jackeown / pythonisnotslow

Simple program to benchmark C vs pure Python vs Numpy vs Cython.

Compare the speed of C, pure Python, Numpy, Cython, and Fortran.

First build the C code, Cython, and Fortran by running make

Next run ./benchmark.sh

Remove built files using make clean

The C code is in benchmark.c
The pure Python code is in naiveBenchmark.py
The numpy Python code is in numpyBenchmark.py
The Cython code is in cythonHelpers.pyx and is imported/run from cythonBenchmark.py
The Fortran code is in benchmark.f90

Here are the results from running ./benchmark.sh on my workstation:

C code:

real    0m0.024s
user    0m0.019s
sys     0m0.005s
#####################################
Naive python:

real    2m16.430s
user    2m15.398s
sys     0m0.676s
#####################################
Numpy python:

real    0m0.793s
user    0m0.213s
sys     0m0.080s
#####################################
Cython python:

real    0m0.036s
user    0m0.028s
sys     0m0.004s
#####################################
(Modern) Fortran code (gfortran):

real    0m0.066s
user    0m0.043s
sys     0m0.006s
#####################################
(Modern) Fortran code (Intel):

real    0m0.114s
user    0m0.110s
sys     0m0.002s
#####################################

pythonisnotslow's Issues

your C code is unoptimized

the reason your C code is slower is because numpy and cython compiles into vectorized instructions. try convert your C code with high performance linear algebra libraries such as Eigen and you'll see very similar performances between the C version and your cython version. (because they would be using the same thing)