Clang LLVM为JIT编译功能提供了不良功能指针
clang+llvm provides a bad function pointers for JIT compiled functions
我正在尝试使用clang llvm 3.6来编译几个C函数(每个C函数最终都可以很大)。
不幸的是,i LLVM提供的功能指针使程序segfault。
到目前为止,我有以下代码:
#include <iostream>
#include <clang/CodeGen/CodeGenAction.h>
#include <clang/Basic/DiagnosticOptions.h>
#include <clang/Basic/TargetInfo.h>
#include <clang/Basic/SourceManager.h>
#include <clang/Frontend/CompilerInstance.h>
#include <clang/Frontend/CompilerInvocation.h>
#include <clang/Frontend/FrontendDiagnostic.h>
#include <clang/Frontend/TextDiagnosticPrinter.h>
#include <clang/Frontend/Utils.h>
#include <clang/Parse/ParseAST.h>
#include <clang/Lex/Preprocessor.h>
#include <llvm/Analysis/Passes.h>
#include <llvm/ExecutionEngine/SectionMemoryManager.h>
#include <llvm/ExecutionEngine/MCJIT.h>
#include <llvm/ExecutionEngine/ExecutionEngine.h>
#include <llvm/IR/Verifier.h>
#include <llvm/IR/Module.h>
#include <llvm/IR/LLVMContext.h>
#include <llvm/IR/LegacyPassManager.h>
#include <llvm/Bitcode/ReaderWriter.h>
#include <llvm/Support/ManagedStatic.h>
#include <llvm/Support/MemoryBuffer.h>
#include <llvm/Support/TargetSelect.h>
#include <llvm/Support/raw_os_ostream.h>
#include <llvm/Linker/Linker.h>
int main(int argc, char *argv[]) {
using namespace llvm;
using namespace clang;
static const char* clangArgv [] = {"program", "-x", "c", "string-input"};
static const int clangArgc = sizeof (clangArgv) / sizeof (clangArgv[0]);
// C functions to be compiled (they could eventually be extremely large)
std::map<std::string, std::string> func2Source;
func2Source["getOne"] = "int getOne() {return 1;}";
func2Source["getTwo"] = "int getTwo() {return 2;}";
llvm::InitializeAllTargets();
llvm::InitializeAllAsmPrinters();
std::unique_ptr<llvm::Linker> linker;
std::unique_ptr<llvm::LLVMContext> context(new llvm::LLVMContext());
std::unique_ptr<llvm::Module> module;
/**
* add each C function to the same module
*/
for (const auto& p : func2Source) {
const std::string& source = p.second;
IntrusiveRefCntPtr<DiagnosticOptions> diagOpts = new DiagnosticOptions();
TextDiagnosticPrinter *diagClient = new TextDiagnosticPrinter(llvm::errs(), &*diagOpts); // will be owned by diags
IntrusiveRefCntPtr<DiagnosticIDs> diagID(new DiagnosticIDs());
IntrusiveRefCntPtr<DiagnosticsEngine> diags(new DiagnosticsEngine(diagID, &*diagOpts, diagClient));
ArrayRef<const char *> args(clangArgv + 1, // skip program name
clangArgc - 1);
std::unique_ptr<CompilerInvocation> invocation(createInvocationFromCommandLine(args, diags));
if (invocation.get() == nullptr) {
std::cerr << "Failed to create compiler invocation" << std::endl;
exit(1);
}
CompilerInvocation::setLangDefaults(*invocation->getLangOpts(), IK_C,
LangStandard::lang_unspecified);
invocation->getFrontendOpts().DisableFree = false; // make sure we free memory (by default it does not)
// Create a compiler instance to handle the actual work.
CompilerInstance compiler;
compiler.setInvocation(invocation.release());
// Create the compilers actual diagnostics engine.
compiler.createDiagnostics(); //compiler.createDiagnostics(argc, const_cast<char**> (argv));
if (!compiler.hasDiagnostics()) {
std::cerr << "No diagnostics" << std::endl;
exit(1);
}
// Create memory buffer with source text
std::unique_ptr<llvm::MemoryBuffer> buffer = llvm::MemoryBuffer::getMemBufferCopy(source, "SIMPLE_BUFFER");
if (buffer.get() == nullptr) {
std::cerr << "Failed to create memory buffer" << std::endl;
exit(1);
}
// Remap auxiliary name "string-input" to memory buffer
PreprocessorOptions& po = compiler.getInvocation().getPreprocessorOpts();
po.addRemappedFile("string-input", buffer.release());
// Create and execute the frontend to generate an LLVM bitcode module.
