|Bill Allombert on Sat, 09 Aug 2014 22:53:10 +0200|
[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]
|Re: New ellL1 implementation by Pascal Molin|
On Wed, Jun 25, 2014 at 02:24:09PM +0200, Bill Allombert wrote: > On Sun, Jun 22, 2014 at 12:21:25PM +0200, Bill Allombert wrote: > > Dear PARI developers, > > > > I have merged Pascal Molin code for computing ellL1 and ellanalyticrank. > > The new code is much faster than the old one for derivatives. > > For the value of L itself, it is slightly slower, but only because the > > older code used unproven (and sometimes wrong) bounds for the summation. > > > > This makes ellheegner faster when the height of the point is large. > > I have commited a second patch that adds an alternative summation strategy > suggested by Pascal. ellL1 will choose the strategy that requires the less > amount of memory. I have commited a third patch that implement Brent and Kung summation for ellheegner too, when the naive height of the point is > 1900*log(2). This is significantly faster than the previous summation strategy. The bound 1900 is not very sharp. It is not very easy to tune this. Cheers, Bill.