[Hplusroadmap] Fwd: [BBF Standards] Input/output

[Bryan Bishop]

On Monday 11 February 2008, Eugen Leitl wrote:
> Have you considered the possibility that there is no human-parseable
> mapping into the problem space? People do emergence and massive
> parallelism very badly. Sequential processes and processivity are
> mutually exclusive. Brew a cup of tea, and think about it for a
> while.

So I thought about it for a while. I see that my response avoids a 
particular issue. While parallel programs can be broken into linear 
problems, necessarily (typical) programming is done linearly, and 
therefore we run into the linear -> parallel conversion problem 
commonly found in computer science. At the moment, the best hack 
solution I can offer is to have a generalized database for known types 
of problems in cs that can be solved in polynomial time and those that 
cannot, and then have a 'compiler' attempt to optimize any linear 
program for a parallel version with the known conversions of linear 
problems into parallel problems. However, in some cases, there will not 
be known solutions and users would have to run programs to search the 
problem space for the new solutions. This is a hacky solution and 
requires more clarification, so let me get back to it once I review my 
compiler architecture and my understanding of hardcore theoretical cs.

- Bryan
________________________________________
Bryan Bishop
http://heybryan.org/