RE: [computer-go] Designing faster, better influence functions

["Mark Boon" <tesuji@xxxxxxxxx>]

I don't think I would want to get into using CPU-specific instructions. But
I must say I've sometimes thought about what could be useful to have in
terms of Go-specific hardware. In Tsume-Goliath I think I found that the
single most expensive function was getNlib(), which would retrieve the
number of liberties of a chain at a certain coordinate and put the
coordinates of those liberties in a little list, with a maximum of 'n'
liberties. A very simple and basic function, used in a lot of different
places. The total time spent in this one function was something like 20 or
30%.

There are others like that, like a kind of flood-fill function which I use a
lot. Wouldn't todays graphics cards be able to do something like a 4-way
flood-fill in hardware? Does anyone know something about this?

> -----Original Message-----
> From: computer-go-bounces@xxxxxxxxxxxxxxxxx
> [mailto:computer-go-bounces@xxxxxxxxxxxxxxxxx]On Behalf Of Frank de Groot
> Sent: Sunday, October 31, 2004 21:52
> To: computer-go
> Subject: Re: [computer-go] Designing faster, better influence functions
>
>
> From: "Mark Boon" <tesuji@xxxxxxxxxxxxxxxxx>
>
>
> > Would you care to explain? I'm not familiar with the terms MMX,
> > morphological kernels or Manhatten distance.
>
> Manhatten distance is what David called "Taxicab distance".
> Not the Pythagorean distance but the sum of H & V distances.
>
> MMX is Multimedia Extensions, 64-bit instructions in all
> Pentium-compatible
> CPU's of today.
> Morphological kernels are "lenses" in the shape (in our case) of a diamond
> which contain modification values.
>
>
> > The faster you can make it, the better. But influence itself doesn't
> really
> > have that much value.
>
> Yes.
> I need it to be used in search so it must be fast.
>
> _______________________________________________
> computer-go mailing list
> computer-go@xxxxxxxxxxxxxxxxx
> http://www.computer-go.org/mailman/listinfo/computer-go/
>

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