i64:
case LIBTCG_op_bswap64_i64: {
Type *SwapType = nullptr;
switch (Opcode) {
case LIBTCG_op_bswap16_i32:
case LIBTCG_op_bswap16_i64:
SwapType = Builder.getInt16Ty();
break;
case LIBTCG_op_bswap32_i32:
case LIBTCG_op_bswap32_i64:
SwapType = Builder.getInt32Ty();
break;
case LIBTCG_op_bswap64_i64:
SwapType = Builder.getInt64Ty();
break;
default:
revng_unreachable("Unexpected opcode");
}
Value *Truncated = Builder.CreateTrunc(InArguments[0], SwapType);
Function *BSwapFunction = Intrinsic::getDeclaration(&TheModule,
Intrinsic::bswap,
{ SwapType });
Value *Swapped = Builder.CreateCall(BSwapFunction, Truncated);
return { Builder.CreateZExt(Swapped, RegisterType) };
}
case LIBTCG_op_set_label: {
revng_assert(ConstArguments[0].kind == LIBTCG_ARG_LABEL);
auto LabelId = ConstArguments[0].label->id;
std::stringstream LabelSS;
LabelSS << "bb." << JumpTargets.nameForAddress(LastPC);
LabelSS << "_L" << std::dec << LabelId;
std::string Label = LabelSS.str();
BasicBlock *Fallthrough = nullptr;
if (!LabeledBasicBlocks.contains(Label)) {
Fallthrough = BasicBlock::Create(Context, Label, TheFunction);
Fallthrough->moveAfter(Builder.GetInsertBlock());
LabeledBasicBlocks[Label] = Fallthrough;
} else {
// A basic block with that label already exist
Fallthrough = LabeledBasicBlocks[Label];
// Ensure it's empty
revng_assert(Fallthrough->begin() == Fallthrough->end());
// Move it to the bottom
Fallthrough->removeFromParent();
TheFunction->insert(TheFunction->end(), Fallthrough);
}
Builder.CreateBr(Fallthrough);
Blocks.push_back(Fallthrough);
Builder.SetInsertPoint(Fallthrough);
Variables.newExtendedBasicBlock();
return {};
}
case LIBTCG_op_br:
case LIBTCG_op_brcond_i32:
case LIBTCG_op_brcond2_i32:
case LIBTCG_op_brcond_i64: {
// We take the last constant arguments, which is the LabelId both in
// conditional and unconditional jumps
revng_assert(ConstArguments.back().kind == LIBTCG_ARG_LABEL);
auto LabelId = ConstArguments.back().label->id;
std::stringstream LabelSS;
LabelSS << "bb." << JumpTargets.nameForAddress(LastPC);
LabelSS << "_L" << std::dec << LabelId;
std::string Label = LabelSS.str();
BasicBlock *Fallthrough = BasicBlock::Create(Context,
Label + "_ft",
TheFunction);
// Look for a matching label
BasicBlock *Target = nullptr;
if (!LabeledBasicBlocks.contains(Label)) {
// No matching label, create a temporary block
Target = BasicBlock::Create(Context, Label, TheFunction);
LabeledBasicBlocks[Label] = Target;
} else {
Target = LabeledBasicBlocks[Label];
}
if (Opcode == LIBTCG_op_br) {
// Unconditional jump
Builder.CreateBr(Target);
} else if (Opcode == LIBTCG_op_brcond_i32
or Opcode == LIBTCG_op_brcond_i64) {
// Conditional jump
revng_assert(ConstArguments[0].kind == LIBTCG_ARG_COND);
Value *Compare = createICmp(Builder,
ConstArguments[0].cond,
InArguments[0],
InArguments[1]);
Builder.CreateCondBr(Compare, Target, Fallthrough);
} else {
revng_unreachable("Unhandled opcode");
}
Blocks.push_back(Fallthrough);
Builder.SetInsertPoint(Fallthrough);
if (Opcode == LIBTCG_op_br) {
Variables.newExtendedBasicBlock();
}
return {};
}
case LIBTCG_op_exit_tb: {
auto *Zero = ConstantInt::get(Type::getInt32Ty(Context), 0);
Builder.CreateCall(JumpTargets.exitTB(), { Zero });
Builder.CreateUnreachable();
ExitBlocks.push_back(Builder.GetInsertBlock());
auto *NextBB = BasicBlock::Create(Context, "", TheFunction);
Blocks.push_back(NextBB);
Builder.SetInsertPoint(NextBB);
