Skip to content

Integration

To integrate Proteus into your application, you must:

  1. Annotate functions (or GPU kernels) for JIT specialization.
  2. Modify your build to include: (a) the Proteus LLVM plugin pass, (b) the include directory with Proteus's headers, and link with the runtime library.

This is done by adding Proteus's plugin pass to Clang compilation and its include directory: 

CXXFLAGS += -fpass-plugin=<install-path>/lib64/libProteusPass.so -I<install-path>/include

Also, linking must include the runtime library (preferrably rpath-ed is using Proteus as a dynamic library) and LLVM libraries: 

LDFLAGS += -L <install-path>/lib64 -Wl,-rpath,<install-path>/lib -lproteus $(llvm-config --libs)

A complete example is: 

clang++ -fpass-plugin=<install-path>/lib64/libProteusPass.so \
    -I<install-path>/include \
    -L <install-path>/lib64 -Wl,-rpath,<install-path>/lib64 \
    -lproteus $(llvm-config --libs) \
    MyAwesomeCode.cpp -o MyAwesomeExe

Using CMake

To use Proteus with CMake, make sure the Proteus install directory is in CMAKE_PREFIX_PATH, or pass it as -Dproteus_DIR=<install-path>. Then, in your project's CMakeLists.txt simply add the following two lines:

find_package(proteus CONFIG REQUIRED)

add_proteus(target)

Where target is the name of your library or executable target.

Warning

You may need to build libproteus with the Position Independent Code (PIC) flag (-fPIC or CMAKE_POSITION_INDEPENDENT_CODE=on in CMake), if you want to integrate Proteus as a static library to a dynamic library target.

When using Proteus for JIT generation of CPU code, and your JIT-compiled code needs to call functions defined in your main program, you must link your main application with the -rdynamic flag. This flag ensures that all symbols from your main program are visible to the Proteus JIT, allowing external function calls from JIT code to resolve correctly at runtime.