Proving that C++ is not slow, it's just a skill issue

This repository contains implementations of Dijkstra algorithm in different programming languages. Quick peek at the results:

All data is included, you are welcome to play with the configuration of the visualizer. The current result was produced with ../script/create_graphics.py --reverse --inverse --relative-slowest.

Goal

The benchmark is a combined benchmark that strives to capture a performance of an average and reasonably-written program. The "reasonable" code is defined as code that:

• Stays readable
• In particular, accesses structure members by names rather than by index
• Stays native, does not delegate the task to other programming languages
• Does not use assembler optimizations
• Does not use hardware acceleration
• Uses standard I/O (one exception)
• Does not use multithreading (one exception)

The execution time is affected by:

• Raw speed
• Efficiency of memory management
• Efficiency of standard library
• Input/output speed
• Initialization time
• etc.

The benchmark does not focus on only one of these aspects. Finding out which exactly part is faster in which languages is outside of the scope of this benchmark.

Notes

The names of the bars on the chart are pretty self-explanatory. Except for:

• C, clang++, release is a C program compiled with clang++ as valid C++.
• C, clang, map+threads is a C program written with multi-threading support and memory mapping enabled. It is written as a demonstration, comparing its performance to other programs is not fair.
• C++, clang++, map+threads is same but with C++.

Hardware

The measurements were performed on Intel Pentium 4415U. To reduce noise, the kernel was booted with isolcpus=1,3, where 1 and 3 are logical cores that correspond to physical core 1.

Software

GCC      13.3.1
Clang    19.1.4
GFortran 13.3.1
MCS      6.12.0.199, Target 4.5
Python   3.12.8
GHC      9.4.8
Node     22.4.1
Matlab   24.2.0.2833386 (R2024b) Update 4
Kernel   6.10.10-zen1-x86_64

Conclusions

• C++ is faster than C (when using standard single-threaded I/O, see "Goal")
• Fortran is not that fast