BLD: Negative zero handling with ifort

[h6197627]

Fixes scipy/scipy#11339

What does this implement/fix?

NumPy/SciPy compiled with Intel Fortran compiler using default flags treat both -0.0 and +0.0 as 0.0, however it is not IEEE 754 compliant and some SciPy logic relies on this -0.0 and +0.0 differentiation.

[charris]

Can you give an example of the problems?

[h6197627]

@charris, one of SciPy test fails because some functionality relies on this difference between -0.0 and +0.0

_____________________________________________ TestExp1.test_branch_cut _____________________________________________
self = <scipy.special.tests.test_exponential_integrals.TestExp1 object at 0x00000296E7953EC8>

    def test_branch_cut(self):
        assert np.isnan(sc.exp1(-1))
>       assert sc.exp1(complex(-1, 0)).imag == (
            -sc.exp1(complex(-1, -0.0)).imag
        )
E       AssertionError: assert -3.141592653589793 == --3.141592653589793
E        +  where -3.141592653589793 = (-1.8951178163559368-3.141592653589793j).imag
E        +    where (-1.8951178163559368-3.141592653589793j) = <ufunc 'exp1'>((-1+0j))
E        +      where <ufunc 'exp1'> = sc.exp1
E        +      and   (-1+0j) = complex(-1, 0)
E        +  and   -3.141592653589793 = (-1.8951178163559368-3.141592653589793j).imag
E        +    where (-1.8951178163559368-3.141592653589793j) = <ufunc 'exp1'>((-1-0j))
E        +      where <ufunc 'exp1'> = sc.exp1
E        +      and   (-1-0j) = complex(-1, -0.0)

self       = <scipy.special.tests.test_exponential_integrals.TestExp1 object at 0x00000296E7953EC8>

C:\Program Files\Python\3.7.6\lib\site-packages\scipy\special\tests\test_exponential_integrals.py:12: AssertionError

After some discussions in scipy/scipy#11339 @rgommers advised to pass this flag at the build system level rather than to specific SciPy module, because it forces NumPy/SciPy to follow closer IEEE floating point standards when building with Intel compilers.

[charris]

Thanks @h6197627 . I note that there is no get_flags_debug for the Intel compiler on non-windows platforms, do you know if we should have that?

[rgommers]

Thanks @h6197627

[rgommers]

Let's not worry about get_flags_debug unless anyone complains about it.