Programming/Python

Programming Details

Python is a widely used high-level programming language used for general-purpose programming.

An interpreted language, Python has a design philosophy that emphasizes code readability (notably using white-space indentation to delimit code blocks rather than curly braces or keywords), and a syntax that allows programmers to express concepts in fewer lines of code than possible in languages such as C++ or JAVA

When programming with Python in an HPC environment you will need to change the first line as shown below

 #!/usr/bin/python

To

 #!/usr/bin/env python

Python example

#!/usr/bin/env python

from mpi4py import MPI

comm = MPI.COMM_WORLD
rank = comm.Get_rank()

if rank == 0:
   data = {'key1' : [7, 2.72, 2+3j],
           'key2' : ( 'abc', 'xyz')}
else:
   data = None

data = comm.bcast(data, root=0)

if rank != 0:
        print ("data is %s and %d" % (data,rank))
else:
        print ("I am master\n")

Due to the limitations of the Python interpreter (GIL) you need either multiprocessing (mpi4py) or use C extensions that release GIL during computations e.g., Like NumPy functions which are compiled binaries for speed.

Modules Available

The following modules are available:

Python provided by the Python Software foundation

• module add python/2.7.11
• module add python/3.5.1

Anaconda python Anaconda is the open data science platform powered by Continuum

• module add python/anaconda/4.0/2.7
• module add python/anaconda/4.0/3.5
• module add python/anaconda/4.1.1/2.7

and miniconda is provided by

• module add python/anaconda/4.3.31/3.6-VE
• module add python/anaconda/4.6/miniconda/3.7
• module add python/anaconda/202111/3.9
• module add python/anaconda/20220712/3.9

Compilation

Python is byte-compiled at runtime by typing for example

[username@login01 ~]$  python myPython.py

Usage Examples

Batch example

 
#!/bin/bash
#SBATCH -J compute-single-node
#SBATCH -N 1
#SBATCH --ntasks-per-node 20
#SBATCH -o %N.%j.%a.out
#SBATCH -e %N.%j.%a.err
#SBATCH -p compute
#SBATCH --exclusive
#SBATCH --mail-user= your email address here

echo $SLURM_JOB_NODELIST

module purge
module add python/anaconda/20220712/3.9
module add openmpi/gcc/1.10.2

mpirun python broadcast.py

[username@login01 ~]$ sbatch python-demo.job
Submitted batch job 289572

Python and OpenMP

The Python interpreter uses a GIL (Global Interpreter Lock) which makes multi-threading almost impossible. Although there are ways around this and one of the most common methods is to use Cython.

Cython supports native parallelism through the cython.parallel module. To use this kind of parallelism, the GIL must be released (see Releasing the GIL). It currently supports OpenMP

• Example with a reduction (on sum):

#!/usr/bin/env python
from cython.parallel import prange

cdef int i
cdef int n = 30
cdef int sum = 0

for i in prange(n, nogil=True):
    sum += i

print(sum)

• Example with a typed memoryview (e.g. a NumPy array):

#!/usr/bin/env python
from cython.parallel import prange

def func(double[:] x, double alpha):
    cdef Py_ssize_t i

    for i in prange(x.shape[0]):
        x[i] = alpha * x[i]

To actually use the OpenMP support, you need to tell the C or C++ compiler to enable OpenMP. For GCC this can be done as follows in a setup.py:

#!/usr/bin/env python
from distutils.core import setup
from distutils.extension import Extension
from Cython.Build import cythonize

ext_modules = [
    Extension(
        "hello",
        ["hello.pyx"],
        extra_compile_args=['-fopenmp'],
        extra_link_args=['-fopenmp'],
    )
]

setup(
    name='hello-parallel-world',
    ext_modules=cythonize(ext_modules),
)

• See Cython Parallelism documentation for more details here.

Numba

Alternatively to Cython, in cases where appropriate (if you are using NumPy this may be useful!) Numba can provide slightly worse performance but is much simpler to use.

For example a matrix multiplication:

@njit(parallel=True)
def matmult(a,b):
        assert A.shape[1] == B.shape[0]
        res = np.zeros((A.shape[0], B.shape[1]), )
        for i in prange (A.shape[0]):
                for k in range (A.shape[1]):
                        for j in range(B.shape[1]):
                                res[i,j]+=A[i,k] * B[k,j]
        return res

Numba also has support for CUDA GPU programming: https://numba.readthedocs.io/en/stable/cuda/overview.html

Visit the Numba website for more details: https://numba.pydata.org/.

Next Steps

• https://en.wikipedia.org/wiki/Python_(programming_language)
• https://www.python.org/
• https://www.continuum.io/anaconda-overview
• OpenMPI
• OpenMP
• https://numba.pydata.org/

If you are trying to speed up your Python but need some help you can contact our RSE team in the Support Portal!



Languages | Main Page | Further Topics