Hope Architecture

The Just-in-time compiling process undergoes several steps. Those steps are explained in the following and detailed in an example and in a dedicated optimization section:

Flow diagram of the step executed during a function call decorated with HOPE.

Start The Python interpreter loads a function or method previously decorated with the @hope.jit decorator.

Cache verification HOPE checks if a compiled version of the requested functions has previously been cached.

Parse function The first time the decorated function is called, the wrapper generates an abstract syntax tree (AST) by parsing the function definition using the Python built-in ast package.

Generate HOPE AST Using the visitor pattern, the Python AST is traversed and a corresponding HOPE specific AST is generated. During the traversal we use the Python built-in inspect package to infer the data types of the live objects such as parameters, variable and return values.

Numerical optimization HOPE traverses the new AST in order to identify numerical optimization possibilities and alters the tree accordingly.

Generate C++ code A C++ code is generated from the HOPE AST.

Compile code to shared object library The Python built-in setuptools package is then used to compile a shared object library from the generated code.

Add library to cache Using the extracted information from the function signature and a hash over the function body the compiled shared object library is cached for future calls.

Load library The shared object library is dynamically loaded into the runtime.

Execute compiled function A call to the function is directed to the function in the shared object library and executed with the passed parameters.

Subsequent function call HOPE analyzes the types of the passed arguments and queries the cache for a function matching the requested name and arguments.

Further reading