clang::EmitLLVMOnlyAction action(context.get());
if (!compiler.ExecuteAction(action)) {
std::cerr << "Failed to emit LLVM bitcode" << std::endl;
exit(1);
}
std::unique_ptr<llvm::Module> module1 = action.takeModule();
if (module1.get() == nullptr) {
std::cerr << "No module" << std::endl;
exit(1);
}
if (linker.get() == nullptr) {
module.reset(module1.release());
linker.reset(new llvm::Linker(module.get()));
} else {
if (linker->linkInModule(module1.release())) {
std::cerr << "LLVM failed to link module" << std::endl;
exit(1);
}
}
}
llvm::InitializeNativeTarget();
llvm::Module* m = module.get();
std::string errStr;
std::unique_ptr<llvm::ExecutionEngine> executionEngine(EngineBuilder(std::move(module))
.setErrorStr(&errStr)
.setEngineKind(EngineKind::JIT)
.setMCJITMemoryManager(std::unique_ptr<SectionMemoryManager>(new SectionMemoryManager()))
.setVerifyModules(true)
.create());
if (!executionEngine.get()) {
std::cerr << "Could not create ExecutionEngine: " + errStr << std::endl;
exit(1);
}
executionEngine->finalizeObject();
/**
* Lets try to use each function
*/
for (const auto& p : func2Source) {
const std::string& funcName = p.first;
llvm::Function* func = m->getFunction(funcName);
if (func == nullptr) {
std::cerr << "Unable to find function '" << funcName << "' in LLVM module" << std::endl;
exit(1);
}
// Validate the generated code, checking for consistency.
llvm::raw_os_ostream os(std::cerr);
bool failed = llvm::verifyFunction(*func, &os);
if (failed) {
std::cerr << "Failed to verify function '" << funcName << "' in LLVM module" << std::endl;
exit(1);
}
#if 1
func->dump(); // Dump the function for exposition purposes.
// JIT the function, returning a function pointer.
void *fPtr = executionEngine->getPointerToFunction(func); ///// BAD function pointer!!!!
// Cast it to the right type (takes no arguments, returns a double) so we
// can call it as a native function.
int (*funcPtr)();
*(int **) (&funcPtr) = *(int **) fPtr;
int v = (*funcPtr)();
std::cout << "return: " << v << std::endl;
#else // THIS DOES NOT WORK EITHER:
// JIT the function, returning a function pointer.
uint64_t fPtr = executionEngine->getFunctionAddress(funcName); ///// BAD function pointer!!!!
if (fPtr == 0) {
std::cerr << "Unable to find function '" << funcName << "' in LLVM module" << std::endl;
exit(1);
}
int (*funcPtr)();
*(int **) (&funcPtr) = *(int **) fPtr;
int v = (*funcPtr)();
std::cout << "return: " << v << std::endl;
#endif
}
}
任何人可以帮助我解决问题吗?
(我正在Linux-ubuntu 15.04中运行此操作)
此任务令人难以置信:
*(int **) (&funcPtr) = *(int **) fPtr;
编写int*
,然后将其用作下一行的函数指针,不仅违反了严格的注意,而且数据指针通常不够大,无法容纳整个代码指针。
安全的方法是
memcpy(funcPtr, fPtr, sizeof funcPtr);
或
funcPtr = reinterpret_cast<decltype(funcPtr)>(fPtr);
相关文章:
- 在执行其他功能的同时播放动画(LED矩阵和Arduino/ESP8266)
- 多态性和功能结合
- 带内存和隔离功能的SQLite
- 在CMakeLists.txt的安装功能中使用.cmake文件有什么用
- 类模板的成员功能的定义在单独的TU中完全专业化
- 有没有一种方法可以创建一个带有哈希表的数据库,该哈希表具有恒定时间查找功能
- 如何在C++中获得"静态纯虚拟"功能?
- 两个文件使用彼此的功能-如何解决
- 我应该实现右值推送功能吗?我应该使用std::move吗
- QML按钮点击功能执行顺序
- 无法理解此 return 语句的功能,没有它就会发生运行时错误
- 有没有可能有一个只有ADL才能找到的非好友功能
- 功能样式转换从 'int' 到 'ItemType' 的匹配转换
- 文件系统:复制功能的速度秘诀是什么
- 在用于格式4的arm模拟器中实现功能时的一个问题
- 如何在Directwrite中获得给定字体的可用OpenType功能
- 对可变参数使用声明.如何选择正确的功能
- 询问在设计我的手臂模拟器功能表示格式1
- 功能原型的目的
- Clang LLVM为JIT编译功能提供了不良功能指针