如何在 LLVM IR 的 switch 指令中使用 char*？答案

【问题标题】：How to use char* in switch instruction of LLVM IR?如何在 LLVM IR 的 switch 指令中使用 char*？
【发布时间】：2020-08-14 13:38:30
【问题描述】：

我正在尝试在 switch 指令中使用字符串。我的想法是使用字符串的地址作为字符串的值（ConstantInt*）传递给switch指令，如果两个地址相等，switch case应该匹配。

但是当我在 IR 代码（由我的程序生成）下运行时，会发生错误：

source_filename = "input/step/step1.ts"
target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-apple-darwin19.6.0"

@aa = private unnamed_addr constant [3 x i8] c"aa\00", align 1
@b = linkonce_odr global i8* getelementptr inbounds ([3 x i8], [3 x i8]* @aa, i32 0, i32 0)
@bb = private unnamed_addr constant [3 x i8] c"bb\00", align 1
define i32 @main() {
  %1 = load i8*, i8** @b
  switch i8* %1, label %4 [
    i8* getelementptr inbounds ([3 x i8], [3 x i8]* @aa, i32 0, i32 0), label %2
    i8* getelementptr inbounds ([3 x i8], [3 x i8]* @bb, i32 0, i32 0), label %3
  ]

2:                                                ; preds = %0
  br label %4

3:                                                ; preds = %0
  br label %4

4:                                                ; preds = %2, %3, %0
  ret i32 0
}

语言代码如下：

var b: string = "aa";  
switch (b) {  
  case "aa":  
    break;  
  case "bb":  
    break;  
}

