[Pachi] Pachi became strongest Go program on desktop systems. Are there any prospects?

Stanisław Halik sthalik at misaki.pl
Sun Sep 7 11:14:00 CEST 2014


On 2014-09-07 11:05, Petr Baudis wrote:
>> That is, will make locality worse?
>    Yes, and IIRC in some places our float ops gets vectorized and that
> should go faster with half the bit size.

Then let's make locality better to the point of not caring about such 
minute details.

First attempt at struct node refactor failed when the 'sibling' struct 
member turned out to be essential. Some logic for node frobbing depends 
critically on that particular member.

I can do the tedious work since you're the brainiac here. As such:

Decide which stuff to reorganize. Whether node 'children' member can be 
safely realloc'd[1] without care for reentrancy. Elaborate whether 
struct nodes are shared between threads here.

Which stuff can't be touched. 'children' are the point here, but let's 
arrive at what to keep, otherwise another refactor failure...

-sh

[1] It's presently just a bunch of pointers. If made to a contiguous 
allocation, as talked about, needs to be updated as grows.


More information about the Pachi mailing list