Re: [boost] [regex, xpressive] interesting(?) perf benchmark

Subject: Re: [boost] [regex, xpressive] interesting(?) perf benchmark
From: Isaac Gouy (igouy2_at_[hidden])
Date: 2010-06-08 11:41:09

On Jun 7, 8:17Â am, John Maddock <boost.re..._at_[hidden]> wrote:
> >> Just an idea: for static xpressive, couldn't you detect at compile-time
> >> that the expression is truly regular, and use a DFA in that case?
>
> > Oh, sure! Why don't you submit a DFA and I'll use it in xpressive? ;-)
>
> > After nosing about on the internet a bit more, I found this interesting
> > comparison:
>
> >http://shootout.alioth.debian.org/u32/benchmark.php?test=regexdna&lan...
>
> > Here we see every language compared on how well it can perform on a
> > particular regex task. The top-performer is <drumroll> Google's
> > JavaScript V8 engine! Wow. C++ is in 5th place. The fastest C++ program
> > submitted to the competition uses static xpressive <pats own back>. I'm
> > not so upset about being beaten by V8. It adaptively improves its native
> > codegen *at runtime*. What really bugs me is that we're skunked by a C
> > library: Tcl. Grrrr. I've read a bit about Tcl's regex library; it does
> > what Mathias is suggesting: implements both a DFA and an NFA, analyzes
> > the pattern and chooses which to use. I've known for a while that this
> > is the way forward, but I just don't have the time for that. (Wasn't
> > there a GSoC project to do that for Boost.Regex?)
>
> My memory fails me.... Â In any case the regex GSOC project never got off the
> ground.
>
> Nosing around the entries to the competition, I wonder how much of the
> performance difference is down to the regex engine, and how much to other
> tricks the entries use: for example I notice the top C program uses a thread
> pool to conduct everything in parallel. Â Cheating I say! ;-)

Does "everything in parallel" gain anything when the program is being
forced onto one core? (Notice the ≈ CPU Load column).

In contrast, when programs are allowed to use quad-core there's a
benefit:

http://shootout.alioth.debian.org/u32q/performance.php?test=regexdna