[Numpy-discussion] Disabling Extended Precision in NumPy (like -ffloat-store)

David Cournapeau cournape@gmail....
Sun Apr 25 07:07:22 CDT 2010


On Thu, Apr 22, 2010 at 12:04 PM, Adrien Guillon <aj.guillon@gmail.com> wrote:

>
> The idea here, is that if I can ensure there is never extended
> precision in the Python code...

This is totally out of reach with numpy is you use the float32 dtype,
for the reasons I have given before. The only solutions I could see
are: using an enviroment geared toward IEEE 754 compliance (CPU
emulation),  simply use one of the existing package to run code on
GPU, or use software-implemented FPU. The latter meaning that you
cannot use linear algebra and so on, at least not with just
numpy/scipy,

David


More information about the NumPy-Discussion mailing list