IR 代码似乎是正确的，但运行时出现错误：

Stack dump:
0.      Program arguments: ./build/tser input/step/step1.ts 
1.      Running pass 'Function Pass Manager' on module 'input/step/step1.ts'.
2.      Running pass 'X86 DAG->DAG Instruction Selection' on function '@main'
0  tser                     0x00000001073a718c llvm::sys::PrintStackTrace(llvm::raw_ostream&) + 60
1  tser                     0x00000001073a7749 PrintStackTraceSignalHandler(void*) + 25
2  tser                     0x00000001073a52b6 llvm::sys::RunSignalHandlers() + 118
3  tser                     0x00000001073ab21c SignalHandler(int) + 252
4  libsystem_platform.dylib 0x00007fff6efa65fd _sigtramp + 29
5  libsystem_platform.dylib 0x00007ffee8c53700 _sigtramp + 18446744071457919264
6  tser                     0x0000000107d22fad llvm::APInt::operator[](unsigned int) const + 141
7  tser                     0x0000000107d22e8d llvm::APInt::isNegative() const + 29
8  tser                     0x00000001071d2d92 llvm::APInt::compareSigned(llvm::APInt const&) const + 242
9  tser                     0x0000000107d2814d llvm::APInt::slt(llvm::APInt const&) const + 29
10 tser                     0x0000000108dda1ee llvm::SwitchCG::sortAndRangeify(std::__1::vector<llvm::SwitchCG::CaseCluster, std::__1::allocator<llvm::SwitchCG::CaseCluster> >&)::$_1::operator()(llvm::SwitchCG::CaseCluster const&, llvm::SwitchCG::CaseCluster const&) const + 62
11 tser                     0x0000000108dd9b53 void std::__1::__sort<llvm::SwitchCG::sortAndRangeify(std::__1::vector<llvm::SwitchCG::CaseCluster, std::__1::allocator<llvm::SwitchCG::CaseCluster> >&)::$_1&, llvm::SwitchCG::CaseCluster*>(llvm::SwitchCG::CaseCluster*, llvm::SwitchCG::CaseCluster*, llvm::SwitchCG::sortAndRangeify(std::__1::vector<llvm::SwitchCG::CaseCluster, std::__1::allocator<llvm::SwitchCG::CaseCluster> >&)::$_1&) + 147
12 tser                     0x0000000108dd9ab5 void std::__1::sort<llvm::SwitchCG::CaseCluster*, llvm::SwitchCG::sortAndRangeify(std::__1::vector<llvm::SwitchCG::CaseCluster, std::__1::allocator<llvm::SwitchCG::CaseCluster> >&)::$_1&>(llvm::SwitchCG::CaseCluster*, llvm::SwitchCG::CaseCluster*, llvm::SwitchCG::sortAndRangeify(std::__1::vector<llvm::SwitchCG::CaseCluster, std::__1::allocator<llvm::SwitchCG::CaseCluster> >&)::$_1&) + 37
13 tser                     0x0000000108dd9a86 void std::__1::sort<llvm::SwitchCG::CaseCluster, llvm::SwitchCG::sortAndRangeify(std::__1::vector<llvm::SwitchCG::CaseCluster, std::__1::allocator<llvm::SwitchCG::CaseCluster> >&)::$_1>(std::__1::__wrap_iter<llvm::SwitchCG::CaseCluster*>, std::__1::__wrap_iter<llvm::SwitchCG::CaseCluster*>, llvm::SwitchCG::sortAndRangeify(std::__1::vector<llvm::SwitchCG::CaseCluster, std::__1::allocator<llvm::SwitchCG::CaseCluster> >&)::$_1) + 54
14 tser                     0x0000000108dd99dd void llvm::sort<std::__1::__wrap_iter<llvm::SwitchCG::CaseCluster*>, llvm::SwitchCG::sortAndRangeify(std::__1::vector<llvm::SwitchCG::CaseCluster, std::__1::allocator<llvm::SwitchCG::CaseCluster> >&)::$_1>(std::__1::__wrap_iter<llvm::SwitchCG::CaseCluster*>, std::__1::__wrap_iter<llvm::SwitchCG::CaseCluster*>, llvm::SwitchCG::sortAndRangeify(std::__1::vector<llvm::SwitchCG::CaseCluster, std::__1::allocator<llvm::SwitchCG::CaseCluster> >&)::$_1) + 45
15 tser                     0x0000000108dd06f3 void llvm::sort<std::__1::vector<llvm::SwitchCG::CaseCluster, std::__1::allocator<llvm::SwitchCG::CaseCluster> >&, llvm::SwitchCG::sortAndRangeify(std::__1::vector<llvm::SwitchCG::CaseCluster, std::__1::allocator<llvm::SwitchCG::CaseCluster> >&)::$_1>(std::__1::vector<llvm::SwitchCG::CaseCluster, std::__1::allocator<llvm::SwitchCG::CaseCluster> >&, llvm::SwitchCG::sortAndRangeify(std::__1::vector<llvm::SwitchCG::CaseCluster, std::__1::allocator<llvm::SwitchCG::CaseCluster> >&)::$_1) + 51
16 tser                     0x0000000108dd0491 llvm::SwitchCG::sortAndRangeify(std::__1::vector<llvm::SwitchCG::CaseCluster, std::__1::allocator<llvm::SwitchCG::CaseCluster> >&) + 225
17 tser                     0x00000001084fe009 llvm::SelectionDAGBuilder::visitSwitch(llvm::SwitchInst const&) + 569
18 tser                     0x00000001084fb508 llvm::SelectionDAGBuilder::visit(unsigned int, llvm::User const&) + 152
19 tser                     0x00000001084fac9d llvm::SelectionDAGBuilder::visit(llvm::Instruction const&) + 141
20 tser                     0x00000001086002cc llvm::SelectionDAGISel::SelectBasicBlock(llvm::ilist_iterator<llvm::ilist_detail::node_options<llvm::Instruction, true, false, void>, false, true>, llvm::ilist_iterator<llvm::ilist_detail::node_options<llvm::Instruction, true, false, void>, false, true>, bool&) + 220
