The source code for GlaurungCCT8 can now be downloaded from my site. Binaries for Windows, Linux and Mac OS X will be added in the next few days. This is almost the exact version which played in the CCT, except that a tiny quiescence bug has been fixed, and that the program finally sends information about hash usage to the GUI. It's funny, Glaurung has been available for download for about one and a half year, and nobody ever complained about the missing hash usage information. Then, suddenly, two people complained during the same day.
The changes compared to Glaurung 1.0.2 are rather unexciting, unless you happen to have a computer with more than one CPU:
- Parallel search has been added, using the technique known as the "Young Brothers Wait Concept" (or at least something very similar). This is limited to two CPUs so far, support for bigger numbers of CPUs will be added in a later version. Of course the current version will also work on a computer with more than two CPUs, but no more than two will be used.
- Several minor bugs which were found during implementation of the parallel search have been fixed. Nothing big, and not likely to have any influence on playing strength.
- Several new bugs were without doubt added during implementation of the parallel search. I hope these also don't have any big influence on the playing strength, but before I find them I cannot know.
Please note that this version almost requires at least two CPUs (or a single dual-core CPU). It will be ridiculously slow with the default settings. If you reduce the "Number of threads" UCI parameter to 1, it will work reasonably well on a single-CPU computer, but it will still be a bit slower and probably weaker than Glaurung 1.0.2. In other words, those of you without a dual shouldn't upgrade, but rather wait for the next Glaurung version.
I have two goals with this version: The first is to show that parallel search is nowhere near as difficult to do than most people seem to believe, and that YBWC isn't necessarily harder than ABDADA and other shared hash table algorithms. I hope we will see many new parallel engines in the next few months, and that my source code will make the job a bit easier for some of the programmers.
The second goal is to try to put an end to the awful naming convention of adding "Deep" in front of the name for parallel versions of chess engines. I wish people would either leave the engine name unchanged (like Crafty, Diep, Zappa and Scorpio, among others), add a simple "SMP" at the end, or do something more creative ("ParSOS" is a fabulous name!). The "Deep" prefix is embarassingly silly, especially for those of us who have been around long enough to remember how it all started.
Tord
