[r-t] Composition search?

[Alan Burlison]

On 23 November 2015 at 18:54, Graham John <graham at changeringing.co.uk> wrote:

> What is required is either a faster processor (using SMC32) or a better
> algorithm. The last entry was in 2010, so a latest hi-end processor would
> run SMC32 faster. SMC32 is single-threaded so a multi-threaded 64-bit
> program has the potential to yield significant improvements.

Yes, if I do have a go I'd be using a high level language and
multithreading, potentially up to hundreds of threads.

> If you are intending to write your own search engine to improve the
> benchmark result, it is worth noting that SMC32 doesn't use place notation,
> or generate & prove rows, when it is composing. It traverses a composing
> tree from calling position to calling position, each in a single step,
> checking the current node for truth against previous nodes using
> pre-calculated node falseness tables. It also utilises a number of other
> optimisations, and its core is written in Assembler.

Hmm, interesting - thanks. I was thinking of just generating then
proving but it sounds like that's unlikely to be a good approach if
SMC32 uses a tree and precaulculation approach.