Regex benchmark

[Tuom Larsen]

Dear list!

I recently looked at the "regex-dna" [1] benchmark at Computer Language Benchmarks Game and apparently Go runs 28x slower (145.22s / 5.22s) than C++ despite both using re2.

Please, can someone help me understand the reason for such a difference?

[1] http://benchmarksgame.alioth.debian.org/u64/performance.php?test=regexdna

[thwd]

On Wednesday, February 11, 2015 at 2:02:14 PM UTC+1, Tuom Larsen wrote:

I recently looked at the "regex-dna" [1] benchmark at Computer Language Benchmarks Game and apparently Go runs 28x slower (145.22s / 5.22s) than C++ despite both using re2.

Please, can someone help me understand the reason for such a difference?

Go only uses re2's syntax, not the library. Go's regexp package only does minor optimizations on expressions and builds a finite automaton at runtime.

Speculation: C++ probably uses metaprogramming (templates) to build the automatons at compile time, taknig full advantage of a compiler's optimization capabilities.

[Ian Lance Taylor]

On Wed, Feb 11, 2015 at 5:02 AM, Tuom Larsen <tuom....@gmail.com> wrote:
>
> I recently looked at the "regex-dna" [1] benchmark at Computer Language
> Benchmarks Game and apparently Go runs 28x slower (145.22s / 5.22s) than C++
> despite both using re2.
>
> Please, can someone help me understand the reason for such a difference?

They both use re2 but they use entirely different implementations of
it. The C++ re2 includes several optimizations that the Go version
does not.

Ian