Message342837
I have also some ideas about algorithmic optimizations (they need to be tested). In classic formula $a_{i+1} = a_i + (n - a_i^2)/(2*a_i)$ we can calculate $n - a_i^2$ as $(n - a_{i-1}^2) - (a_i^2 - a_{i-1})^2 = (n - a_{i-1}^2) - (a_i^2 - a_{i-1})*(a_i^2 + a_{i-1})$. $n - a_i^2$ usually is much smaller than $n$, so this can speed up subtraction and division. Things become more complicated when use shifts as in your formula, but I think that we can get benefit even in this case. This can also speed up the final check $a_i^2 <= n$. |
|
Date |
User |
Action |
Args |
2019-05-19 07:16:22 | serhiy.storchaka | set | recipients:
+ serhiy.storchaka, mark.dickinson |
2019-05-19 07:16:22 | serhiy.storchaka | set | messageid: <1558250182.76.0.794719861419.issue36957@roundup.psfhosted.org> |
2019-05-19 07:16:22 | serhiy.storchaka | link | issue36957 messages |
2019-05-19 07:16:22 | serhiy.storchaka | create | |
|