This is pretty much to keep track on languages progress when it comes to performance. It's just a small test, with almost no real world scenario. But again, it is just to give us some idea and reference between those languages being tested.
The algorithm implemented in all languages is:
- Read the command line arguments, the first argument is an intergen
n(the value used in the results is 800000) - Convert
nto aintor equivalent in the language - Create an
arrayof sizen(when the language support setting a fixed size) - For each position in the
array, fill it with a randomintbetween0andn - then sort the
array
We are using the unix time program to measure the benchmark, so JIT languages with slow cold starts will suffer, that's just part of the benchmark
First language to exit wins.
This benchmark tests memory allocation, memory access and memory management. We are not testing threads, nor IO, but I assume IO won't change much on the result board (we are on 2021 and now every language supports async IO anyways), and threads would possibly only benefit GO for a small margin, not worth the hassle.
Pop_OS! 21.04 LTS
Docker version 20.10.7, build 20.10.7-0ubuntu1~21.04.2
16 Gb RAM, DDR 4, Dual Channel, 2400Mhz
nVidia GeForce GTX 1070 8GB GDDR5
AMD Ryzen 7 1700 3.1GHz
250GB SSD
| # | Language | Time |
|---|---|---|
| 1 | C++ | 0.07s |
| 1 | Rust | 0.07s |
| 1 | C | 0.07s |
| 4 | Haxe (C++) | 0.09s |
| 5 | C# | 0.10s |
| 6 | GO | 0.16s |
| 7 | JavaScript | 0.21s |
| 8 | Nim | 0.22s |
| 9 | PHP | 0.27s |
| 10 | Haxe (HL/JIT) | 0.30s |
| 11 | Java | 0.34s |
| 12 | Ruby | 0.36s |
| 13 | Lua | 0.42s |
| 14 | Dart | 0.56s |
| 15 | GDScript | 0.80s |
| 16 | Swift | 0.94s |
| 17 | Python | 1.09s |
Unfortunately I could not run the docker-container with godot since it depends on x11 even for "no-window" runs... So I've computed the value from my own Pop_OS! version, which is pretty much the same as if I've runned from the container.
Considering that most of that is due to the graphical engine booting up, the raw performance from GDScript would take around 0.24s, which would put it close to native-compiled language like Nim and even faster than PHP. The GDScript performance is simply the most impressive to me, specially because it's a very simple language mainted by a couple of people and it's just blazing fast!
So, Apple claims swift is as fast as C and way faster than Obj-C, so why on earth is it so slow compared to them in this test? There's probrably two reasons for that, I assume. First reason: maybe swift's compiler is just better optimzed for MacOS and iOS, and we are currently testing on linux not darwin. Second reason: if you count only the user-space time, you get around 0.11s, which would put swift close to "C", the issue is that at kernel level it takes another ~0.80s, I'm not really sure why that happens, this behaviour can also be seen on "Java"
Make sure you have make and docker-compose installed.
Most languages will use the newest availalbe docker image (which means, newest language version), with the exception of .net, haxe and java, since the oficial maintaners don't supply a generic, non-version-specific alpine-based docker image or in the case of haxe, it's just a mess.
Just type:
$ make result
This is a list of languages that I want to create versions for (as soon as I have the free time):
- julia
- F#
- closure
- kotlin