Re: Supercompiled Perl

Helmut Enck-Radana wrote:

> Not caching results, but specialized versions of the original program.
> Consider following simple example: A command line tool takes 2 arguments
> and produces some output dependent on them. In 50% of all invocations the
> 2nd argument is the same. The supercompiler can transform the program to a
> specialized one, which has only 1 argument. The 2nd argument has become a
> constant within the program. The new program is as efficient as if it had
> been designed from scratch for being dependent on only 1 argument. AFAIU
> this is possible, because the semantics of the whole program is analyzed, a
> model is created, and a complete new program with the same semantics is
> generated.
 [ ... ]
 
>"Andrei P. Nemytykh" wrote:
>  I remember an idea ( by Sergei Romanenko ). He suggests
>  to use supercompiler together with a meta-program. The meta-program
>  has to be a part of a concrete operating system. It is a dispatcher
>  of supercompilation.
>  The program has to observe the behavior of users:
>   - how often a concrete Perl-program is used ?
>   - maybe, some arguments of a program have a constant intrivial
>     properties, so its specialized version makes sense ?
>   - maybe, something else ...
>
>  If the meta-program decides to transform a Perl-program, then it does
>  that and creates a new residual version. After that any call to
>  the original Perl-program ( within the chosen context ) is replaced
>  with the residual version automatically.
> [...]

  This idea solves a number of problems of supercompilation.
  For example,
   -1-  Why are we in need to supercompile a program ?
   -2-  Supercompilation can take much time to do deep transformations,
        because of users do not wait of ending  the process at all.
   -3-  What is more, a supercompiler can try different strategies of
        the transformations to choose the best one.

Sincerely,
Andrei Nemytykh.