How to install numpy on a box without hardware FPU
Hello, Is it possible to configure/install numpy on a box without a hardware FPU? When I try to install it using pip,I get a bunch of compile errors since floating-point exceptions (FE_DIVBYZERO etc) are undefined on this platform. How do I get numpy installed and working on such a platform? Thanks,Emel
On 01.09.2014 10:33, Emel Hasdal wrote:
Hello,
Is it possible to configure/install numpy on a box without a hardware FPU? When I try to install it using pip, I get a bunch of compile errors since floating-point exceptions (FE_DIVBYZERO etc) are undefined on this platform.
How do I get numpy installed and working on such a platform?
If its just that you can try replacing all the fenv stuff with stubs doing nothing. You only lose some runtime warnings about special cases. Why do you want to run numpy on such a system? Numpy is not really intended to run on such devices. But it is possible the debian armel port is so far I know softfloat and numpy seems to be running fine, though it probably also emulates fenv.
I am trying to run a python application which performs statistical calculations using Pandas which seem to depend on Numpy. Hence I have to install Numpy to get the app working. Do you mean I can change numpy/core/src/npymath/ieee754.c.src such that the functions referencing exceptions (npy_get_floatstatus, npy_clear_floatstatus, npy_set_floatstatus_divbyzer, npy_set_floatstatus_overflow, npy_set_floatstatus_underflow, npy_set_floatstatus_invalid) do nothing? Could there be any implications of this on the numpy functionality? Thanks a lot, Emel
Date: Mon, 1 Sep 2014 23:45:02 +0200 From: jtaylor.debian@googlemail.com To: numpy-discussion@scipy.org Subject: Re: [Numpy-discussion] How to install numpy on a box without hardware FPU
On 01.09.2014 10:33, Emel Hasdal wrote:
Hello,
Is it possible to configure/install numpy on a box without a hardware FPU? When I try to install it using pip, I get a bunch of compile errors since floating-point exceptions (FE_DIVBYZERO etc) are undefined on this platform.
How do I get numpy installed and working on such a platform?
If its just that you can try replacing all the fenv stuff with stubs doing nothing. You only lose some runtime warnings about special cases.
Why do you want to run numpy on such a system? Numpy is not really intended to run on such devices. But it is possible the debian armel port is so far I know softfloat and numpy seems to be running fine, though it probably also emulates fenv. _______________________________________________ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion
On Tue, Sep 2, 2014 at 8:29 AM, Emel Hasdal <emel_hasdal@hotmail.com> wrote:
I am trying to run a python application which performs statistical calculations using Pandas which seem to depend on Numpy. Hence I have to install Numpy to get the app working.
Do you mean I can change
numpy/core/src/npymath/ieee754.c.src
such that the functions referencing exceptions (npy_get_floatstatus, npy_clear_floatstatus, npy_set_floatstatus_divbyzer, npy_set_floatstatus_overflow, npy_set_floatstatus_underflow, npy_set_floatstatus_invalid) do nothing?
Could there be any implications of this on the numpy functionality?
AFAIK, few people have ever tried to run numpy on CPU without an FPU. The generic answer is that we do not know, as this is not a supported platform, so you are on your own. I would suggest you just try adding stubs to see how far you can go, and come back to this group with the result of your investigation. David
Sure. I would be happy to try it and report back. Just to make sure I understand it: I will try defining the following functions for my platform such that they do nothing: npy_get_floatstatus, npy_clear_floatstatus, npy_set_floatstatus_divbyzer, npy_set_floatstatus_overflow, npy_set_floatstatus_underflow, npy_set_floatstatus_invalid Thanks,Emel Date: Tue, 2 Sep 2014 09:00:37 +0100 From: cournape@gmail.com To: numpy-discussion@scipy.org Subject: Re: [Numpy-discussion] How to install numpy on a box without hardware FPU On Tue, Sep 2, 2014 at 8:29 AM, Emel Hasdal <emel_hasdal@hotmail.com> wrote: I am trying to run a python application which performs statistical calculations using Pandas which seem to depend on Numpy. Hence I have to install Numpy to get the app working. Do you mean I can change numpy/core/src/npymath/ieee754.c.src such that the functions referencing exceptions (npy_get_floatstatus, npy_clear_floatstatus, npy_set_floatstatus_divbyzer, npy_set_floatstatus_overflow, npy_set_floatstatus_underflow, npy_set_floatstatus_invalid) do nothing? Could there be any implications of this on the numpy functionality? AFAIK, few people have ever tried to run numpy on CPU without an FPU. The generic answer is that we do not know, as this is not a supported platform, so you are on your own. I would suggest you just try adding stubs to see how far you can go, and come back to this group with the result of your investigation. David _______________________________________________ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion
participants (3)
-
David Cournapeau
-
Emel Hasdal
-
Julian Taylor