Real Semantics

Building the custom LLVM interpreter (lli)

Prerequisites

• A modern C++ toolchain (>=g++-4.8 or a recent version of clang)
• MPFR library (and also GMP, but it should install automatically if you get it through some sort of package manager)
• List of dependencies described in the LLVM Getting Started page. A few comments...
  • Install the build-essential package if using Ubuntu Linux: that should install most of the required packages
  • If you're building on Mac, make sure to install all packages required by LLVM, including something called m4 and a bunch of compression libraries
  • Be sure to install libffi, or library calls such as printf() won't work. On Ubuntu Linux systems it should come with the box. If not, the apt-get package name is libffi-dev. Sorry but I don't know its OSX equivalence...
  • Install g++-multilib, if you are using gcc to build LLVM
  • Install cmake, which is used to configure the build system

After you get all the tools...

1. Clone this repository and checkout the llvm submodule:

   $ git clone [email protected]:danking/real-semantics.git
   $ cd real-semantics
   $ git submodule init && git submodule update
   $ cd llvm-3.6.0

2. If you would like to get the latest LLVM code (which could be broken), switch to master branch and do a pull. You can skip this step if you just want to try it out.

   $ git checkout master
   $ git pull

3. Now we can start building LLVM:

   $ mkdir build
   $ cd build
   build$ cmake .. -DLLVM_ENABLE_EH=ON -DLLVM_ENABLE_RTTI=ON -DLLVM_ENABLE_FFI=ON
   build$ make -j2 # or whatever number of processors you have

   During the third (cmake) step, you can pass -DCMAKE_BUILD_TYPE=DEBUG -DLLVM_ENABLE_ASSERTIONS=ON as additional arguments if you wish to do a debug build.

4. Keep your fingers crossed and hope that it will build without error... Actually it will probably fail at a final linking step because I haven't figured out the Makefile yet. If that happens, go to build/tools/lli/CMakeFiles/lli.dir/link.txt and add -lmpfr -lgmp to the end of that file, and run make (in the build directory!) again.

5. After it successfully builds, the executable we care about is located at build/bin/lli.

Compiling C/C++ programs into LLVM byte code and run them

1. Make sure that you have clang or clang++ installed because we didn't actually build clang when we built LLVM.

2. Generating byte code is easy:

   # this will generate the actual "byte" code
   $ clang -emit-llvm -c program.c -o program.bc
   # this will generate a human readable LLVM IR program
   $ clang -S -emit-llvm -c program.c -o program.ll

   For complete debugging information, use the clang-3.6 executable:

   $ ./clang-3.6 -I/usr/include/clang/3.4/include -emit-llvm -c -g program.c -o program.bc

   Notes:

   • Be sure to supply the -c flag; -emit-llvm doesn't work without it.
   • Supply the -g flag if you want extra debugging information (e.g. line number)
   • Supply the -fno-use-cxa-atexit flag if you encounter a __dso_handle related error in the next step.

3. To run the byte code with lli, use the .bc file. DO NOT forget to invoke lli with the -force-interpreter flag:

   # make sure you are using the correct lli executable...
   $ ./lli -force-interpreter program.bc