Skip to content
Performance benchmarks of Python, Numpy, etc. vs. other languages such as Matlab, Julia, Fortran.
Python Fortran CMake MATLAB C Meson Other
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
NASA_JulesKouatchou
cmake
iter
mandelbrot
matmul
pisum
pythonperformance
tests
.appveyor.yml
.coveragerc
.flake8
.gitattributes
.gitignore
.travis.yml
CMakeLists.txt
LICENSE.txt
Matmul.py
NoneVsNan.py
NothingVsNan.jl
Pisum.py
README.md
RunHypot.py
WindowsStderr.py
carriage_return.py
checkversions.py
dis_blank.c
dis_div.c
dis_mult.c
h5py_write_speed.py
hypot.f90
isoctave.m
kind.f90
meson.build
mypy.ini
nans.py
netcdf_write_speed.py
no_traceback_error.py
perf.f90
pyproject.toml
setup.cfg
setup.py

README.md

image Coverage Status image

Python Performance

All benchmarks are platform-independent (run on any computing device with appropriate hardware). CuPy tests require an NVIDIA GPU with CUDA toolkit installed.

Install

This command prepares Python prereqs:

pip install -e .

Using the C and/or Fortran tests requires compilation using CMake.

cmake -B build

cmake --build build --target test  # non-Visual Studio

cmake --build build --target RUN_TESTS  # Visual Studio

Usage

Iterative benchmarks, here using the pisum algorithm:

python Pisum.py

Pi (Machin) benchmark Windows 10

Pi (Machin) benchmark

Matrix Multiplication benchmarks:

python Matmul.py

Hypotenuse

Observe that hypot() is faster from 1 to a few hundred elements, then sqrt(x^2+y^2) becomes slightly faster. However, hypot() does not overflow for arguments near REALMAX. For example, in Python:

from math import sqrt, hypot

a=1e154; hypot(a,a); sqrt(a**2+a**2);

1.414213562373095e+154
inf

Execute the Hypot speed test by:

./RunHypot.py

Python 3.6 hypot() vs rsq()

Python 2.7 hypot() vs rsq()

Python 3.5 hypot() vs rsq()

Notes

Julia

Julia binaries are often downloaded to a particular directory. Python doesn't pickup .bash_aliases, which is commonly used to point to Julia.

Compiler selection

Intel Fortran:

FC=ifort cmake ..

GNU Fortran (gfortran ≥ 6 required):

FC=gfortran cmake ..

MKL selection

https://software.intel.com/en-us/articles/intel-mkl-link-line-advisor

We give a hint to CMake where your MKL libraries on. For example:

MKLROOT=/opt/intel/mkl cmake ..

Of course this option can be combined with FC.

You can set this environment variable permanently for your convenience (normally you always want to use MKL) by adding to your ~/.bashrc the line:

export MKLROOT=/opt/intel/mkl
You can’t perform that action at this time.