I have some negative results to report. I do not have access to the final partials right now.
2c/7 odd symmetric width 19 - no ships found, longest partial - ~300 knight2 phases, length ~75
3c/7 asymmetric width 14 - no ships found, longest partial - 90 knight3 phases, length ~30
3c/6 asymmetric width 18 - no ships found, longest partial - 36 knight3 phases, length ~18
The actual length of the longest partial is several beyond what I report, but my search program forgets it before printing it out.
I wrote a variation of my knight2 program I call knight3. Knight3 is base architected to check every 3rd phase. This gives me the ability to search for c/3 and 3c/6 ships. I then check for every 2nd phase some of the time and this allows me to search for 3c/7 and 2c/5 ships. Knight3 is very preliminary and can currently only search for asymmetric and doubled period glide symmetric ships. Getting the odd, even, and gutter modes working will occur as time permits.
I checked for glide symmetric 4c/10 ships using knight3. I was all excited that I had found the following new type of ship. Then I checked the 2c/5 thread and I guess it has been known for a while.
Code: Select all
x = 15, y = 20, rule = B3/S23
4bo5bo$3b3o3b3o$2bo2bo3bo2bo$b3o7b3o$2bobo5bobo$4b2o3b2o$o4bo3bo4bo$5b
o3bo$2o3bo3bo3b2o$2bo2bo3bo2bo$4bo5bo2$6b3o$4b2o3b2o$7bo3bo$2bo2b2o$5b
o3bo$bo3bo3b2o$10bo$2b2o5bo!
I suspect my search programs are sometimes faster and sometimes slower than glife. I am curious about what some example benchmarks for glife are. For knight2 I can report the following execution times to first ship found:
c/4 asymmetric width 10 - 90 seconds
c/5 gutter width 19 - 2 hours
c/5 asymmetric - 24 hours
These are all single thread scalar core numbers and as far as I can tell all cores run about the same speed today.
The searches I reported at the beginning of this post all took between 2 and 4 weeks of total CPU time. I chopped the last two searches into pieces and distributed them among ~8 cores do to my impatience.
I am currently developing "knightcuda" which will be massively parallel and run on graphics cards. This is slow going, as I have never programmed a massively parallel graphics card before and great care must be taken in order to get decent performance.
Have a happy day,
-Tim Coe