21 tser                     0x00000001085ffcbe llvm::SelectionDAGISel::SelectAllBasicBlocks(llvm::Function const&) + 6110
22 tser                     0x00000001085fc9f2 llvm::SelectionDAGISel::runOnMachineFunction(llvm::MachineFunction&) + 2130
23 tser                     0x0000000107a281d5 (anonymous namespace)::X86DAGToDAGISel::runOnMachineFunction(llvm::MachineFunction&) + 277
24 tser                     0x00000001089ebee6 llvm::MachineFunctionPass::runOnFunction(llvm::Function&) + 550
25 tser                     0x00000001075cb369 llvm::FPPassManager::runOnFunction(llvm::Function&) + 681
26 tser                     0x00000001075cb945 llvm::FPPassManager::runOnModule(llvm::Module&) + 117
27 tser                     0x00000001075cc0a8 (anonymous namespace)::MPPassManager::runOnModule(llvm::Module&) + 776
28 tser                     0x00000001075cbbd5 llvm::legacy::PassManagerImpl::run(llvm::Module&) + 293
29 tser                     0x00000001075cc6d1 llvm::legacy::PassManager::run(llvm::Module&) + 33
30 tser                     0x0000000107775275 llvm::orc::SimpleCompiler::operator()(llvm::Module&) + 309
31 tser                     0x000000010786ed64 decltype(std::__1::forward<llvm::orc::TMOwningSimpleCompiler&>(fp)(std::__1::forward<llvm::Module&>(fp0))) std::__1::__invoke<llvm::orc::TMOwningSimpleCompiler&, llvm::Module&>(llvm::orc::TMOwningSimpleCompiler&, llvm::Module&) + 68
32 tser                     0x000000010786ecd4 llvm::Expected<std::__1::unique_ptr<llvm::MemoryBuffer, std::__1::default_delete<llvm::MemoryBuffer> > > std::__1::__invoke_void_return_wrapper<llvm::Expected<std::__1::unique_ptr<llvm::MemoryBuffer, std::__1::default_delete<llvm::MemoryBuffer> > > >::__call<llvm::orc::TMOwningSimpleCompiler&, llvm::Module&>(llvm::orc::TMOwningSimpleCompiler&, llvm::Module&) + 68
33 tser                     0x000000010786ec84 std::__1::__function::__alloc_func<llvm::orc::TMOwningSimpleCompiler, std::__1::allocator<llvm::orc::TMOwningSimpleCompiler>, llvm::Expected<std::__1::unique_ptr<llvm::MemoryBuffer, std::__1::default_delete<llvm::MemoryBuffer> > > (llvm::Module&)>::operator()(llvm::Module&) + 68
34 tser                     0x000000010786daf3 std::__1::__function::__func<llvm::orc::TMOwningSimpleCompiler, std::__1::allocator<llvm::orc::TMOwningSimpleCompiler>, llvm::Expected<std::__1::unique_ptr<llvm::MemoryBuffer, std::__1::default_delete<llvm::MemoryBuffer> > > (llvm::Module&)>::operator()(llvm::Module&) + 67
35 tser                     0x000000010781d127 std::__1::__function::__value_func<llvm::Expected<std::__1::unique_ptr<llvm::MemoryBuffer, std::__1::default_delete<llvm::MemoryBuffer> > > (llvm::Module&)>::operator()(llvm::Module&) const + 103
36 tser                     0x000000010781d08f std::__1::function<llvm::Expected<std::__1::unique_ptr<llvm::MemoryBuffer, std::__1::default_delete<llvm::MemoryBuffer> > > (llvm::Module&)>::operator()(llvm::Module&) const + 63
37 tser                     0x000000010781bb26 decltype(fp(decltype(std::__1::__declval<llvm::Module&>(0)) std::__1::declval<llvm::Module&>()())) llvm::orc::ThreadSafeModule::withModuleDo<std::__1::function<llvm::Expected<std::__1::unique_ptr<llvm::MemoryBuffer, std::__1::default_delete<llvm::MemoryBuffer> > > (llvm::Module&)>&>(llvm::Module&) + 182
38 tser                     0x000000010781b84b llvm::orc::IRCompileLayer::emit(llvm::orc::MaterializationResponsibility, llvm::orc::ThreadSafeModule) + 187
39 tser                     0x0000000107847cc6 llvm::orc::BasicIRLayerMaterializationUnit::materialize(llvm::orc::MaterializationResponsibility) + 470
40 tser                     0x0000000107873280 llvm::orc::MaterializationUnit::doMaterialize(llvm::orc::JITDylib&) + 112
41 tser                     0x0000000107786dc0 llvm::orc::ExecutionSession::materializeOnCurrentThread(llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >) + 32
42 tser                     0x00000001077931a4 decltype(std::__1::forward<void (*&)(llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >)>(fp)(std::__1::forward<llvm::orc::JITDylib&>(fp0), std::__1::forward<std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> > >(fp0))) std::__1::__invoke<void (*&)(llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >), llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> > >(void (*&)(llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >), llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >&&) + 84
