These are chat archives for symengine/symengine

3rd
Mar 2016
Isuru Fernando
@isuruf
Mar 03 2016 02:37
@certik, how about making our mpz_wrapper the default for the integer storage?
Also we can stop building teuchos when WITH_SYMENGINE_RCP=no and WITH_BFD=no right?
Akash Trehan
@CodeMaxx
Mar 03 2016 13:25
@isuruf Do you know of any references for finding the taylor_term for inverse hyperbolic functions( acsch to be precise )
I mean the algorithm
Isuru Fernando
@isuruf
Mar 03 2016 13:31
from sympy import *
var("x")
asinh(1/x).diff(x).series(x, 0, 10).integrate(x)
That won't work
Kalevi Suominen
@jksuom
Mar 03 2016 13:37
The behaviour of acsch is very bad at 0, which lies in the middle of the branch cut. One can at best hope for asymptotic expansions with logarithmic terms, and probably different ones for the positive and negative sides. I don't think it would make much sense to introduce them to SymPy.
Akash Trehan
@CodeMaxx
Mar 03 2016 14:00
oh…ok
Ondřej Čertík
@certik
Mar 03 2016 21:11
@isuruf, yes we don't need to build Teuchos with WITH_SYMENGINE_RCP=yes and WITH_BFD=no.
Yes, we should make mpz_wrapper the default. This will also make symengine build just with the gmp C library, the gmpxx is not needed.
Akash Trehan
@CodeMaxx
Mar 03 2016 21:13

I was reading up on Teuchos RCP , RCP and UniquePtr.
In design.md in the conclusions it says
However, the issue is that even manual new/delete (or the equivalent UniquePtr) is slow, so we want to avoid it, or only do it as little as possible. I am still thinking if we could perhaps use UniquePtr instead of RCP.

So is RCP slower or faster than UniquePtr ?

@certik

Ondřej Čertík
@certik
Mar 03 2016 21:14
RCP is slower than UniquePtr.
(RCP is doing reference counting in addition to what UniquePtr does.)
Akash Trehan
@CodeMaxx
Mar 03 2016 21:16
Even after using this PR...certik/trilinos#1
I mean in Debug mode
and in what context is the UniquePtr slow(as mentioned above) then ?