Re: Experiences about speed in tcl and STk.

From: Erick Gallesio <eg_at_kaolin.unice.fr>
Date: Fri, 10 Mar 1995 17:18:28 +0000

> Johann Friedrich Heinrichmeyer writes
>
> I was a little disappointed by the speed of my port of my schematic
> editor interface for spice (not faster than the tcl version). I
> experienced that regexp-replace seems to be very slow. Maybe "case" and
> "do" should be compiled ...? The loading of a schematic file increased
> in speed very much when i used a more "scheme like" input format and
> therefore only the "read .." command. When i read in the old tcl format
> using regexp heavily the program slows down.
>

Are you sure that you do not compile several time the same regular expression?
Regexp compiling is very time consuming. In STk, compilation is done with
the string->regexp primitive. For instance,

        (do ((i 0 (+ i 1)))
            ((= i 10000))
          ((string->regexp "a*b*c*d") "XXXXXX ad YYYYY")))

takes 3266.67ms on a sparc10 whereas

        (define pattern (string->regexp "a*b*c*d"))
        (do ((i 0 (+ i 1)))
            ((= i 10000))
           (pattern "XXXXXX ad YYYYY")))

where the regexp is compiled only once takes 1850.00ms.

> Here three questions.
>
> 1. SCM and STk seem to have the same parents?

Yes, they share a SIOD origin.
 
> 2. How difficult is it to put speedups from SCM to STk ( case and do
> construct are compiled in SCM)?

This not too difficult (but not too simple too, this explain why it is not
already done. However, I hope to integrate them soon.
 
> 3. Has someone already tried to bring scm compilers and STk together?
> I think of meroon, hobbit .... . I dont want to compile everything
> but it would be nice to transform selected and proven procedures
> more or less in dlls ("remove-if" from lisp or similar things).

Hobbit could be probably ported for generating STk code. But as far as I know,
nobody has started such a job.

                -- Erick
Received on Fri Mar 10 1995 - 17:18:54 CET