A LLVM pass that prints bytecode programs into the dot format.

This is a DCC888 class project assignment.

You must write your own pass, and it must have a few requirements:

• Each instruction must be printed with an opcode;
• All the arguments of an instruction must be printed next to this instruction;
• Arguments that do not have names, such as getelementptr in function calls, should not be printed;
• Type information should not be printed.

To build this project, you need to run the following shell commands:

Assuming you have ninja-build installed...

$ mkdir -p build
$ cd build
$ cmake -G Ninja -DLLVM_INSTALL_DIR=/path/to/llvm-project/build ..
$ cmake --build .

Once you build the project, all source files are generated as follows:

• All tests' sources are compiled into LLVM IR code, using the following command:

$ clang -S -emit-llvm -o "Filename.c.ll" "Filename.c" -Xclang -disable-O0-optnone

• Since the pass is built out-of-tree, the libBytecodeToDot.so file is compiled and loaded dynamically as plugin using opt by the following command:

$ opt -load-pass-plugin "build/libBytecodeToDot.so" -passes=mem2reg,lcssa,BytecodeToDotPass -S "FileName.c.ll" -o "FileName.c.ll" 2>&1

Note: both commands were added as custom commands in the CMakeLists.txt file.

Also, every DOT file is generated in the same directory as its relative .ll file.

To generate the PDF files from the DOT files, go to the root of the repo again and run the following commands:

$ chmod +x pdf_gen.sh
$ ./pdf_gen.sh

The PDF files for each CFG can be found in the tests/Stanford/TestName.ll_Files/FunctionName.dot.pdf folder.

If python's img2pdf package is installed, pdf files get merged into a single file called tests/Stanford/TestName.ll_merged.pdf.