43 tser                     0x0000000107793127 void std::__1::__invoke_void_return_wrapper<void>::__call<void (*&)(llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >), llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> > >(void (*&)(llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >), llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >&&) + 71
44 tser                     0x00000001077930d7 std::__1::__function::__alloc_func<void (*)(llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >), std::__1::allocator<void (*)(llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >)>, void (llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >)>::operator()(llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >&&) + 71
45 tser                     0x0000000107791ec6 std::__1::__function::__func<void (*)(llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >), std::__1::allocator<void (*)(llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >)>, void (llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >)>::operator()(llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >&&) + 70
46 tser                     0x000000010778d01a std::__1::__function::__value_func<void (llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >)>::operator()(llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >&&) const + 106
47 tser                     0x000000010778cec5 std::__1::function<void (llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >)>::operator()(llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >) const + 69
48 tser                     0x0000000107783143 llvm::orc::ExecutionSession::dispatchMaterialization(llvm::orc::JITDylib&, std::__1::unique_ptr<llvm::orc::MaterializationUnit, std::__1::default_delete<llvm::orc::MaterializationUnit> >) + 163
49 tser                     0x0000000107784c50 llvm::orc::ExecutionSession::runOutstandingMUs() + 320
50 tser                     0x0000000107781f4c llvm::orc::ExecutionSession::lookup(llvm::orc::LookupKind, std::__1::vector<std::__1::pair<llvm::orc::JITDylib*, llvm::orc::JITDylibLookupFlags>, std::__1::allocator<std::__1::pair<llvm::orc::JITDylib*, llvm::orc::JITDylibLookupFlags> > > const&, llvm::orc::SymbolLookupSet, llvm::orc::SymbolState, llvm::unique_function<void (llvm::Expected<llvm::DenseMap<llvm::orc::SymbolStringPtr, llvm::JITEvaluatedSymbol, llvm::DenseMapInfo<llvm::orc::SymbolStringPtr>, llvm::detail::DenseMapPair<llvm::orc::SymbolStringPtr, llvm::JITEvaluatedSymbol> > >)>, std::__1::function<void (llvm::DenseMap<llvm::orc::JITDylib*, llvm::DenseSet<llvm::orc::SymbolStringPtr, llvm::DenseMapInfo<llvm::orc::SymbolStringPtr> >, llvm::DenseMapInfo<llvm::orc::JITDylib*>, llvm::detail::DenseMapPair<llvm::orc::JITDylib*, llvm::DenseSet<llvm::orc::SymbolStringPtr, llvm::DenseMapInfo<llvm::orc::SymbolStringPtr> > > > const&)>) + 892
51 tser                     0x0000000107787b34 llvm::orc::ExecutionSession::lookup(std::__1::vector<std::__1::pair<llvm::orc::JITDylib*, llvm::orc::JITDylibLookupFlags>, std::__1::allocator<std::__1::pair<llvm::orc::JITDylib*, llvm::orc::JITDylibLookupFlags> > > const&, llvm::orc::SymbolLookupSet const&, llvm::orc::LookupKind, llvm::orc::SymbolState, std::__1::function<void (llvm::DenseMap<llvm::orc::JITDylib*, llvm::DenseSet<llvm::orc::SymbolStringPtr, llvm::DenseMapInfo<llvm::orc::SymbolStringPtr> >, llvm::DenseMapInfo<llvm::orc::JITDylib*>, llvm::detail::DenseMapPair<llvm::orc::JITDylib*, llvm::DenseSet<llvm::orc::SymbolStringPtr, llvm::DenseMapInfo<llvm::orc::SymbolStringPtr> > > > const&)>) + 324
52 tser                     0x0000000107787dc3 llvm::orc::ExecutionSession::lookup(std::__1::vector<std::__1::pair<llvm::orc::JITDylib*, llvm::orc::JITDylibLookupFlags>, std::__1::allocator<std::__1::pair<llvm::orc::JITDylib*, llvm::orc::JITDylibLookupFlags> > > const&, llvm::orc::SymbolStringPtr) + 307
53 tser                     0x00000001078533fd llvm::orc::LLJIT::lookupLinkerMangled(llvm::orc::JITDylib&, llvm::StringRef) + 173
54 tser                     0x0000000106fcdcf9 llvm::orc::LLJIT::lookup(llvm::orc::JITDylib&, llvm::StringRef) + 169
55 tser                     0x0000000106fc1acb llvm::orc::LLJIT::lookup(llvm::StringRef) + 75
56 tser                     0x0000000106fc0888 main + 2632
57 libdyld.dylib            0x00007fff6edadcc9 start + 1
58 libdyld.dylib            0x0000000000000002 start + 18446603338656326458
make: *** [start] Segmentation fault: 11