Variables.newExtendedBasicBlock();
return {};
}
case LIBTCG_op_goto_tb:
case LIBTCG_op_goto_ptr:
// Nothing to do here
return {};
case LIBTCG_op_add2_i32:
case LIBTCG_op_sub2_i32:
case LIBTCG_op_add2_i64:
case LIBTCG_op_sub2_i64: {
Value *FirstOpLow = nullptr;
Value *FirstOpHigh = nullptr;
Value *SecondOpLow = nullptr;
Value *SecondOpHigh = nullptr;
IntegerType *DestinationType = Builder.getIntNTy(RegisterSize * 2);
FirstOpLow = Builder.CreateZExt(InArguments[0], DestinationType);
FirstOpHigh = Builder.CreateZExt(InArguments[1], DestinationType);
SecondOpLow = Builder.CreateZExt(InArguments[2], DestinationType);
SecondOpHigh = Builder.CreateZExt(InArguments[3], DestinationType);
FirstOpHigh = Builder.CreateShl(FirstOpHigh, RegisterSize);
SecondOpHigh = Builder.CreateShl(SecondOpHigh, RegisterSize);
Value *FirstOp = Builder.CreateOr(FirstOpHigh, FirstOpLow);
Value *SecondOp = Builder.CreateOr(SecondOpHigh, SecondOpLow);
Instruction::BinaryOps BinaryOp = opcodeToBinaryOp(Opcode);
Value *Result = Builder.CreateBinOp(BinaryOp, FirstOp, SecondOp);
Value *ResultLow = Builder.CreateTrunc(Result, RegisterType);
Value *ShiftedResult = Builder.CreateLShr(Result, RegisterSize);
Value *ResultHigh = Builder.CreateTrunc(ShiftedResult, RegisterType);
return { ResultLow, ResultHigh };
}
case LIBTCG_op_mulu2_i32:
case LIBTCG_op_mulu2_i64:
case LIBTCG_op_muls2_i32:
case LIBTCG_op_muls2_i64: {
IntegerType *DestinationType = Builder.getIntNTy(RegisterSize * 2);
Value *FirstOp = nullptr;
Value *SecondOp = nullptr;
if (Opcode == LIBTCG_op_mulu2_i32 or Opcode == LIBTCG_op_mulu2_i64) {
FirstOp = Builder.CreateZExt(InArguments[0], DestinationType);
SecondOp = Builder.CreateZExt(InArguments[1], DestinationType);
} else if (Opcode == LIBTCG_op_muls2_i32 or Opcode == LIBTCG_op_muls2_i64) {
FirstOp = Builder.CreateSExt(InArguments[0], DestinationType);
SecondOp = Builder.CreateSExt(InArguments[1], DestinationType);
} else {
revng_unreachable("Unexpected opcode");
}
Value *Result = Builder.CreateMul(FirstOp, SecondOp);
Value *ResultLow = Builder.CreateTrunc(Result, RegisterType);
Value *ShiftedResult = Builder.CreateLShr(Result, RegisterSize);
Value *ResultHigh = Builder.CreateTrunc(ShiftedResult, RegisterType);
return { ResultLow, ResultHigh };
}
case LIBTCG_op_muluh_i32:
case LIBTCG_op_mulsh_i32:
case LIBTCG_op_muluh_i64:
case LIBTCG_op_mulsh_i64:
case LIBTCG_op_setcond2_i32:
revng_unreachable("Instruction not implemented");
case LIBTCG_op_extract_i32: {
auto *Const32 = ConstantInt::get(Type::getInt32Ty(Context), 32);
Value *Length = InArguments[1];
Value *Offset = InArguments[2];
Value *ShlAmount = Builder.CreateSub(Const32,
Builder.CreateAdd(Offset, Length));
Value *Shl = Builder.CreateShl(InArguments[0], ShlAmount);
Value *LShr = Builder.CreateLShr(Shl, Builder.CreateSub(Const32, Length));
return { LShr };
}
case LIBTCG_op_sextract_i32: {
auto *Const32 = ConstantInt::get(Type::getInt32Ty(Context), 32);
Value *Length = InArguments[1];
Value *Offset = InArguments[2];
Value *ShlAmount = Builder.CreateSub(Const32,
Builder.CreateAdd(Offset, Length));
Value *Shl = Builder.CreateShl(InArguments[0], ShlAmount);
Value *AShr = Builder.CreateAShr(Shl, Builder.CreateSub(Const32, Length));
return { AShr };
}
case LIBTCG_op_extract_i64: {
auto *Const64 = ConstantInt::get(Type::getInt64Ty(Context), 64);
Value *Length = InArguments[1];
Value *Offset = InArguments[2];
