#19797: Build GSL in IEEE 754 compliant mode
-------------------------------------+-------------------------------------
Reporter: jdemeyer | Owner:
Type: enhancement | Status: new
Priority: major | Milestone: sage-7.0
Component: packages: | Resolution:
standard | Merged in:
Keywords: | Reviewers:
Authors: Jeroen Demeyer | Work issues:
Report Upstream: N/A | Commit:
Branch: | 694feb2d71ebf521cf5f32dba69e0406d16e9f11
u/jdemeyer/ticket/19797 | Stopgaps:
Dependencies: #19796 |
-------------------------------------+-------------------------------------
Comment (by vbraun):
So the question is: Should we guarantee that `x*x.conj()` has no numerical
noise as imaginary part.
I'm not sure.
* Even besides FMA there is always the possibility of 80-bit x87
instructions messing things up, though thats being retired now.
* Blas are more than happy to use FMA, so //if// we make guarantees for
scalars then they'll most likely be broken by diagonal matrices, say.
What I am pretty sure about is: multiplying `x*x.conj()` as complex
numbers is a bad way of evaluating the absolute value. Fast callables
using abs should return real and not complex values. And for that we need
to hold abs. At which point the issue of the numerical noise in the
imaginary part is pretty meaningless.
--
Ticket URL: <http://trac.sagemath.org/ticket/19797#comment:7>
Sage <http://www.sagemath.org>
Sage: Creating a Viable Open Source Alternative to Magma, Maple, Mathematica,
and MATLAB
--
You received this message because you are subscribed to the Google Groups
"sage-trac" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To post to this group, send email to [email protected].
Visit this group at https://groups.google.com/group/sage-trac.
For more options, visit https://groups.google.com/d/optout.
