Python is just an idiom

Christophe Brun, PapIT (christophe.brun@papit.fr)

Abstract

This article and code repository aim to highlight that Python is just an idiom, a syntax. A langage is just a langage. It is not directly executed.

It is easily highlighted when the same code is executed with compltly different performances depending on their implementation.

Regardless of what can be said by those many experts benchmarking langages for any purpose. The same Python code will be here executed in different environnements, in their latest versions, highlighting performance differences and potential outcomes.

Those environnements are just a set of relevant Python environments but are not by far exhaustive.

An eye-opener was the 2015 Green Code Lab Challenge¹, 82 teams in 8 countries, where the langages where ranked the following:

Python
C++
Python
Javascript
Python
C++
Python
Python
Python
Python

Each team has a raspberry as server model, running Greenspector. Keep in mind: Lot of teams don’t success in having measures or functional application. The reasons : no agility for certain teams, tunnel effect with a rush the Friday, application not robust… i.e. the challenge was functionally complex enough to use simple idiom like Python.

A common misconception to compare langages without carrying about the runtime environnement

Many articles, including academics, are comparing the efficiency of a programming langage under multiple aspects, such as execution time, energy, etc. Some of them not even mentioning the environment and the langage implementation like João Saraiva et al.². This raise many questions for a langage such as Python having numerous implementations, 3 of them, CPython, Cython, Pypy being really mainstream. This paper measure the CPU time and state the Energy = Power x Time, but not sure, they refer to article stating the contrary.

Alibabacloud performed a survey on energy efficiency, without detailed method³

Overwhelming results in Google:

Multiplying the wrong way

120000000 * 5

The multiplication above is almost instantaneous in the CPython interpreter, hopefully.

from dis import dis


def add5(n):
    s = 0
    for _ in range(n):
        s += 5
    return s


add5(120000000)
dis(add5)

The above program is doing the same plus trying to disassemble it. This is obviously slower when interpreted in CPython. But a smart compiler is probably able to optimize those addition turning it to a multiplication. Let's try different Python implementations compiled or interpreted to look for major performance differences in the same langage.

Cython⁴

A static compiler for Python programming langage. Many would say a transpiler to C/C++. The code is then compiler, here with clang.

Compilation

cython --embed -o multiply.c -3 multiply.py
clang -O3 multiply.c -o multiply -I/usr/include/python3.10 -lpython3.10

Environnement

chrichri@chrichri-HKD-WXX:~/python-is-a-syntax$ cython -V
Cython version 0.29.32
chrichri@chrichri-HKD-WXX:~/python-is-a-syntax$ clang -v
Ubuntu clang version 14.0.0-1ubuntu1
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin
Found candidate GCC installation: /usr/bin/../lib/gcc/x86_64-linux-gnu/10
Found candidate GCC installation: /usr/bin/../lib/gcc/x86_64-linux-gnu/11
Selected GCC installation: /usr/bin/../lib/gcc/x86_64-linux-gnu/11
Candidate multilib: .;@m64
Selected multilib: .;@m64

Speed and disassembly

chrichri@chrichri-HKD-WXX:~/python-is-a-syntax$ time ./multiply
600000000

real    0m6,646s
user    0m6,505s
sys     0m0,012s

Pypy⁵

A fast, compliant alternative implementation of Python... On average, PyPy is 4.5 times faster than CPython. Use a JIT (Just In Time) Compiler. Guido van Rossum would have said at 2015 EuroPython that Pypy is the workaround to GIL issues .

Slightly faster than CPython but no disassembly.

Environnement

Python 3.9.12 (05fbe3aa5b0845e6c37239768aa455451aa5faba, Mar 29 2022, 08:15:34)
[PyPy 7.3.9 with GCC 10.2.1 20210130 (Red Hat 10.2.1-11)] on linux

Speed and disassembly

chrichri@chrichri-HKD-WXX:~/python-is-a-syntax$ time ./pypy3.9-v7.3.9-linux64/bin/python3 multiply.py 
  5           0 LOAD_CONST               1 (0)
              2 STORE_FAST               1 (s)

  6           4 LOAD_GLOBAL              0 (range)
              6 LOAD_FAST                0 (n)
              8 CALL_FUNCTION            1
             10 GET_ITER
        >>   12 FOR_ITER                12 (to 26)
             14 STORE_FAST               2 (_)

  7          16 LOAD_FAST                1 (s)
             18 LOAD_CONST               2 (5)
             20 INPLACE_ADD
             22 STORE_FAST               1 (s)
             24 JUMP_ABSOLUTE           12

  8     >>   26 LOAD_FAST                1 (s)
             28 RETURN_VALUE

real    0m0,413s
user    0m0,372s
sys     0m0,025s

CPython⁶

Original implementation from 1991 by Guido van Rossum.