Value *ShlAmount = Builder.CreateSub(Const64,
Builder.CreateAdd(Offset, Length));
Value *Shl = Builder.CreateShl(InArguments[0], ShlAmount);
Value *LShr = Builder.CreateLShr(Shl, Builder.CreateSub(Const64, Length));
return { LShr };
}
case LIBTCG_op_sextract_i64: {
auto *Const64 = ConstantInt::get(Type::getInt64Ty(Context), 64);
Value *Length = InArguments[1];
Value *Offset = InArguments[2];
Value *ShlAmount = Builder.CreateSub(Const64,
Builder.CreateAdd(Offset, Length));
Value *Shl = Builder.CreateShl(InArguments[0], ShlAmount);
Value *AShr = Builder.CreateAShr(Shl, Builder.CreateSub(Const64, Length));
return { AShr };
}
case LIBTCG_op_extract2_i32:
case LIBTCG_op_extract2_i64: {
Value *Low = InArguments[0];
Value *High = InArguments[1];
Value *Offset = InArguments[2];
auto *ConstSize = ConstantInt::get(RegisterType, RegisterSize);
Value *Shift = Builder.CreateLShr(Low, Offset);
Value *Result = genDeposit(Builder,
RegisterSize,
Shift,
High,
Builder.CreateSub(ConstSize, Offset),
Offset);
return { Result };
}
case LIBTCG_op_clz_i32: {
Type *Int1Ty = Type::getInt1Ty(Context);
auto *One = ConstantInt::get(Int1Ty, 1);
auto *Zero = ConstantInt::get(RegisterType, 0);
Value *Arg = InArguments[0];
Value *ZeroVal = InArguments[1];
CallInst *Ctlz = Builder.CreateBinaryIntrinsic(Intrinsic::ctlz, Arg, One);
Value *ICmp = Builder.CreateICmp(CmpInst::ICMP_EQ, Arg, Zero);
Value *Select = Builder.CreateSelect(ICmp, ZeroVal, Ctlz);
return { Select };
}
case LIBTCG_op_clz_i64: {
Type *Int1Ty = Type::getInt1Ty(Context);
auto *One = ConstantInt::get(Int1Ty, 1);
auto *Zero = ConstantInt::get(RegisterType, 0);
Value *Arg = InArguments[0];
Value *ZeroVal = InArguments[1];
CallInst *Ctlz = Builder.CreateBinaryIntrinsic(Intrinsic::ctlz, Arg, One);
Value *ICmp = Builder.CreateICmp(CmpInst::ICMP_EQ, Arg, Zero);
Value *Select = Builder.CreateSelect(ICmp, ZeroVal, Ctlz);
return { Select };
}
case LIBTCG_op_ctz_i32: {
Type *Int1Ty = Type::getInt1Ty(Context);
auto *One = ConstantInt::get(Int1Ty, 1);
auto *Zero = ConstantInt::get(RegisterType, 0);
Value *Arg = InArguments[0];
Value *ZeroVal = InArguments[1];
CallInst *Cttz = Builder.CreateBinaryIntrinsic(Intrinsic::cttz, Arg, One);
Value *ICmp = Builder.CreateICmp(CmpInst::ICMP_EQ, Arg, Zero);
Value *Select = Builder.CreateSelect(ICmp, ZeroVal, Cttz);
return { Select };
}
case LIBTCG_op_ctz_i64: {
Type *Int1Ty = Type::getInt1Ty(Context);
auto *One = ConstantInt::get(Int1Ty, 1);
auto *Zero = ConstantInt::get(RegisterType, 0);
Value *Arg = InArguments[0];
Value *ZeroVal = InArguments[1];
CallInst *Cttz = Builder.CreateBinaryIntrinsic(Intrinsic::cttz, Arg, One);
Value *ICmp = Builder.CreateICmp(CmpInst::ICMP_EQ, Arg, Zero);
Value *Select = Builder.CreateSelect(ICmp, ZeroVal, Cttz);
return { Select };
}
default:
// For debugging purposes printing the actual opcode
// really helps.
std::stringstream ErrSS;
ErrSS << "Unknown libtcg opcode [" << Opcode
<< "]: " << LibTcg.instructionName(Opcode);
revng_unreachable(ErrSS.str().c_str());
}
}
void IT::handleExitTB() {
auto &Context = TheModule.getContext();
auto *Zero = ConstantInt::get(Type::getInt32Ty(Context), 0);
Builder.CreateCall(JumpTargets.exitTB(), { Zero });
Builder.CreateUnreachable();
ExitBlocks.push_back(Builder.GetInsertBlock());
auto *NextBB = BasicBlock::Create(Context, "", TheFunction);
Blocks.push_back(NextBB);
Builder.SetInsertPoint(NextBB);
Variables.newExtendedBasicBlock();
}