[pythran] cx-limited-range

  • From: Mehdi AMINI <joker.eph@xxxxxxxxx>
  • To: pythran@xxxxxxxxxxxxx
  • Date: Fri, 14 Nov 2014 12:02:15 -0800

Hi,

from pythran/pythonic/patch/README.rst
"""
The implementation of std::complex is very slow, due to the complex limited
range (see `-fcx-limited-range`) feature. Numpy does not implement it, so we
have to conform to numpy's version.
"""

I don't get it. To begin with, the implementation of std::complex is very slow due the ABSENCE of complex limited range.
Then what it is about the Numpy statement?
Complex in Python (Core, not Numpy), implements the division with range reduction: https://github.com/python-git/python/blob/master/Objects/complexobject.c#L92 This is far from the patch in C++ complex: (__c * __a + __d * __b, __c * __b - __d * __a) / __e;

Moreover I don't even get the validity of that. You are replacing an implementation in a standard library, really???? This looks like a really bad thing to do.

If your problem is that NT2 expect std::complex, then patch NT2!

Mehdi



Other related posts: