Gaviota tablebases, Probing Code Release (Finally)
Posted: Wed Feb 10, 2010 7:49 pm
http://sites.google.com/site/gaviotache ... e/releases
Enjoy, it comes with zero support
You are free to do whatever the MIT License allow you to do, which is almost everything.
Version is v0.1, meaning, I hope there is no bugs, but if there is any, let me know and v1.0 won't have them. Some other features are intended to be added before 1.0. In any case, this is what I am using currently in Gaviota and works well.
It was a painful job to excise the whole code from Gaviota to release this as a module. The code is still messy and should be considered a useful prototype. It works (or at least it is supposed to), but do not go to look for beauty in it. I feel quite embarrassed about it.
The performance is acceptable, IMHO, but there are lots of room for improvement. In fact, some functions were designed on purpose to be simple so they will work the first time. Still, some of those were not optimized and are primitive. Since the bottleneck is HD access, you will probably won't notice it. The cache system is quite efficient, IMHO though, which is important.
If there is anything good about this, is the function tb_probe_soft, which will probably set this apart from the Nalimov scheme. In addition, the compression of the scheme 4 is better.
Good luck with them!
Miguel
PS: The generator code release will come later.
Enjoy, it comes with zero support
You are free to do whatever the MIT License allow you to do, which is almost everything.
Version is v0.1, meaning, I hope there is no bugs, but if there is any, let me know and v1.0 won't have them. Some other features are intended to be added before 1.0. In any case, this is what I am using currently in Gaviota and works well.
It was a painful job to excise the whole code from Gaviota to release this as a module. The code is still messy and should be considered a useful prototype. It works (or at least it is supposed to), but do not go to look for beauty in it. I feel quite embarrassed about it.
The performance is acceptable, IMHO, but there are lots of room for improvement. In fact, some functions were designed on purpose to be simple so they will work the first time. Still, some of those were not optimized and are primitive. Since the bottleneck is HD access, you will probably won't notice it. The cache system is quite efficient, IMHO though, which is important.
If there is anything good about this, is the function tb_probe_soft, which will probably set this apart from the Nalimov scheme. In addition, the compression of the scheme 4 is better.
Good luck with them!
Miguel
PS: The generator code release will come later.