没有任何教程，我找不到它，非常感谢任何人的帮助。

【问题讨论】：

您不能使用字符串，因为它们的值是一个地址，并且为时已晚。只有在编译器将开关转换为一系列机器代码指令之后，才在链接时分配。

标签： c++ clang llvm llvm-ir

【解决方案1】：

您将不得不阅读 llvm 文档：https://llvm.org/docs/LangRef.html#switch-instruction。请注意，在 switch 语句中使用 strings 是不正确的。它必须在switch 语句的参数中是一个整数，这样llvm 字节码才能正常工作。

您必须将给定的switch 语句（如您向我们展示的那样）转换为与llvm 字节码中的switch 等效的if..else-if..else，因为将字符串直接传递到switch 语句中不会像您预期的那样执行。当 switch 表达式不是 int 时，将 switch 转换为等效的 if..else-if..else 并且它应该可以工作。

【讨论】：

我终于找到了另一种方法，我认为它比纯if-else实现更好。它改为使用案例索引，并结合 switch 指令和if-else 指令。你可以看看我的回答：stackoverflow.com/a/63453227/11354404.

【解决方案2】：

最后我找到了一个更好的方法，我认为它比纯 if-else 实现更好。它改用index of cases，并结合了switch指令和if-else指令：

将case的字符串值转换为case的索引，类型为ConstantInt。
计算case值的匹配索引，这一步使用if-else，在switch指令之前运行。
然后使用switch指令（索引为键）。

如以下代码：

var a: number = 11;

var b: string = "aa";
switch (b) {
  case "aa":
    a = a + 100;
    break;
  case "bb":
    a = a + 200;
    break;
}

将其转换为以下 IR 代码：

@a = linkonce_odr global i32 11
@aa = private unnamed_addr constant [3 x i8] c"aa\00", align 1
@bb = private unnamed_addr constant [3 x i8] c"bb\00", align 1
@b = linkonce_odr global i8* getelementptr inbounds ([3 x i8], [3 x i8]* @aa, i32 0, i32 0)

define i32 @main() {
  %1 = alloca i32
  %2 = load i8*, i8** @b
  %3 = icmp eq i8* %2, getelementptr inbounds ([3 x i8], [3 x i8]* @aa, i32 0, i32 0)
  br i1 %3, label %4, label %5

4:                                                ; preds = %0
  store i32 0, i32* %1
  br label %5

5:                                                ; preds = %4, %0
  %6 = load i8*, i8** @b
  %7 = icmp eq i8* %6, getelementptr inbounds ([3 x i8], [3 x i8]* @bb, i32 0, i32 0)
  br i1 %7, label %8, label %9

8:                                                ; preds = %5
  store i32 1, i32* %1
  br label %9

9:                                                ; preds = %8, %5
  %10 = load i32, i32* %1
  switch i32 %10, label %17 [
    i32 0, label %11
    i32 1, label %14
  ]

11:                                               ; preds = %9
  %12 = load i32, i32* @a
  %13 = add i32 %12, 100
  store i32 %13, i32* @a
  br label %17

14:                                               ; preds = %9
  %15 = load i32, i32* @a
  %16 = add i32 %15, 200
  store i32 %16, i32* @a
  br label %17

17:                                               ; preds = %9, %11, %14
  ret i32 0
}

【讨论】：