[plt-scheme] Reexpansion of modules

From: Jens Axel Søgaard (jensaxel at soegaard.net)
Date: Mon Aug 21 15:56:44 EDT 2006

Matthew Flatt skrev:
> At Sat, 19 Aug 2006 19:29:01 +0200, Jens Axel Søgaard wrote:
>> The recent thread on MzScheme and Larceny were thought provoking.
>> Even though Matthias has lost hope to see a bridge between MzScheme
>> and Larceny, maybe it is possible cross the gap using a tightrope?
>> Intended not for the masses, but for the adventurous.
> 
> How much speed improvement would you expect to get that way?
> 
> MzScheme isn't slow:
> 
>   http://tmp.barzilay.org/mzbench.html

Not much. It just seemed easier to build than Felleisen's suggestion:

   > 2. I still believe that a whole-program compiler for PLT Scheme
   >   targeting the LLVM in addition to the JIT compiler would be
   >   extremely beneficial to the overall project. The less C people
   >   have to write for performance reasons the better.

Also, when I originally wrote the code, I were hoping to use
the MzScheme module+macro system to write portable code. Now I don't
worry about that.

Thanks for the numbers - I hadn't registered that the 3m version
had become faster than the standard Böhm-collector version.

> Well, it's still slow on programs that use continuations at a very fine
> granularity. Otherwise, the gap between MzScheme and
> performance-oriented implementation has closed considerably, especially
> for the benchmarks that are supposed to represent real programs (e.g.,
> "dynamic"). I see no particular obstacle to closing the gap further.

Good news.

> It's difficult to know whether these benchmarks measure anything that
> you care about, though. Maybe you have some code that runs much faster
> in Larceny than MzScheme? If so, I'm interested to take a look.

Not yet :-)

-- 
Jens Axel Søgaard




Posted on the users mailing list.