How to load LLVM bitcode file from an ifstream?

问题

I'm trying to load an LLVM module defined in a .bc file at runtime but have run into a snag.

The bitcode of interest has been generated from hello.cpp:

// hello.cpp
// build with:
// clang-3.4 -c -emit-llvm hello.cpp -o hello.bc
#include <iostream>

void hello()
{
  std::cout << "Hello, world!" << std::endl;
}

When the program below attempts to load it at runtime, it crashes inside llvm::BitstreamCursor::Read():

// main.cpp
// build with:
// g++ main.cpp `llvm-config-3.4 --cppflags --ldflags --libs` -ldl -lpthread -lcurses
#include <llvm/IR/Module.h>
#include <llvm/IRReader/IRReader.h>
#include <llvm/IR/LLVMContext.h>
#include <llvm/Support/SourceMgr.h>
#include <llvm/Support/MemoryBuffer.h>
#include <llvm/Support/raw_ostream.h>
#include <fstream>
#include <iostream>

llvm::Module *load_module(std::ifstream &stream)
{
  if(!stream)
  {
    std::cerr << "error after open stream" << std::endl;
    return 0;
  }

  // load bitcode
  std::string ir((std::istreambuf_iterator<char>(stream)), (std::istreambuf_iterator<char>()));

  // parse it
  using namespace llvm;
  LLVMContext context;
  SMDiagnostic error;
  Module *module = ParseIR(MemoryBuffer::getMemBuffer(StringRef(ir.c_str())), error, context);

  if(!module)
  {
    std::string what;
    llvm::raw_string_ostream os(what);
    error.print("error after ParseIR()", os);
    std::cerr << what;
  } // end if

  return module;
}

int main()
{
  std::ifstream stream("hello.bc", std::ios_base::binary);
  llvm::Module *m = load_module(stream);
  if(m)
  {
    m->dump();
  }

  return 0;
}

I'm building against LLVM v3.4 using the command lines mentioned in the comments.

Any idea what I'm doing wrong?

回答1:

One possible solution is to replace the ParseIR() call with

ParseIRFile("hello.bc", error, getGlobalContext())

Doing this also makes the code a lot simpler since you can skip the part where you try to manually load the bitcode.

The following is sufficient to get a dump for the module (you should add in some error handling to it)

#include <llvm/IR/Module.h>
#include <llvm/IRReader/IRReader.h>
#include <llvm/IR/LLVMContext.h>
#include <llvm/Support/SourceMgr.h>

using namespace llvm;
int main()
{
  LLVMContext context;
  SMDiagnostic error;
  Module *m = ParseIRFile("hello.bc", error, context);
  if(m)
  {
    m->dump();
  } 

  return 0;
}

回答2:

There are two problems:

The lifetime of the LLVMContext needs to outlast the lifetime of the Module. Otherwise the Module will refer to a LLVMContext which no longer exists.
The StringRef referring to the IR should be constructed from the std::string containing the IR, not a zero-terminated string. Otherwise ParseIR won't find the end of the IR correctly.

Here's the corrected version of load_module:

llvm::Module *load_module(std::ifstream &stream, llvm::LLVMContext &context)
{
  if(!stream)
  {
    std::cerr << "error after open stream" << std::endl;
    return 0;
  }

  // load bitcode
  std::string ir((std::istreambuf_iterator<char>(stream)), (std::istreambuf_iterator<char>()));

  // parse it
  using namespace llvm;
  SMDiagnostic error;
  Module *module = ParseIR(MemoryBuffer::getMemBuffer(StringRef(ir)), error, context);

  if(!module)
  {
    std::string what;
    llvm::raw_string_ostream os(what);
    error.print("error after ParseIR()", os);
    std::cerr << what;
  } // end if

  return module;
}

来源：https://stackoverflow.com/questions/22239801/how-to-load-llvm-bitcode-file-from-an-ifstream

标签

c++

llvm

iostream