CADO
Hi,
Would anyone who attended the CADO workshop care to
summarize what went on? Were there any revelations/
insights/new ideas ??
I now have a proof that using a constant area sieve region
for the lattice sieve is not optimal. One should use a larger
region for the smaller special q's and gradually decrease the size
of the region as q increases. However, the speedup obtained does
not seem to be large for a practical reason.
Owing to the way computers address 2D arrays, it is
advantageous to have a sieve region be 2^n x 2^m.
An attempt to use (say) a sieve region that is 1.537 x 2^n
by 3.682 x 2^m would slow down the addressing.....
Thus, while the total number of sieve points would be reduced
by following an optimal "size strategy", addressing those points
would be slower.
I can write this up and submit a paper, but since the optimization
is not (very) effective, I am not sure that it is worth doing.
