pybind11 demo

results (release build, i5-4670 @ 3.4Ghz x 4)

Disclaimer: these benchmark were done for fun only. No serious optimization has been done.

ackermann

numbers m=4 n=1 -> 2862984010 calculations

# pure python
scripts/ackermann.py		~10 mins

# numba jit compiled python
scripts/ackermann_numba.py 	~30 secs

# pybind calling c++ code
scripts/ackermann_pybind.py 	~2.6 secs

csvreader

simplistic csv reader, reads ~4096000 lines of csv file, 10 ints each

# pure python
scripts/csvreader.py        ~4.3 secs

# pybind calling c++ code
scripts/csvreader_pybind.py ~1.5 secs

split

splits string. Test with 1048576 chunks

# pure python, must be heavily optimized damn
scripts/split.py               0.050 secs

# pybind calling c++ code, c++ code not really optimized for speed
# this copies the entire vector to a python object, possibly very costly?!
scripts/split_pybind.py	       ~0.070 secs

# pybing calling c++ code, c++ code not really optimized for speed
# not copying should happen here, same performance strangely
scripts/split_mut_pybind.py    ~0.065 secs  

# using naive implementation. This is actually faster for strings with <= ~2000 chunks
# e.g. in csv reader
scripts/split_naive_pybind.py  ~38 secs

install

Dependencies:

python >= 3
cmake >= 3.1
g++, clang supporting c++11

git clone --recursive https://github.com/juliangaal/pybind_demo
cd pybind_demo && mkdir build && cd build
cmake ..
make

Default CMAKE_BUILD_TYPE=Release

Custom Options

LIB_GEN_PATH: Where generated library will be placed. Default: scripts/

Note You may need to manually set the python version to build against, e.g. when encountering a PythonLibs and PythonInterpreter mismatch, in build directory:

cmake -DPYTHON_LIBRARY=$(python-config --prefix)/lib/libpython<version>.dylib -DPYTHON_INCLUDE_DIR=$(python-config --prefix)/include/python<version> ..
make

use from python

To run the compiled c++ lib from python code, in scripts/: run

python-<version you compiled against> <file>.py

The easiest way to use the generated library from anywhere is to add the output path (default scripts) to PYTHONPATH

PYTHONPATH="${PYTHONPATH}:/path/to/this/repo/scripts"
export PYTHONPATH

or change LIB_GEN_PATH to an existing path in $PYTHONPATH