Bill Gosper's original test pattern was a puffer train. For those who have Golly, the initial pattern is: x = 19, y = 10, rule = B3/S23 b3o11b3o$bo2bo10bo2bo$bo6b3o4bo$bo5bo2bo4bo$2bobo2bo3bo4bobo$6b2o2bobo $8bo$9b2obo$9b3o$9b3o! Golly's implementation of Hashlife is so fast that it can run this pattern to 10^308 generations in less than 30 seconds (and using about 180 megabytes of memory) on my 3.2-GHz Intel Core i7. It's hard to time this simply because I can't get the "+" key to auto-repeat fast enough. Beyond 10^308, the generation count displays as "inf" because it overflows in the conversion to floating-point in the status display routine. (I have fixed this, and I need to send the patches to the other Golly folks). By comparison, in his 1984 paper Bill reported that "an early model Symbolics 3600" (a LISP Machine) had run Hashlife "several million steps, at a rate which doubled every two or three minutes" (once the initial non-periodic part of the exhaust trail had been fully computed). Golly does 1024 doublings in less than 30 seconds. - Robert On 5/24/12, Bill Gosper <billgosper@gmail.com> wrote:
Just try http://en.wikipedia.org/wiki/Hashlife . I think Elsevier still has the 1984 Physica D article paywalled. I scanned the seven separate pages as http://gosper.org/physicad0.png .. http://gosper.org/physicad6.png, (didn't somebody "staple" these?), but Tom Rokicki (et al)'s Golly<http://golly.sourceforge.net/>has exceeded my wildest expectations (and blew away Conway when I showed him).
-- Robert Munafo -- mrob.com Follow me at: gplus.to/mrob - fb.com/mrob27 - twitter.com/mrob_27 - mrob27.wordpress.com - youtube.com/user/mrob143 - rilybot.blogspot.com