PyCCA: Pure-python C compiler and assembler
Luke Campagnola, 2014
Python is an excellent platform for numerical computing but relies heavily on compiled modules to provide optimized functions. For distributed packages, this either increases the burden on the developer to produce compiled binaries for a variety of platforms, or increases the burden on the end user to compile the package or its binary dependencies. Consequently, many Python developers avoid optimzed code, preferring instead to advertise "pure-python" as a feature of their packages.
The objective of pycca is to provide a pure-python approach that allows assembly and C functions to be compiled and executed at runtime with no external dependencies.
PyCCA allows assembler code to be compiled and executed within Python with no external dependencies. This works by:
- Allocating a block of memory with execute privileges.
- Compiling assembly instructions into machine code and writing to executable memory.
- Using the built-in ctypes package to create a python function that points to the compiled machine code.
Can load executable machine code into memory pages and call this executable code via ctypes.
Functional assembly compiler with a relatively limited set of instructions (see examples.py and pycca/asm/instructions.py). All instructions are tested to produce identical output to the GNU assembly compiler.
C compiler in early development
Assembly examples have been tested on:
Linux OSX Windows IA-32 Python 2.7 X X Python 3.4 X Intel-64 Python 2.7 X X Python 3.4 X X
Unit tests pass on 64-bit and 32-bit Linux under python 2.7 and 3.4
- Version 0.3: Basic C compiler (based on pre-parsed data structures) with support for 32- and 64-bit calling conventions on Linux, OSX, and Windows. This will require adding many more floating point and other arithmetic instructions.
- Version 0.4: Parser supporting a subset of C language including functions, control flow, and basic data types.
- Version 0.5: Add support for more SSE2, AVX instructions.