Re: [computer-go] Hardware-Instruction.

["chrilly" <chrilly@xxxxxxxx>]

>
>I don't know of anyone who  has ever compared a pure serial program to
>a pure  parallel program.  Despite the  claims, the only  way this can
>happen is if  someone writes the best serial  program they can without
>giving ANY consideration to a parallel implementation, then converting
>it in a 100% faithful way.
>
I think this comparisons are a little bit an academic art pour la art. A
parallel programm must be in some aspects weaker than the best serial one. I
know some parallel papers where they authors say, that they have switched
off for comparison-reasons search-extensions and pruning. Just plain
Alpha-Beta. This is pointless, because such a programm is not competitive. A
parallel programm is concerning pruning/extensions principially handicaped,
because some pruning/extension techniques depend on dynamic features of the
search tree. One uses the information gathered so far in the search. But
this information is - at least partly - not available on a parallel system.

In the Single-Processor version of Brutus I used some sort of
singular-extensions. I had to skip this concept in the parallel version,
because the search become too unstable (fail-high/fail-low effects). One can
therefore not really compare the original single- and the parallel programm.
One can of course start now a 1-processor and a 16-processor version. But
this 1-processor version is optimized to make the 16-processor version as
efficient as possible. It is definetly not the best possible 1-processor
version. This is not any more my goal/problem.

Vincent exaggeration:
Hint: Adjust with logarithm.

Chrilly


_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/