A comparison of programming language implementation performance. We focus on the performance of the built-in primitives the language provides.
-
Each test program should implement the same algorithm.
-
No FFI.
-
No inline assembly (unless you're submitting an assembly test program!)
-
No external libraries.
-
No rolling your own datatypes. We're interested in the performance of those already provided.
-
Different language implementations (e.g. GCC and Clang) may have different test programs.
Check out the code, and run run.
$ git clone git@github.com:Wilfred/the_end_times.git
$ cd the_end_times
$ ./run
You will need Python 2, and all the languages you are interested in testing.
Pidigits tests numeric performance for arbitrary sized integers.
GHC does really well here as it uses GMP underneath. Go's impressive performance is due to some of the math/big library being written in assembly. Julia also uses GMP, so I suspect there is low hanging fruit in optimisations still.
CPython beats out pypy as pypy's JIT can't help in this benchmark.
Run on an AMD Phenom X4 965 with 4 GiB RAM with Linux x86_64 3.16.4.
| Language | Implementation | Time in seconds |
|---|---|---|
| Common Lisp | SBCL 1.2.2 | 12.049 |
| Go | Go 1.3.3 | 2.899 |
| Haskell | GHC 7.8.3 | 1.583 |
| Java | OpenJDK 1.7.0_71 | 19.618 |
| Julia | Julia 0.3.1 | 7.395 |
| PHP | PHP 5.6.2 | 438.732 |
| Python3 | CPython 3.4.2 | 4.568 |
| Python3 | pypy 2.4.0 | 6.463 |
| Ruby | JRuby 1.7.16 | 29.197 |
| Ruby | MRI 2.1.3p242 | 27.721 |
| Ruby | Rubinius 2.2.10 | 19.010 |
| Rust | Rust 0.12.0-dev | 11.793 |
| Scheme | Gambit 4.7.3 | 14.631 |
| Scheme | Guile 2.0.11 | 9.413 |
| Scheme | Racket 6.1 | 17.369 |
Code is under the BSD license, and includes code originally from The Computer Language Benchmarks Game.