Thanks for all those tests. I wonder why it does not run on your Nexus S's default browser. Does it support JavaScript? Maybe not fully?
p10047 takes a long time finding the right "x" for a strong discriminant.
I have been testing version 0.8.4. Here is a result on Chrome at 2.7GHz:
PRP 2^69337-3: 3456571ms
That is just under an hour. It hardly touches the O(n^2) partialModReduction(). I guess it is O(n) for that particular number. Most of the time is going to be in O(n^1.585) karatsuba().
I have made improvements to the sumArray() and put the karatsuba threshold back to 64 words. I can't see what can be improved next. Toom-Cook-n will only complicate the program.
Clearly, a native implementation is going to win by many miles, and it is good to see BPSW wins overall. Maybe I should try to convert my JavaScript code into C, to see if it is any faster than GMP
Version 0.8.4: