Over in comp.lang.c++.moderated it was suggested that I'd castrated C++ or hobbled the DMC++ compiler in order to get the benchmark results in www.digitalmars.com/d/cppstrings.html. Since the conventional wisdom is that C++ using STL can't be beat for efficiency, I can understand their skepticism. I'd be skeptical too.

So David Friedman has stepped in to compare his excellent gdc compiler with g++, where the compilers also share a common optimizer and code generator. As everyone can see on the updated web page, the results are comparable. D runs away with the win by any measure.

Many thanks to David!

Walter wrote:
> Over in comp.lang.c++.moderated it was suggested that I'd castrated C++ or
> hobbled the DMC++ compiler in order to get the benchmark results in
> www.digitalmars.com/d/cppstrings.html. Since the conventional wisdom is that
> C++ using STL can't be beat for efficiency, I can understand their
> skepticism. I'd be skeptical too.
> 
> So David Friedman has stepped in to compare his excellent gdc compiler with
> g++, where the compilers also share a common optimizer and code generator.
> As everyone can see on the updated web page, the results are comparable. D
> runs away with the win by any measure.

FWIW, I decided to run my own benchmarks using DMC (with STLPort) and DMD.  I tried the default code plus a quick hack of the C++ version to dump everything in a stringstream before parsing.  Here are my results:

D:

Execution time: 0.077 s
Execution time: 0.071 s
Execution time: 0.079 s

CPP 1 (original):

Execution time: 0.136 s
Execution time: 0.125 s
Execution time: 0.132 s

CPP 2 (stringstream version):

Execution time: 0.146 s
Execution time: 0.155 s
Execution time: 0.157 s

This does seem a fair test of the code the average person is likely to write using each language.  I know each could be optimized, but doing so would require some skill and/or nonstandard library code.  A refcounting string and a hashmap would be the obvious improvements for C++, but while the hashmap will soon be part of the standard, there is a lot of contention over whether a refcounting version of basic_string could ever be standard-compliant.  If I have the time I'll see about using STLPort's hashmap (I'm pretty sure it has one) to flatten the differences a bit.


Sean

Sean Kelly wrote:
> If I have the time I'll see about using STLPort's hashmap (I'm pretty sure it has one) to flatten the differences a bit.

Well that turned out to be easier than I thought.  I just replaced "map" with "hash_map" and included the appropriate header file.  Here are the results:

CPP 3 (with hash_map):

Execution time: 0.102 s
Execution time: 0.110 s
Execution time: 0.102 s

So D is still the clear winner.


Sean

"Sean Kelly" <sean@f4.ca> wrote in message news:ck9qcp$1a5s$1@digitaldaemon.com...
> Sean Kelly wrote:
> > If I have the time I'll see about using
> > STLPort's hashmap (I'm pretty sure it has one) to flatten the
> > differences a bit.
>
> Well that turned out to be easier than I thought.  I just replaced "map" with "hash_map" and included the appropriate header file.  Here are the results:
>
> CPP 3 (with hash_map):
>
> Execution time: 0.102 s
> Execution time: 0.110 s
> Execution time: 0.102 s
>
> So D is still the clear winner.
>
>
> Sean

It would be interesting to also rehash the assoc array in the D code and see if that speeds up anything. Also it would be neat to pre-allocate the hash table so that the array is faster to define, too.

"Sean Kelly" <sean@f4.ca> wrote in message news:ck9q0a$1a0u$1@digitaldaemon.com...
> This does seem a fair test of the code the average person is likely to write using each language.  I know each could be optimized, but doing so would require some skill and/or nonstandard library code.

And that's part of the point. Both the D and C++ versions are straightforward implementations using the standard features of each language. Both languages offer the ability to supertune the source, etc., but I don't think that would produce any useful information as few are ever likely to write such code.

How about implement all the tests on the "The Great Computer Language Shootout" web-page, and submit to:

http://shootout.alioth.debian.org/ http://dada.perl.it/shootout/

(On the win32 page, it only shows D 0.42, without any acutal test.)

D will then gain much more visibility.

<some@where.com> wrote in message news:ckbuo6$2q7v$1@digitaldaemon.com...
>
> How about implement all the tests on the "The Great Computer Language
Shootout"
> web-page, and submit to:
>
> http://shootout.alioth.debian.org/ http://dada.perl.it/shootout/
>
> (On the win32 page, it only shows D 0.42, without any acutal test.)
>
> D will then gain much more visibility.

I agree, and it's being worked on.