Re: [sage-devel] Dependence of RealDoubleElement on GSL

2021-10-13 Thread Matthias Koeppe
William, Dima, thanks for these first microbenchmarks. I agree, of course, that consistency, speed and precision are all important. It sounds like keeping GSL as an option may still be desirable. It may be possible to create a common base class RealDoubleElement_base and have two implementation

Re: [sage-devel] Dependence of RealDoubleElement on GSL

2021-10-13 Thread Dima Pasechnik
On Wed, Oct 13, 2021 at 12:41 AM William Stein wrote: > Hi, > > In addition to numerical correctness considerations, it is important > to also consider the impact on **performance**. > > > Would we be able to eliminate this dependency? Some of the functions, > like isnan, are available in the sta

Re: [sage-devel] Dependence of RealDoubleElement on GSL

2021-10-12 Thread William Stein
Hi, In addition to numerical correctness considerations, it is important to also consider the impact on **performance**. > Would we be able to eliminate this dependency? Some of the functions, like > isnan, are available in the standard C library since C99. Do functions like > gsl_sf_sin have a

Re: [sage-devel] Dependence of RealDoubleElement on GSL

2021-10-12 Thread François Bissey
After looking a bit at gsl’s doc I don’t think there is any advantage to using them if we are not using the error handling and reporting of gsl (by that I mean error estimates on the results). The only interesting detail is, quoting, “consistency across platforms”. If we are not doing high preci

Re: [sage-devel] Dependence of RealDoubleElement on GSL

2021-10-12 Thread Matthias Koeppe
On Tuesday, October 12, 2021 at 3:08:51 PM UTC-7 Dima Pasechnik wrote: > On Tue, Oct 12, 2021 at 9:46 PM Matthias Koeppe > wrote: > >> Elements of RealDoubleField (RDF) have some methods that are implemented >> using GSL. >> > >> Would we be able to eliminate this dependency? >> > > Do you me

Re: [sage-devel] Dependence of RealDoubleElement on GSL

2021-10-12 Thread Dima Pasechnik
On Tue, Oct 12, 2021 at 9:46 PM Matthias Koeppe wrote: > Elements of RealDoubleField (RDF) have some methods that are implemented > using GSL. > > Would we be able to eliminate this dependency? > Do you mean, specifically for RDF? (gsl is used in Sage's calculus/, not only in RDF) > Some of

[sage-devel] Dependence of RealDoubleElement on GSL

2021-10-12 Thread Matthias Koeppe
Elements of RealDoubleField (RDF) have some methods that are implemented using GSL. Would we be able to eliminate this dependency? Some of the functions, like isnan, are available in the standard C library since C99. Do functions like gsl_sf_sin have advantages over using functions from the mat