Environnement

Python 3.10.6 (main, Nov  2 2022, 18:53:38) [GCC 11.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.

Speed and disassembly

chrichri@chrichri-HKD-WXX:~/python-is-a-syntax$ time python3 multiply.py
  5           0 LOAD_CONST               1 (0)
              2 STORE_FAST               1 (s)

  6           4 LOAD_GLOBAL              0 (range)
              6 LOAD_FAST                0 (n)
              8 CALL_FUNCTION            1
             10 GET_ITER
        >>   12 FOR_ITER                 6 (to 26)
             14 STORE_FAST               2 (_)

  7          16 LOAD_FAST                1 (s)
             18 LOAD_CONST               2 (5)
             20 INPLACE_ADD
             22 STORE_FAST               1 (s)
             24 JUMP_ABSOLUTE            6 (to 12)

  8     >>   26 LOAD_FAST                1 (s)
             28 RETURN_VALUE

real    0m8,465s
user    0m8,442s
sys     0m0,016s

GraalVM Python⁷

The polyglot JVM by oracle. It aims to move as many possible application to Oracle GraalVM in cloud services, CoBOL, CPP, Java, Python. Python in GRaalVM can access AWT! On performance side, according to Oracle there is On average, Python in GraalVM (entreprise) is 8.92x faster than CPython

Environnement

chrichri@chrichri-HKD-WXX:~/python-is-a-syntax$ ./graalvm-ce-java11-linux-amd64-22.3.0/graalvm-ce-java11-22.3.0/bin/graalpy 
Python 3.8.5 (Thu Oct 20 10:52:00 UTC 2022)
[Graal, GraalVM CE, Java 11.0.17] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>>

Speed and disassembly

chrichri@chrichri-HKD-WXX:~/python-is-a-syntax$ time ./graalvm-ce-java11-linux-amd64-22.3.0/graalvm-ce-java11-22.3.0/bin/graalpy multiply.py 

real    0m3,078s
user    0m8,520s
sys     0m0,434s

The dis disassembly appears to be missing. Somehow it look mocked as it can be imported but nothing is ever returned.

Performance discussion

Implementations are ordered by user time.

Not obvious as the disassembly of CPython Pypy where pretty much the same but there is a 20 time speed improvement with Pypy. Pypy JIT understand at some point this addition is in fact a multiplication.

In the same langage, Python, the CPython can be more than 20 time slower than Pypy implementation:

Implementation	real	user	Ratio user time to faster
Pypy	0,413	0,372	1
GraalVM	3,078	8,52	7,45
Cython	6,653	6,645	16,11
CPython	8,465	8,442	20,5

Relevance of comparing a langage to another

The famous website landing page statement is unambiguous Which programming language is fastest? Let's go measure … benchmark programs !. This website only compares different version of same implementation. It should probably better compare different implementations. Plus the CPU is the usr+sys time giving a false sense the smallest time is the fastest. Different figures from the Java source⁸ and Python 3 source⁹ Mandelbrot computation from the benchmark game are plotted (Warning time axis in logarithmic scale due to GraalVM Python). Implementations are ordered by user time.

It highlights the two Java implementations consume less resources but from real time point of view, both CPython and Cython are faster. Which fact is completely hidden by the benchmark results. Python is also the slower when running in Graal VM.

phe-sto/Python-is-just-a-syntax

Python is just an idiom

Abstract

A common misconception to compare langages without carrying about the runtime environnement

Multiplying the wrong way

Cython4

Compilation

Environnement

Speed and disassembly

Pypy5

Environnement

Speed and disassembly

CPython6

Environnement

Speed and disassembly

GraalVM Python7

Environnement

Speed and disassembly

Performance discussion

Relevance of comparing a langage to another

Footnotes

Cython⁴

Pypy⁵

CPython⁶

GraalVM Python⁷