My understanding is Trial Factoring doesn't use FFTs, it eliminates most factors then does the powering algorithm using integer and floating point math.

Unfortunately there is no built in benchmark for trial factoring ( that I know of ).

One could be made, it could use some fixed mersenne number and a fixed number of iterations per screen output, say 1000, it would have to be run, result recorded then stopped with the bits factored for the mersenne upped in the worktodo.ini, this way the performance at 57,58...67 could be used.

There are various plateaus of efficiency at different bit depths -62, 63+64, 65+.

I guess maybe just 61,63,65 could be used, because 62 and below it doubles the time as bit depth increments ( by 1 ), same for 63 and 64, and again at 65 and above. ( This is only for non SSE2 CPUs which is what you were mentioning. )

The times are always very close to 2x the previous except at 62 to 63 and 64 to 65 where it is close to 4x.

The data could be a table like the benchmark page on mersenne.org or the informative graph guido72 made

http://www.mersenneforum.org/jpg/bench.jpg
from the Xyzzy's benchmark challenge thread

http://www.mersenneforum.org/showthr...=&threadid=992
or even a small program that asks for some information, CPU and clock speed, and returns the top 3 versions.

If no exact match either say so and show the closest.

A table is probably more informative and you can use your own judgement rather than hoping the algorithm got it correct.