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

cd build

ctest

Usage

Iterative benchmarks, here using the pisum algorithm:

python Pisum.py

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

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