[Ecls-list] strange things with declarations, ECL 9.6.2

[Juan Jose Garcia-Ripoll]

2009/8/14 Gábor Balázs <gabalz at gmail.com>:
> ECL:
>   real time : 119.914 secs
>   run time  : 111.667 secs
>   gc count  : 8798 times
>   consed    : 30900318944 bytes
> SBCL
>   23.515 seconds of real time
>   5,114,640,016 bytes consed
> I also tried my library's speed test and that showed me that ECL is 100-300
> times slower than SBCL.

I get other values on my Mac OS X. ECL does 50s with 150,000,000
consed and SBCL 16s with more consing. But that is not the point.

Right now ECL is not inlining function calls. That means your code is
boxing and unboxing numbers all the time, which is the cause for the
slowdown.

> I know that ECL is not primarily designed for scientific computation
> purposes, but I want to ask a few things.
> - What kind of experience do you have about speed differences between ECL
> and an other Lisp implementations (e.g. SBCL)?

Right now this is not a fair comparison. ECL's type inferencer and
inlining capabilities are in its infancy and work is in progress on
that camp.

> - Can you recommend me a good profiler for ECL?  Should I use Slime or gprof
> somehow?

ECL has its own profile package, which you can load using (REQUIRE 'PROFILE)

Juanjo

-- 
Instituto de Física Fundamental, CSIC
c/ Serrano, 113b, Madrid 28006 (Spain)
http://juanjose.garciaripoll.googlepages.com